Color MGMT Concepts

Color Management Concepts
Michael Stokes Microsoft Corporation Draft v. 0.7, March 5, 2001 Contents

Foreword...............................................................................................................................3 Introduction...........................................................................................................................6 Color Science: Models and Algorithms................................................................................9 Models and Algorithms....................................................................................................9 Human Visual System................................................................................................... Viewing !n"ironment Models........................................................................................ 3 Measurement Methods................................................................................................. # $e%ect and Arti%act &ecognition.................................................................................... ' Color Management............................................................................................................. ( Color $ata .................................................................................................................... ( Metadata Structures......................................................................................................)6 *rocessing Se+uences.................................................................................................3( Functional Structures....................................................................................................,) Color Imaging -or.%lows..............................................................................................,# Color /uality.......................................................................................................................## Im0ortance o% 1esting Methods....................................................................................## Considerations in !sta2lishing 1est Methods...............................................................#' Common Image Characteristics....................................................................................6) Color /uality Metrics.....................................................................................................6' A *rocess %or !"aluating Image /uality.......................................................................'( Image /uality 3imitations o% Com0uter Systems..........................................................'9 *ractical Suggestions....................................................................................................(6 I!C Color Standards...........................................................................................................(( A00roach.......................................................................................................................(( *ossi2le Architecture %or Color Management in 40en Systems..................................(9 &esults o% International Standardi5ation in I!C61C 77...............................................97 How Color Management 8e &eali5ed...........................................................................9 1erminology........................................................................................................................93 A2out the Author............................................................................................................... 7, !ndnotes........................................................................................................................... 76
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
1he in%ormation contained in this document re0resents the current "iew o% Microso%t Cor0oration on the issues discussed as o% the date o% 0u2lication. 8ecause Microso%t must res0ond to changing mar.et conditions< it should not 2e inter0reted to 2e a commitment on the 0art o% Microso%t< and Microso%t cannot guarantee the accuracy o% any in%ormation 0resented. 1his document is %or in%ormational 0ur0oses only. MIC&4S4F1 MA=! 94 -A&&A91I!S< !>*&!SS 4& IM*3I!$< I9 1HIS $4C?M!91. Microso%t Cor0oration may ha"e 0atents or 0ending 0atent a00lications< trademar.s< co0yrights< or other intellectual 0ro0erty rights co"ering su2@ect matter in this document. 1he %urnishing o% this document does not gi"e you any license to the 0atents< trademar.s< co0yrights< or other intellectual 0ro0erty rights eAce0t as eA0ressly 0ro"ided in any written license agreement %rom Microso%t Cor0oration. Microso%t< -indows< and -indows 91 are trademar.s or registered trademar.s o% Microso%t Cor0oration in the ?nited States and6or other countries. 4ther 0roduct and com0any names mentioned herein may 2e the trademar.s o% their res0ecti"e owners. B )77 Microso%t Cor0oration. All rights reser"ed. Microso%t than.s the International !lectrotechnical Commission CI!CD %or 0ermission to use material %rom the I!C 6 966 series o% international standards. All eAtracts are co0yright B I!C< :ene"a< Swit5erland. All rights reser"ed. Further in%ormation on the I!C< its 0u2lications and its role is a"aila2le %rom www.iec.ch. I!C ta.es no res0onsi2ility %or and will not assume lia2ility %or damages resulting %rom the readerEs misinter0retation o% the re%erenced material re%erenced due to its 0lacement and conteAt in this 0u2lication. 1he material is re0roduced or rewritten with their 0ermission.
2001 Microsoft Corporation. All rights reserved IEC 61966 extracts are copyright IEC !eneva "#it$erland. All rights reserved.
Foreword
The field of digital color reproduction is a congruence of several older industries merging together. These industries include broadcast television, motion pictures, slide reproduction, still photography, photofinishing, computer graphics, desktop publishing, paint formulation, presentation graphics, and graphics arts. Color science has provided a scientific foundation for all of these industries with varying degrees of significance, but each industry has extended this foundation with empirical results specialized to its particular needs. Therefore, each of these industries encompasses a significant body of knowledge with respect to color reproduction issues. Much has been written about the traditional aspects of each field 1, ,!,",#,$,% &esearchers in color science",',(,1) have continued to advance the scientific foundations over the last several decades, mostly independent of modern computer operating systems and networks. This has raised some issues in transitioning from the traditional methods to open computing environments, and the constraints they imposed. The digital color reproduction emphasis began in 1(') with the Mavica *+,-./ announcement. This led to a shift in many traditional imaging &01 budgets away from analog technologies to a strong investment in digital imaging. 2or most of these companies, this investment has yet to yield a net profit. The advent of digital color processing applications in open systems, and in particular the 3orld 3ide 3eb, forced these industries to work within open computing environments and with each other. This has created a new technology field called digital color reproduction. This new field inherited many of the older analog4based industries5 methods and standards. 2urthermore, it combines the contributions of color science researchers with the constraints imposed by software operating systems, networks, applications, and devices that make up today5s digital computing environment. The tensions among the traditional industries, together with the new digital technology, created an interesting and often conflict4filled new technical environment for digital color reproduction. Most current practitioners trace experiences directly to either one of the color or computer industries, and many claim priority in setting direction and standards in this new field. Many members of traditional industries are threatened by the idea of control of digital color processing by the operating system vendors or other traditional industries. 6mid these tensions, there has been a great reluctance to open up solutions for the betterment of the end users. 3ith a few notable exceptions, this has resulted in the current amalgam of end4user
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
solutions, none of which meet the need for transparent, predictable color reproduction, and most of which are incompatible with each other and with open, non4proprietary solutions. The current chaos in this new field can be attributed to several mid4level companies who had control in one of the many contributing industries. 7n this new field, those companies are fighting to survive. To see the effects of the more general analog to digital conversion in the generic media industries, simply try and find a family owned photofinisher store or print shop in your neighborhood. 7n the 1($)s, such individual shops were common in almost every retail center, but this is no longer true. 3hile there has been incredible consolidation in these industries, a ma8or portion of the consumer work has been 9uickly migrating to the actual consumers and their personal computers. 7t is common for home users to take digital pictures with their digital camera and to create their own greeting cards or invitations. These are 8ust two examples of the more general analog to digital media transition that has stripped the traditional analog industries of their most profitable product mix. :iven this business climate, it is not surprising that these traditional analog companies are often the most vocal opponents to open standards for digital color reproduction. This is apparent in attacks made on standards without constructive proposals and attempts to control and divert more general open standards to primarily serve a single analog industry5s needs by making the standard essentially emulate the traditional analog workflow. +ome of this resistance is due to a business strategy based on proprietary solutions instead of open technology. Many traditional industries have thrived on using trade secrets and patents to protect niche markets or to control positions. 3hile intellectual property remains a core component of technological innovation, it is important to have a productive balance with standards to help allow technologies to thrive. The ;.+. <atent ,ffice5s lack of expertise in granting software patents related to color reproduction significantly hampered progress. Today, many companies seem to be as concerned about protecting or violating intellectual property rights as they are about creating technical breakthroughs or more importantly, providing any consumer benefits. 7nterestingly, chaos and in4fighting are not new to the color reproduction industry. 7n fact, 6drian Cornwell4Clyne, in his classic monograph Color Cinematography, summarized a similar set of conditions in 1(#1 with the following statement, The public history of =processes> of color cinematography is on the whole discouraging and disconcerting, but the reader may be assured that the private history is hardly credible, and will, if ever it be made known, constitute a singular commentary upon the least rational aspects of our society and its culture.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7t can only be hoped that the digital color reproduction field will mature and become as technical and commercially successful as the color motion picture industry has become. The difficult part for the digital color reproduction field has been combining all these divergent fields together with the capabilities of current computer systems and networks in an open architecture. There is some evidence of this maturity occurring with the development of 7?C multimedia color measurement and management standards that provide basic standards in a heterogeneous device environment. 7 am fortunate to have worked in many of these divergent fields and been in the midst of several efforts to resolve the open architectural issues. These experiences have allowed me to publish many algorithms that in the past have been rarely published or explained, though well known to those skilled in this field. ,pen standards for digital color reproduction are inevitable. @oth the end users and the ma8or color vendors want results that only standards can provide. 7 hope this paper will be one more step on the path of making open standards for digital color reproduction a reality.
Michael Stokes Color Architect Microso%t Cor0oration
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
Introduction
This paper is a practical discussion of digital color reproduction issues, based on the author5s experience in this field.
WHATS
IN
THIS PAPER?
This paper combines conceptual explanations and practical guidance with an overview of devices, communication protocols, user intentions, viewing conditions, and color management systems. 7t also provides extensive references to prior art and current research. 2or areas covered ade9uately elsewhere, this paper includes overviews with references to other sources of information. The goal is to provide clear explanations of the ma8or issues involved in implementing a digital color reproduction system. ?ach component in a digital color reproduction system is described in this paper, with explanations of the underlying concepts. This paper also describes how to characterize color capabilities with respect to how the human visual system perceives color. The common protocols to communicate color between devices and across systems are also described. This paper also addresses the critical component of preserving the users5 intentions throughout the reproduction process, together with a discussion of issues and methods for compensating for each device5s viewing conditions.
WHO IS THIS PAPER FOR?

This paper will help you solve color reproduction problems, whether you are building a color printer, writing a device driver for a digital camera, attempting to calibrate a color monitor, designing an application5s interface with a color management system, or simply trying to understand and improve your system5s color performance. Many books have been written about various aspects of digital color reproduction, but most have only provided overviews of particular concepts or implementation details for particular software products. -one have provided all the relevant information to support creating a complete digital color reproduction system that works in an open environment such as the 3orld 3ide 3eb. 6 color engineer for a device manufacturer is often responsible for providing device profiles for the product along with suggestions for improving product design. This paper provides real4 world examples of how to build color4managed devices.
Device Engineers. Color Researchers. 6
color scientist or academic researcher typically understands the relevant science and theories involved in color
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
'
reproduction but is often astounded at the practical implementations and limitations in real4world products, software, and systems. This paper provides insight into how conceptual theory is often implemented in practical products and why such decisions are made. This information could lead to the development of more relevant theories and research for real4world products.
Application rogra!!ers. 6n
application programmer is fre9uently responsible for providing color support for a product with little information on how to do. This paper provides this information.
"raphics an# Me#ia rofessionals. The
graphics arts prepress separator, television video colorist, and motion picture film engineer fre9uently find themselves interfacing with a digital world that seems naAve about their field of expertise. This paper provides explanations to establish common ground between the legacies of analog color reproduction systems and the reality of digital color systems today. This includes explanations on when and how to use device profiles and when to edit them, along with comparisons of current software utilities.
$e% Designers. 3eb
page authors simply want to use color graphics and images with assurance that their clients will see the same thing they see, without any unpleasant surprises. This paper provides practical help in improving the current situation and insight into what the future will bring. The 3orld 3ide 3eb Consortium has standardized on using 7?C $1($$4 41 *s&:@/ as the default color space for BTMC, C++, +M7C and other web standards that are fundamental to web designers. This paper provides insight into s&:@ and provides methods web designers can use to verify the color content they design will be delivered with appropriate fidelity to end users. 3ithout this standard in place, it is easy for web designers to capture an image of a piece of clothing and design the product web page on their computer system and display and then have end users perceive a variety of different colors in the final browsed page. This leads to customer returns and complaints about mismatching color where if proper standards were in place could have been avoided. This is potentially unpleasant when presenting the final web design to a client and their system is different from the designers and no standard color management was used to account for this mismatch.
En# &sers. ?nd users
of applications 8ust want color to work without surprises, preferably without intervention. 7n spite of all the research and resources devoted to digital color reproduction, and the many press releases about new products and standards, most users are confused and almost never achieve the color results they want. This paper explains why it is difficult to achieve desired results today, what is being done about it, and what users can do to improve their results.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
INTRODUCTION TO DIGITAL COLOR REPRODUCTION

1igital color reproduction is derived from several traditional industries with conflicting needs, priorities, and solutions. The field of digital color reproduction can be divided into three broad categoriesD Color Science' Mo#els an# algorith!s. The various models and algorithms for color reproduction represent mostly independent pieces of digital color reproduction systems. This includes what is known about modeling the human visual system. Color Manage!ent. Color management takes the models and algorithms of color science and provides the practical engineering necessary to transform these into real world products. Color management consists of device model processing se9uences, data and metadata structures, functional structures and workflow designs. 1evice model processing se9uences are the processing se9uences take these algorithms and connect them together in appropriate se9uences to address particular devices and situations. The data and metadata structures provide a means for communicating the color information as well as the parameters of each individual model or algorithm in the processing se9uences, within the limitations of the overall software environment. The functional structures provide software support within the overall software environment to allow the data and software to communicate and function. The workflow designs provide practicable limitations for both the functional software and color reproduction results. Color ()alit*. 2inally, color 9uality provides the metrics and support to 8udge whether the models, algorithms and color management has produced an acceptable color reproduction. 7ssues and limitations in each of these categories have a profound impact on the reasonable solutions in the other categories. 7t is difficult to attempt to implement solutions in digital color reproduction without a clear understanding of all the issues involved. To properly understand individual issues in detail, one must first have a broad understanding of all the issues and how they are interrelated. @ecause digital color reproduction is a relatively young field, many issues have not been resolved.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
Color Science: Models and Algorithms

MODELS
AND
ALGORITHMS
The various models and algorithms for color reproduction represent mostly independent pieces of digital color reproduction systems and represent the basic aspects of color science. The models involved in digital color reproduction includeD The human visual system Color appearance models, to compensate for differing viewing conditions :amut mapping methods, to compensate for limited device gamuts 1evice mapping and measurement methods, to correlate device space color states with human visual system colors 6 set of user intent algorithmsEcolor enhancement and media intents Channel4generation algorithmsEblack generation Continuous4to4discrete algorithmsEhalf4toning, error diffusion, and so on Mechanism compensation issuesEbanding compensation 7nkFmedia compensation issuesEink limiting ?ach model or algorithm should have a clear input and output color space definition, including viewing conditions. This is what these transformations are aboutEtransforming colors from one color space or viewing condition into another. 7n some instances, the actual dimensions stay the same for input and output, but the meaning of the color values is 9uite different, as in the case of tone correction algorithms. The basic idea of color transformation algorithms is that they are the building blocks that can be logically pieced together in se9uence to provide analytical device models. These device models can then be integrated into functional structures along with the appropriate data structures to pass parameter information between algorithms as necessary. The device models can also be integrated into communication protocols to determine what is being processed. Many implementations today use some subset of the methods described later in this section. Most use a significant subset and only provide the crudest device modeling capabilities. ,ne reason for this is the lack of basic knowledge available to implementation engineers for whom digital color reproduction is only a secondary responsibility11. 6lthough there are researchers in this field, there is a division of experience between the implementation color engineer and the color scientist. There are few individuals in the field of digital color
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
reproduction who have formal training in color science and research, and who have written device driver color code that has shipped in products. ;nfortunately, there is also relatively little opportunity for these two groups to share the limitations with each other. 7t is hoped that this paper and can begin to rectify this situation. There is a reasonably small set of these fundamental algorithms that with proper application and se9uencing can provide accurate analytical device models for devices of very different technologies. 2or example, a set of less than fifty algorithms are more than ade9uate to provide sound analytical models for diverse devices as digital cameras, flat4bed scanners, cathode ray tubes, flat panel displays, ink 8et printers, dye4 sublimation printers and offset presses. +everal of the algorithms described later in this section are appropriate for a single class of devices such as printers, while others are more general, such as tone reproduction algorithms. +ome general algorithms, such as tone reproduction, provide a simple functionality as a single4 dimensional lookup table that can be applied in other algorithms such as flare compensation. These underlying implementation models reduce the set of basic implementation algorithms to a fairly small number. 7t is important not to confuse the implementation algorithms with the color transformation. The implementation has no associated color meaning and therefore no limitations or guidelines for use. The color transformations have explicit meanings and guidelines associated with them. This inherent meaning contained with the color transformation algorithms provide the meaning and physical association with the devices. This also allows researcher, engineers, developers, and even end users to control different aspects of the digital color reproduction process in an intuitive manner. The lack of intuitive control is one of many current weaknesses in several digital color reproduction products. 3ith proper analytical models contained within processing se9uences implemented by functional structures, and interchanging appropriate data structures within applicable workflow designs, any level of intuitive controls can be exposed or set to defaults in a straightforward manner. 7fG Eheuristic or empirical models are used instead of analytical models, Eor inappropriate processing se9uences are used, Eor functional structures restrict ade9uate implementations of the appropriate processing se9uences, Eor the data structures do not allow ade9uate interchange of information Eor inappropriate workflow designs are used,
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
Git is virtually impossible for anyone in the development process to control significant elements ade9uately, much less control intuitive aspects of the digital color reproduction system. These flaws, however, are common in most products and systems on the market today.
HUMAN VISUAL SYSTEM

The human visual system is complex and poorly modeledH yet it provides a fundamental metric and common denominator for all color reproduction systems 1 ,1!. 6lthough most references on color reproduction begin with overviews of the human visual system, few explain the fundamental part the system plays throughout the reproduction process. ?very digital color reproduction application is ultimately 8udged on how well it appears to the end user. This means that the 9uality metric for any color reproduction device, system, or application must be based on the human visual system. To create such a metric, a reasonable mathematical model of the human visual system is re9uired. Bowever, no one completely understands today how humans see color, therefore there are no complete models of the human visual system. This is an understandable state of affairs, because it is estimated that over ") percent of the human brain is spent interpreting visual information. This forces developers to approximate the human visual system. The differences in these approximations are often at the heart of problems in connecting products from different vendors together.
FIn later "ersion: add in%ormation detailing com0leAity o% the 2rain. Cornea< lens< iris< cones< %o"ea< ganglion< 3:9< Visual CorteA.G
1espite this, there are several theoretical models that provide a reasonable approximation for how the human visual system works. The most practical of these with respect to digital color reproduction are in the form of color spaces, or color appearance models that include color spaces.1",1#,' These spaces typically provide perceptually uniform dimensions of lightness, chroma, and hue in a polar coordinate system1#. 2rom these dimensions, a 9uality metric can be derived to determine how well one color matches another *delta ?/. This metric can be used to determine how well different color reproduction methods work with respect to a reference image or target colors. 6 fundamental concept in digital color reproduction is that the human visual system provides a common denominator for all color systems. This not only involves transforming from one device color space to
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
another *for example, from monitor to printer/, but also between viewing conditions *for example, change in brightness levels/. This human visual system4based color space and metric provide the ability to separate out one device model from another, and also one viewing condition from another. 7f this were not the case, engineers would have to characterize and calibrate each single combination of devices with each other and in all combinations of viewing conditions, which is impossible. 7nstead, device models provide a transformation between a native device color space and a particular human visual system4based color space such as C7?I.J or C7?C6@. @ecause all the C7?4based color spaces assume a particular viewing condition and media, it is also necessary that transformation to a color appearance space be applied to achieve independence from any device or viewing condition. 6gain, without such a color appearance space, even an imperfect one, it is impractical to provide digital color reproduction solutions in open systems such as the 3orld 3ide 3eb. The term device-independent color refers to the fact that the color and its underlying color space are not in any particular hardware device color space, but instead are based a color space that is fundamentally based on the human visual system. Thus, the human visual system provides the common denominator between devices and between viewing conditions within color reproduction systems. 7n addition to the technical and resource limitations of cross4calibrating devices, most device vendors are unwilling to provide prototypes to other device vendors to characterize. .et this is the necessary action re9uired if device vendors were unwilling to provide transformations to a device4independent color space based on the human visual system. 6lthough this might seem obvious, solutions must be feasible within the current or future business models of the leading companies in this field. @ecause many of these companies are direct competitors, this re9uires either solutions for competing in the open market, open standards to reduce potential proprietary advantages, or some combination of these two directions. This is a classic scenario between proprietary intellectual property controlling a technical field and open standards. 1epending on each individual company5s business model, market share, and in4house intellectual property portfolio and capabilities, a company5s stance on any of the issues in the field of digital color reproduction will vary dramatically and at times seem to be in direct conflict with other positions a company takes. The end4user will only benefit if progress is made in providing practical solutions, and this can be accomplished by using proprietary solutions or open standards. This theme of proprietary solutions versus open standards is common in relatively new technical fields, and is exacerbated in the field of digital
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
color reproduction by the many traditional industries involved and the size of the markets. 7n addition to transforming between devices, transforming between viewing conditions also re9uires a practical model based on the human visual system. This is the domain of color appearance models in the next section. +imply providing a transformation to a single viewing condition is neither a practical solution nor a feasible answer. 6gain, a device4and4viewing4condition4independent color space must provide a common denominator in communication color between devices and software. ?ven so, if companies adopt significantly different human visual system4based color spaces, the results will be ob8ectionably different1$. Therefore, it would be in the industry5s best interest if a single color appearance model were adopted.
VIEWING ENVIRONMENT MODELS

Kiewing environment models fall into two broad categories, simple transformations and color appearance models. @oth types attempt to model some aspects of the visual system, mostly to account for differences in viewing conditions, such as white point and luminance '. 2ormally, color appearance models are defined to be vision models that account for particular visual effects and have predictions for lightness, chroma, hue, brightness, and saturation",'. +imple viewing environment models compensate only for white point changes or surround effects. The goal of color appearance models is to predict various visual effects and provide a model that predicts color attributes of brightness, lightness, chroma, hue, and saturation accurately. The accuracy of such models is tested in psychophysical experiments that relate the physical stimulus to the human response. Bistorically, color scientists have worked with single color targets, which are the basis for C7? 1(!1 and C7? 1(%$ observers and color spaces 1%. 7n contrast, many of the contributing industries such as photography have long experimented with complex imagery1',1(. ,nly recently have researchers begun to combine these two legacies of research. This is because the computational power and digital output devices have only recently been available. The researchers who have been performing this research also view the human visual system as the fundamental common denominator for digital color reproduction. The disadvantage of this concept if implemented as an exchange format is that the image cannot be displayed easily ). 7t can be argued that the entire field of densitometry was created by a lack of computation power to create colorimetric densitometric measurements efficiently. Bowever, the legacy created by the
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
photographic and graphics arts industries in following this somewhat arbitrary measurement standard has created a strong resistance to converting these industries to BK+4based measurements. 2undamentally, a color appearance transformation transforms colors between C7?I.J values in particular viewing conditions *and sometimes media/ and a device4and4viewing4condition4independent color space. This color space possesses dimensions of hue, chroma, and lightness, and possibly brightness and saturation. This device4and4 viewing4condition4independent color space is the common denominator for all color interchange transformations. 6n excellent summary is given by 2airchild' of the concepts involved in modern color appearance models. The most common and simplest of such algorithms is the Kon Lries white point adaptation algorithm. This algorithm transforms colors in a C7?I.J color space with one white point into a C7?I.J color space with a different white point by transforming the C7?I.J values into cone space values and multiplying the resultant cone values by a ratio of the different white points in the two viewing conditions before transforming the new cone values back into C7?I.J values. This simple algorithm, which can be implemented as a single matrix multiplication, provides reasonable results in many circumstances. Bowever, many companies assume this accounts for all viewing condition issues, such as surround and luminance levels, or they simply ignore such issues. This is the primary difference between color appearance models and simple viewing environment transformations. Color appearance models account for most or all of the relevant viewing conditions, not 8ust one or two of them. 6s with the Kon Lries white point adaptation algorithm, the C7?C6@ and C7?C;K color spaces contain a very simple, yet flawed white point adaptation algorithm'. @ecause these are C7? standard color spaces, people fre9uently misuse them for inappropriate applications. The ma8or misuse occurs in white point adaptations. The issue of surround is e9ually important. @oth C7?C6@ and C7?C;K assume a medium gray surround, while the photographic and television industries have shown repeatedly 1, , ! that this is inade9uate for pro8ected transparencies *such as slides or overheads/ and for television and monitor viewing environments. 6 classic example of the resistance of colorimetry was the transition of the -T+C standards away from sulfide phosphors in the early 1(#)s after the entire standard was adopted by assuming such standards !, ". This transition moved broadcast television away from an open, colorimetric system to a metameric system almost overnight, and vendor resistance to agreeing on a standard set of phosphors can be traced to this period ". This causes users to assume falsely that the shade of red on one monitor matches the shade on another vendor5s monitor, instead of one being relatively orange to the other.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
MEASUREMENT METHODS
2re9uently overlooked aspects of digital color reproduction include the methods, standards, and relationships between the measurement of the device characteristics and the human visual system. ,ften standards provide a self4consistent means of measurement, but with little relationship to how we see color. These include issues of gloss, texture, fluorescence, and translucency, among others 1". Three basic types of color measurement devices provide results relative to an BK+D spectroradiometers, spectrophotometers, and colorimeters. The advent of CC1s has advanced the miniaturization of these devices significantly and therefore reduced the cost. 7t is important to remember that some colorimeters *CMT/ provide more accurate measurements than some spectroradiometers #. ,ne example of problems with measurement involves attempting to maintain compatibility with measurement standards and real4world conditions. 2or example, many graphics arts standards re9uire a black backing *C:6T+/, but this is seldom the case in real4world viewing. 7f compensations are made for this difference, it will adversely affect the resulting device characterization. Many companies routinely use several sheets of identical media to the target media to represent a more common real4world situation. 6nother example of difficult measurement is fluorescing papers. Most measurement standards ignore fluorescence or re9uire it be avoided in measurements. This is clearly impractical when a product media contains fluorescing brighteners. 6nother example is glossy papers. 6lthough many standards recommend a )F"# degree geometry, a )Fdiffuse geometry is more representative of how we see the media. This is a ma8or difficulty because the need for smaller measurement e9uipment virtually eliminates )Fdiffuse geometry instruments. This occurs because the diffusing sphere takes up too much space. 7n addition, as lateral diffusion begins to play a significant role in measurement accuracy degradation, the target sample sizes decrease. .et if one does not follow current measurement standards, it is impractical to exchange data with anyone else. ?ven if measurement standards are followed, cross4instrument calibration procedures are rarely implemented between companies. This results in measurements from the same instrument model being significantly different visually without being detected by measuring devices. Therefore, different companies can use the same calibration utility software, but the difference between the outputs of their measurement instruments *even if the same model/ will result in poor 9uality. 2ortunately, some instrument manufacturers are beginning to address this issue, at least within a single product line. To properly sample the device gamut with empirical models, from 1$ to a few hundred samples must be measured. This re9uires some type of
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
automated measurement stages. 2or displays, a single patch can be displayed se9uencing through a predetermined color target every few seconds. 2or printed materials and for materials to be captured with scanners or digital cameras, automated I. stages are the only practical solution. Two basic designs have been usedD The first uses a relatively small spectrophotometer that is moved, keeping the target media stationary. The second moves the target media over the aperture of the spectrophotometer. The advantage of the second is that larger instruments can be used and therefore larger patch sizes, eliminating the possibility of mechanical or microstructure artifacts causing inaccurate readings. 6 practical example of these problems can be easily illustrated, such as printing a uniform target grid using an ink 8et printer on premium ink 8et paper to create a simple three4dimensional lookup table between C7?I.J values and printer CM. values. 6 common small aperture )F"# degree spectrophotometer is used following the C:6T+ measurement standards. 2rom this table, a series of neutral patches is then printed. The surprising result is that the near white patches are visually very yellow. ;pon investigation, the following anomalies were foundD The paper was significantly fluorescent The paper measured bluish, while the samples measured neutral in direct conflict with the visual appearance 3hen measured with multiple paper samples as a backing, a slight improvement from the yellow result was found, but not significant 3hen viewed under the black backing or multiple paper backing, the yellow samples appeared close in to the same color There are two problems with this result that are not apparentD The scaling performed in the processing se9uence scaled in the lightness dimension only to the media white and black to C7?C6@ values of 1)) and ), respectively 3hen individual samples were viewed in direct comparison on different backings, there was a significant difference The first problem caused the near white colors to be out of gamut and thus mapped maintaining lightness and hue to yellow values. The very bluish media, as compared to the neutral ink set, caused this.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
'
Figure 1. Flu re!"e#$ %&i$e ' i#$ ()''i#g The second problem exaggerated the yellow chroma value by using a simplistic viewing condition algorithm contained in the C7?C6@ color space. 3hen a more sophisticated viewing condition algorithm was used in con8unction with a rotational alignment of the neutral axis with the lightness scaling, the problem was resolved. ,ne could argue that this was not a measurement problem, but a gamut mapping or scaling problem, but the basic assumption was consistent with common practice in that this was a measurement problem. This is one example of the subtleties and complexities involved in color measurements. 7n summary, it is almost impossible to measure color samples in a manner that properly simulates viewing conditions, and is consistent with any existing measurement standards.
DEFECT
AND
ARTIFACT RECOGNITION
This paper provides a starting point for investigating color image 9uality issues. 6n emphasis has been placed on photographic image 9uality. This paper does not intend to offer the best answers for every situation. 6ny suggestions for improving or expanding it would be greatly appreciated. ?ssentially, this paper should help answer the following line of 9uestioningD 7s this a good image or notM 7f not, what is not good about itM 3here did these problems come fromM Bow do they fit into the =entire> scheme of thingsM 3hat are some common approaches to fixing themM 3hat are the ramifications of these fixesM
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
Color Management
Color management takes the models and algorithms of color science and provides the practical engineering necessary to transform these into real world products. Color management consists of device model processing se9uences, data and metadata structures, functional structures and workflow designs. 1evice model processing se9uences are the processing se9uences take these algorithms and connect them together in appropriate se9uences to address particular devices and situations. The data and metadata structures provide a means for communicating the color information as well as the parameters of each individual model or algorithm in the processing se9uences, within the limitations of the overall software environment. The functional structures provide software support within the overall software environment to allow the data and software to communicate and function. The workflow designs provide practicable limitations for both the functional software and color reproduction results.
COLOR DATA
Color data represents the actual core color media information. 2or example, this could be the actual &:@ pixel values or CM.L ink levels themselves. The precision of the color data is dependent upon the capabilities of the file format that is transporting the information. Most file formats support a variety of color precision. This can range from palletized data to 14bit per channel integer data to $"4bit floating point per channel. 7t is important to match the various system variablesEsuch as bandwidth, available memory, performance re9uirements, and 9uality needsEto color data precision. 7f the color information is to be repurposed, it is often necessary to encode the information in a higher precision that is necessary for the immediate usage needs. ;nderstanding or predicting future uses of the color information helps determine what color data to encode. 2or example, if the information could potentially be used in both web4based and print4based scenarios, it is usually better to encode it as high 9uality &:@ values instead of press4centric CM.L ink levels. This is because it is usually easier and of higher 9uality to transform from &:@ to CM.L than the reverse. There exist two basic families of color spaces that almost all other colors spaces can be categorized intoH physical color spaces like sc&:@ and C7? 1(!1 I.J and perceptual color spaces like s&:@ and C7?C6@. The primary distinguishing characteristic of physical spaces is linearity
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
with respect to luminance *simple gamma N 1.)/, while perceptual color spaces are linear with respect to lightness *simple gamma N . /. ?ach of these basic types of color spaces have their appropriate uses and using one color space in processes natively intended for the other results in less than optimal color 9uality. 2or example, computing the perceptual color difference is usually done is C7?C6@, where color differences are more uniformly distributed. Computing color differences in C7? 1(!1 I.J will lead to misleading results at best. +imilar arguments can be made for visualizing gamuts in three4 dimensional C7?C6@ or C7?C;K color spaces instead of two4 dimensional .xy chromaticity charts. There are many other ways to categorize color spaces in addition to physical and perceptual, but this division appears to be most practical for engineering purposes. +ome prefer to device spaces into input and output categories, or scene and rendered spaces, but as one digs into those categories, they 9uickly break down. This has led to new terms such as scene4referred, scene4 based, scene4rendered, and many others that are difficult to explain in terms of current products and implementations. N $e* Two common operational spaces are discussed hereD The linear intensity space in which most optical and synthetic visualization operations work in 6 nearly visually perceptible uniform space in which visual4based operations, compression, and many devices work in This difference is important to remember because both types of operations are re9uired, and thus interactions occur fre9uently between these two types of spaces. Physical Data Oust as there are processes best performed in perceptual spaces, there are also processes best performed in physical spaces, such as sc&:@ or C7? 1(!1 I.J. This include processes that are fundamentally based on linear processing of light, such as transparency blending, anti4aliasing, convolution, light rendering, and so on. 7n fact, most virtual reality and computer visualization processes are performed in physical color spaces before being converted for display or output as a final step in the process. This allows for much more efficient and effective processing. 7t is already clear that these types of processing are migrating into the consumer spaces. 2or example, +ony has recently announced the Oohnny Mneumonic Kirtual &eality application available on the web and Mitsubishi has announced a <C7 visualization processor. @oth of these products fundamentally rely on physical processing for their success.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) 7
7n the future, the line between such realistic effects and processing and traditional imaging based on perceptual spaces will be blurred. To provide potential and reasonable paths forward for vendors to understand and investigate the impact of these trends, it is critical to have robust and standard support for both types of color spaces, both physical and perceptual. 6lthough one could theoretically use C7?C6@ and C7? 1(!1 I.J to address these investigations, most consumer workflows depend upon &:@ solutions, and thus s&:@ and sc&:@ are ideally suited to address these future needs.
CIE+Y,
The C7?I.J space a mathematical convenience developed from psychophysical experiences in the late 1( )s. 6t that time, it was difficult to integrate sums with negative numbers, so it was agreed to arbitrarily create a non4negative set of visual response numbers from the experimental data. The other advantage of C7?I.J is that the . dimension represents the monochrome luminance channel. 7n the 1(#)s, additional experiments were performed that led to modifications, especially in the blue areas of the C7?I.J values. This has led to two different C7?I.J colors space, one formalized in 1(!1 and the other formalized in 1($". 6lthough the 1($" standard is closer to the human visual system, legacy has kept the 1(!1 standard as the dominant one in most implementations. The differences are usually swamped by other variables in most real4world implementations.
SCRG-
The 7?C $1($$4 4 *sc&:@/ color space is designed to complement current color management strategiesEsuch as 7nternational Color Consortium *7CC/, CM.L, and s&:@Eby enabling a method of handling color in the operating systems, device drivers and the 7nternet that utilizes a simple and robust e9uipment independent color definition. This will provide good 9uality, large color gamut, and large bit precision and extended tonal range. @ased on 7?C $1($$4 41 *s&:@/, this color space is well suited for graphics arts &:@ workflows, professional digital photography, computer gamuts and computer graphics. Traditional graphics arts and prepress markets have successfully used CM.L color management solutions for several decades now. 2or many very high4end graphics arts markets, this solution still has significant advantages of all others. To take full advantage of CM.L workflow solutions, users must be well trained in this field. 3ith the advent of distributed printing and proliferation of new capture and marking technologies, there arose a need for a more flexible solution for parts of these markets. To address these needs, the 7CC proposed a color device characterization profile specification. This specification describes a profile format to help
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
in communicating color in graphics arts systems. Currently, the 7CC has one means of tracking and ensuring that a color is correctly mapped from the input to the output color space. This is done by attaching a profile for the input color space to the image or document in 9uestion. This solution is appropriate for some high4end users. Bowever, there is still a broad range of users that do not re9uire this level of flexibility and control in an embedded profile mechanism and have &:@4based, not CM.L4based, workflows. 7nstead, it is possible to create a single, standard default color space definition that can be processed as an implicit 7CC s&:@ profile. 6dditionally, most existing file formats do not, and may never, support color profile embedding, and finally, there are a broad range of uses that actually discourage people from appending any extra data to their files. 6 common standard &:@ color space addresses these issues and is useful and necessary. This approach maintains the advantage of a clear relationship with 7CC color management systems while minimizing software processes and support re9uirements. To address issues, a default &:@ color space E s&:@ E was standardized by 7?C $1($$4 41 as a standard color space solution for office, home, and web markets. The s&:@ standard addresses these concerns, serves the needs of <C4 and 3eb4based color imaging systems and is based on the average performance of cathode ray tube displays. The s&:@ solution is supported by the following observationsD Most computer displays are similar in their key color characteristics E the phosphor chromaticities *primaries/ and transfer function. &:@ spaces are native to displays, scanners and digital cameras, which are the devices with the highest performance constraints. &:@ spaces can be made e9uipment independent in a straightforward way. They can also describe color gamuts large enough for all but a few applications. .et, neither CM.L workflows nor the 7CC profile format nor the s&:@ standard color space provide a complete solution for all situations. 7n particular, the computer graphics and game industries desired a standard &:@ color space that was linear with respect to luminance. 6s a $"4bit encoding, sc&:@ allows for 1$4bit per channel encoding for computer graphics operations. This extended color standard space provides a robust solution to these needs and thus completes a complementary set of color management solutions for multimedia markets. This extended standard &:@ color space solution accomplishes these goals by extending the s&:@ tonal range and bit precision and encoding the values linearly with respect to luminance. 6ll other aspects of the sc&:@ solution are directly inherited from the 7?C $1($$4 41 s&:@ standard.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) )
The basic advantage of sc&:@ is to allow work in a large gamut, large dynamic range linear space. 2or processes that re9uire or desire such a space, the potentially simple 11 look4up4table *C;T/ conversion is a small price to pay in for the computational simplicity gained. 7n addition to the linear luminance aspect of sc&:@, the trends of most consumer devices are toward higher 9uality as is evidenced by improvements in resolution by both consumer printers and digital cameras. This also applies to bit precision as evidences by consumer scanners output 1 bits and 1$ bits per channel. 6s digital camera 1F6s become more affordable, it is clear that even consumer digital cameras will be able to output 1) bits, 1 bits, and possibly 1" bits per channel in the next five years. +ome color printers today can already accept larger bit precision than ' bits per channel. There is also active discussion in both the high end consumer digital photography market and even some of the professional digital photography markets for a &:@4based color space that truly contains not only the two dimensional color gamut pro8ections of photography film, but indeed to actual three dimensional color gamut volumes of photographic film without extensive warping of the gamut shapes. ,ne concern that has been expressed about using perceptual spaces is the use of alpha masking. Typically, computer graphics effects, including alpha masking, operate in an optical intensity environment as opposed to a visually uniform one. This mandates using a linear gamma of 1.) in most computer graphics operations, which is incompatible with the visually uniform encoding using a gamma of . . 7n a "4bit encoding scheme, encoding a mid4level gray using a gamma of 1.) would result in a digital count of "$. +uch an encoding scheme would create visually ob8ectionable artifacts such as contours. Therefore, we recommend that effects such as alpha masking be performed either prior to encoding or by decoding to a color resolution greater than " bits, and then converting into linear intensity space. The sc&:@ color space allows for negative values, which allows for some significant advantages. 2or gamut encapsulation, one might consider display gamuts to be shaped like apples, with the bulk of the color gamut in the bright colors due to the additive nature of the devices and similarly one might consider printer gamuts to be shaped like pears, with the bulk of the color gamut in the shadow colors due to the subtractive nature of the devices. 6lthough some have attempted to stretch the primaries of the apple shapes wide enough to encapsulate the pear shapes, this wastes even more space and re9uires significant movement of colors when gamut mapping from this =working space> to the actual device space. @y allowing negative values to be encoded in an unsigned encoding, this creates a black region instead of a black point. Baving a black region provides for easy encapsulation of very dark, but saturated colors while maintaining a reasonable use of gamut volume and coverage.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) 3
The negative encoding has a second significant advantage in simplifying color4processing applications. Most current applications have a hard limit of black and white at ).) and 1.). This means than any color processing that overflows or underflows these limits will be clipped or gamut mapped. +uch processes are very common in color correcting images, such as additive color changes. Baving an unsigned encoding of negative values *along with encoding values greater than one/ allows the application to persist the overflows and underflows and allows for multiple =undos> without having to cache a large number of interim history images in the process. This in turn significantly reduces the memory re9uirements for simple image processing applications that often are provided with digital imaging devices. @ecause the shape of sc&:@ is similar to the shape of s&:@ and efficiently encapsulates the shapes of printer gamuts, gamut mapping will actually be simpler and more efficient than some alternatives that significantly warp the gamut shape and thus re9uire complex gamut mapping. 6s for the negative numbers, because negative photons do not make physical sense, it is expected that when gamut mapping, these values will be treated appropriately. 3hen compressing the sc&:@ color space for transport or storage in a file format, it is recommended to convert to a .CC space using a gamma . correction to obtain optimal compression. The sc&:@ color space is intended to address the high4end consumer market problems of integrating virtual realistic solutions with more traditional perceptual imaging solutions. This includes business solutions in the entertainment, archival, medical and other industries where the developers do not want to understand or implement complex or proprietary color solutions but do want to have the flexibility to process content in the most appropriate and efficient manner. The sc&:@ color space will not replace s&:@, because s&:@ is already ideally suited for the low4end consumer and enterprise markets. 7t will also not replace 7CC solutions since it does not provide the immense flexibility contained in 7CC workflows. Perceptual Data Color encoding standards to specify ranges, limits, and viewing conditions of different color spaces have begun to play a larger role. Bistorically, the broadcast television industry has led this effort with both the -T+C and %)( encoding standards. The human visual system is usually viewed as the common denominator for color science research. This is in contrast to others who propose that a virtual device be used $.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) ,
The advantage of a virtual device is that for devices that are close to the virtual device, the image can be displayed 9uickly with little modification and produce a reasonable, if imperfect result. The disadvantage is that extra processing must be performed for accurate reproduction by transforming into and out of an BK+ space an extra time. 6nother issue in encoding is compression artifacts. Most modern compression methods re9uire an BK+ space or approximate, but unfortunately the developers seldom bother to research color spaces or color science or even standards. This was the case in the development of O<?:. The researchers decided to recommend an arbitrary .CbCr space that was incompatible with all existing standards and has since proven inferior to even C7?C6@.
CIELA2rom the establishment of the 1(!1 C7?I.J standard, researchers developed a large number of color spaces that attempted to be perceptually uniform. +uch as space would theoretically provide the basis for color difference metrics using simple mathematical constructs, such as ?uclidean distance. 7n 1(%$, the C7? established the C7?C6@ space as the premier space for these applications. +ince 1(%$, additional research resulted in an improved color difference e9uation, designated at the C7? 1((" color difference e9uation. Bowever, this space is not very easy to implement in real hardware, where an &:@ space is more suitable. 7n 1(%' the C7? adopted a standard called CPaPbP. 7t uses I,.,J as inputs. These values are linear transforms of the sensitivities of human cones. The I,.,J space has the valuable property that it can identify whether two ad8acent patches on the retina will match. Bowever if one wants to represent colors as they appear in everyday life, I,.,J space is a very poor space. The spectral sensitivities overlap. That means that a single wavelength will generate a response in all three axes. 6ll colors, with the exception of white and black, fall in the center of the color space. The C7? CPaPbP space addresses the problems with I,.,J +pace. 2irst, it scales white to black using appearance data. Munsell and many others have made e9ually spaced white4to4black scales. These gray appearance scales are fit by the cube root functions of radiance of patches in the display. +tiehl et al. showed that when these e9ually spaced displays are corrected for scattered light a log radiance function fits the data. 7n other words, the cube root function corrects for the effect of scatter in the eye. CP is the cube root of . normalized by a maximum .. This is not derived from the scene. 7t is a separate measurement of a known white. CP is scaled from 1)) to ). C7? aP is an axis perpendicular to CP that represents a red green axis. 7t is the difference of the cube root of normalized I and the cube root of
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) #
normalized .. 7t the stretches this red green axes by multiplying the difference by #)). C7? bP is an axis perpendicular to CP that represents a red green axis. 7t is the difference of the cube root of normalized J and the cube root of normalized .. 7t the stretches this yellow4blue axes by multiplying it by )). The selection of these coefficients is based on color difference data. This formula is called the ;niform Color +pace and Color 1ifference. Kiewing condition influence color appearance of colors. ?ach color space works best for the conditions used to define the space. Many of us need a uniform color space for real world viewing conditions. 6n extremely good example is the problem of mapping color on a computer monitor on colors on a reflection print. The 9uestion is whether CPaPbP is an isotropic color space in ordinary viewing condition.
SRG-
Microsoft and others led the effort to develop an international standard &:@ space for use in heterogeneous workflows. This developed into the 7?C $1($$4 41 s&:@ color space standard published in ,ctober of 1(((. This standard provides the color foundation and default color space for all Microsoft products, the 7nternet, as well as many other vendors and standards. The aim of this color space is to complement the current color management strategies by enabling a third method of handling color in the operating systems, device drivers, and the 7nternet, that utilizes a simple and robust device4independent color definition. This provides good 9uality and backward compatibility with minimum transmission and system overhead. @ased on a calibrated colorimetric &:@ color space well suited to C&T monitors, television, scanners, digital cameras, and printing systems, such a space can be supported with minimum cost to software and hardware vendors. The intent is to promote its adoption by showing the benefits of supporting a standard color space, and the suitability of the standard color space, 7?C $1($$4 41, s&:@. 6pplication developers and users who do not want the overhead of embedding profiles with documents or images should convert them to the s&:@ color space and store them in that format or its .CC derivative. Currently, there is a plethora of &:@ monitor color spaces attempting to fill this void with little guidance or attempts at standards. There is a need to merge the many standard and non4standard &:@ monitor spaces into a single standard &:@ color space. The s&:@ standard has dramatically improved the color fidelity in the desktop environment. 2or example, with Microsoft5s native support for s&:@ color space, the input and output device vendors that support s&:@ can easily and confidently communicate color without further color management overhead in the most common situations.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) 6
The three ma8or factors of this &:@ space are the colorimetric &:@ definition, the e9uivalent gamma value of . and the well4defined viewing conditions, along with a number of secondary details necessary to enable the clear and unambiguous communication of color. The dichotomy between the device4dependent color spaces *that is, amounts of ink expressed in CM.L or digitized video voltages expressed in &:@/ and device4independent color spaces *such as C7?C6@ or C7?I.J/ has created a performance burden on applications that have attempted to avoid device color spaces. This is primarily due to the complexity of the color transforms they need to perform to return the colors to device dependent color spaces. This situation is exacerbated by a reliability gap between the complexity and variety of the transforms, making it hard to ensure that the system is properly configured. To address these concerns and serve the needs of <C4 and 3eb4based color imaging systems, the s&:@ standard is based on the average performance of personal computer displays. This solution is supported by the following observationsD Most computer monitors are similar in their key color characteristicsE the phosphor chromaticities *primaries/ and transfer function &:@ spaces are native to displays, scanners, and digital cameras, which are the devices with the highest performance constraints &:@ spaces can be made device independent in a straightforward way. They can also describe color gamuts that are large enough for all but a small number of applications. This combination of factors makes s&:@ space uni9uely well suited for wide adoption because it can both describe the colors in an unambiguous way and be the native space for actual hardware devices. This, many readers will recognize, describes in a roundabout way what has been the practice in color television for some "# years. This proven methodology provides excellent performance where it is needed the mostD the fast display of images in C&T monitors.
METADATA STRUCTURES
7n recent years, there has been some progress in standardizing digital color reproduction data structures, much of this lead by the 7?C. These 7?C standards or drafts include color spaces s&:@, sc&:@, and s.CC. 7n addition, the 7nternational Color Consortium developed the original 7CC profile format in 1((". To communicate what type of data is being exchange, some additional descriptive information must be provided to determine all of the details that are incorporated into the color space, such as dimensions, viewing conditions, gamma, white point, and so on. There are two basic
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) '
approaches in communicating these re9uired details, either implicitly specifying the information or explicitly embedding the information in the workflow. Most s&:@ workflows implicitly associate all of the necessary details about the exact meaning of the color data in the file format or device itself. 7CC workflows by definition must explicitly associate all of these details along with the color data or device. Implicit Metadata The simplest and most robust method to communicate color information is to implicitly associate color metadata. 2or example, the ?I72 file format by definition defaults to the s&:@ color space, as does <-:, @M<, BTMC, <12, C++, and many other file formats, markup languages and document description formats. This allows all of the necessary color information to be determined by reference. ,ne philosophical approach to exchanging color information is to encode all intermediate information into a single, well4defined virtual device, such as a monitor or a printed page. The s&:@ color space is by far the most prevalent use of this method. The advantage of this approach is to provide easy access to data by devices that are close to the standard reference.
STANDARD REFERENCE MEDIA DESCRIPTIONS

D./ WHITE POINT 7n order to be consistent with common practice in most consumer electronic imaging fields *television, computers, digital photography/, a white point of 1$# was chosen. Two other alternatives would have been to not define the white point at all or choose a more press4centric white point such as 1#). 6 white point was defined in order to provide ease of use to implementers. 2or many implementers, the only processing between 7?C $1($$4 4 *sc&:@/ and 7?C $1($$4 41 *s&:@/ will be a simple tone correction, as it is in B1TK. 7f a white point correction had also been re9uired, this would have led to many complexities such as which white point adaptation to use and how to communicate this down stream effectively. Currently the 7CC has only 8ust begun to discuss this problem in their more complex workflows. @y defining the white point to 1$#, this processing is left to the expertise of the device vendor. 6 white point of 1$# was chosen instead of a more press4centric 1#), because in many cases, 1$# has been show to provide a better display match to typical or even professional prints. +everal color management experts have compared different white points in their professional work and found 1$# to be superior for a working space or display work. 7n addition, the 1#) white point is usually limited to press4centric standards and is thus as the end of the workflow.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) (
Therefore, standardizing upon 1$# white point for display4centric workflows allows maximum flexibility both compatibility with other industries and redirecting content to different output media. GAMMA AND THE DESIRED CRT GAMMA OF 0.0 The choice of gamma . for perceptual space is consistent with legacy implementations and standards, and with the ideal characteristics of C&Ts, including professional broadcast monitors. 6 simple analysis of this gamma value has also shown it is very close to C7? CP when viewing conditions for softcopy display are accounted for. These conditions are defined in 7?C $1($$4 41. This makes the choice of a gamma of . ideal, based upon compatibility, legacy, and scientific evidence.
SRG- AND
ITU1R -T.234 COMPATI-ILITY The 7T;4& @T.%)( primaries were chosen to be consistent with s&:@ and common practices in most consumer markets today. 7n addition to consistency with current television practice, this choice also simplifies gamut mapping issues. :amut mapping remains a complex issue that has active research interest from academic institutions. 6 simple viewpoint is to imagine device gamuts as fruit shapes. 2or example, a display additive device gamut is similar to an apple, while a printer subtractive gamut is similar to a pear. ;nfortunately, many extended gamut color spaces assume all devices are similar to oranges and perform either very simple or very proprietary gamut mapping. 7n an open workflow, this can result in gamuts similar to a pineapple with unexpected sharp points around the gamut and unwanted extrusions in specular highlights. The approach to this problem taken by the maintenance of 7?C $1($$4 4 1 is to use the same gamut and encapsulate the larger gamut using additional bit precision. C. 1avid Tobie has named this the =baby bear, momma bear, papa bear> approach to large gamut color space workflows. The advantage is a significant simplification of gamut mapping issues as the cost of additional bit depth or precision. There has been some confusion in previous versions of the s&:@ standard with respect to compatibility with the 7T;4& @T.%)( standard. Many readers have also expressed concern for compatibility with the %)( recommendation in general. Bowever, the 7T;4& @T.%)( standard can be somewhat confusing for many readers. This section attempts to clarify and reduce this confusion. 2or a single color space to achieve acceptance, it must be ob8ective, that is, have a tightly defined relationship with the C7? standards. 3e are fortunate to have obtained in 6pril 1(() unanimous worldwide agreement on a calibrated nonlinear &:@ space for B1TK production
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
) 9
and program exchangeD &ec. 7T;4& @T.%)(. This recommendation specifies the encoding of real4world scene tristimulus values into a standard monitor &:@ color space assuming a dark viewing condition. These parameters should be used as the basis for the s&:@ color space, but with a dim viewing condition that is closer to typical viewing environments for computer4displayed imagery. Bowever, the 7T; specification is rather vague on defining the target monitor. The 7T;4& @T.%)( standard specifically describes the encoding transfer function for a video camera that when viewed on a =standard> monitor will produce excellent image 9uality. The implicit target of this encoding is a standard video monitor whose transfer function or C&T gamma is not explicitly delineated. 7nstead a typical monitor setup is assumed. The s&:@ standard a standard monitor setup that is compatible with the 7T;4& @T.%)( encoding standard. This is illustrated in 2igures , !, and ". 2igure is derived directly from the 7T;4& @T.%)( standard. This standard provides mathematical methods to transform from tristimulus values of the scene by using a video camera into a reference monitor device space.
+ig)re 2
2igure ! expands the implicit step of these methods and shows the transformation between the original scene tristimulus values into the target monitor tristimulus values. @ecause these two viewing conditions are different, an implicit compensation is made to account for these differences *for example, flare and ambient luminance/. To provide an independent monitor reference color space, the monitor compensation methods must be extracted from this confounded compensation. This is precisely the goal of the s&:@ color space.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 7
+ig)re ,
2igure " illustrates both the s&:@ color space and the extraction of the monitor specifications implicit in the 7T;4& @T.%)( standard. @y producing such a monitor space, one can transfer the 7T;4& @T.%)( encoded signals to other devices. @y building on this system, the s&:@ color space provides a monitor definition that can be used independently from the 7T;4& @T.%)( standard while maintaining compatibility. This allows for the well4 defined transfer of color information across the 3orld 3ide 3eb.
+ig)re -
This s&:@ standard essentially defines the second part of this transformation between the target &:@ monitor space and the monitor C7?I.J tristimulus values in a dim viewing environment. This is re9uired to maintain a consistent monitor centric color reproduction process that is typical of the 3eb, and is consistent with recommendations of the 7CC. This is also consistent with maintaining the preferred reproduction encoding of the 7T; standard. 7n summary, there has been some concern with the choice of a . C&T gamma with a 1.) C;T gamma as opposed to a 1.#%1 * . F1."/ or a 1. (" * . F1.%/ display gamma. 3e feel that there are many reasons to support a . C&T, includingD 6 better fit with 3eber5s fraction Closer to native C&Ts gamma Compatibility with a large legacy of images Compatibility with numerous standards Consistency with a larger market of displays Compatibility and consistency forD 6nalog televisionH analog video 6pple Macintosh video imagery CC7& $)1 images 1igital TKH B1TK ?I72H T722F?<
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
<Cs with #$Q colors and their desktop color schemes and icons <hotoC1 ;ltra4large image collections ;nix workstations Explicit Metadata ICC
THE HISTORY
OF THE
ICC
7n recent years, hopes of resolving the chaos of color reproduction in open systems and the 3orld 3ide 3eb have slowly become focused on de facto industry standards. 6pple Computer led an initiative starting in the spring of 1((!, known as the Color+ync Consortium, to resolve this chaos. ,ver the next eighteen months, this initiative produced an open, cross4platform device color characterization profile format specification based on the 6pple Color+ync profile format, and it set the groundwork for unambiguous interaction among color devices and vendors in open systems. The following year was spent transforming this informal consortium into the 7nternational Color Consortium as a formal, non4profit organization. The next eighteen months were spent establishing clear goals for the consortium and struggling with intellectual property issues. 2ollowing this effort, the next six months saw the genesis of new work to inventory all known problems with the 7CC specification, create a set of reference implementation, and establish guidelines for conformance testing. 7f these initiatives are successful, it appears that the 7CC might finally meet many of its ambitious initial goals. 6s founding chairperson of the Color+ync Consortium, color architect for Color+ync and recent past chairperson of the 7CC, 7 have been intimately involved in all of the developments described earlier. 7 have also been involved in other related open color standards activities. This paper represents my personal viewpoint on these developments and the 7CC in general. 7t does not necessarily represent the official position of the 7CC, 6pple Computer, or Bewlett4<ackard Company.
INTRODUCTION 5 -EFORE COLORSYNC

The field of digital color reproduction is a congruence of several much older industries merging together. ?ach of these industries has its individual aspects of color reproduction that have evolved within the constraints of that industry5s particular production workflows. Color science has provided a scientific foundation for all of these industries with varying degrees of significance, but each industry has extended this foundation with empirical results that are specialized to its particular needs. Thus, each of these industries individually
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 )
encompasses a significant body of knowledge with respect to color reproduction issues. Much has been previously written about the traditional aspects of each field, usually from an analog point of view. 7n addition, researchers in color science have continued to advance the scientific foundations over the last several decades 9uite independently from any of these industries. Bowever, until 9uite recently, most of these efforts have also been independent of modern computer operating systems and digital networks. This has caused significant transition problems between the traditional methods and the constraints imposed by open computing environments and in particular the 3orld 3ide 3eb. The advent of digital color processing applications in open systems, and in particular the 3orld 3ide 3eb, has forced all of these industries into working within open computing environments and with each other. This created a new technology field, digital color reproduction. This relatively new field has inherited many of the methods and standards from each of its contributing industries. This is in addition to contributions from researchers in the color science community and in combination with the constraints imposed by the various software operating systems, networks, applications, and devices that compose the digital computing environment today. The tensions between the traditional industries with each other, and along with the new digital technology, have created an interesting and often conflict4filled new technical environment for digital color reproduction. Most of the current practitioners trace their experience directly to either one of the color or computer industries listed earlier and many claim authority in setting direction and standards in this new field. +ome traditional imaging companies feel threatened by the control of color by operating system venders or other traditional imaging industries. There has been a great reluctance to open solutions for the betterment of the end users. This has created an amalgam of solutions for end users, none of which have fully answered the desire to have transparent, predictable color reproduction and most of which are incompatible with each other. 7n particular, the current chaos in this new field can be attributed to a few companies who had control or near monopolies in one of the many traditional analog imaging industries. 7n this new digital color reproduction field, these companies are fighting to survive. These same traditional analog companies are often the most vocal opponents to open standards for digital color reproduction. This is exemplified by attacks on open standards activities without constructive counter4proposals. +ome of this resistance is due to a strong business strategy based on proprietary solutions instead of open technology. Many traditional industries have thrived on using trade secrets and patents to protect small niche markets or monopolistic positions. The ;.+. <atent 6gency5s lack
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 3
of control and expertise in granting software patents in the fields of digital color reproduction has also hampered progress in this field. Today, many companies seem to be as concerned about protecting or violating intellectual property rights as they are about creating technical breakthroughs. 3hether this is due to the huge unreturned investment in digital technology or simply to entrenched attitudes of traditionally being in control and monopolizing their markets will probably never be known. The 7CC initiative circumvented many of the traditional standards processes. This was done intentionally after several previous efforts on standardizing open color communication within the 7+,F7?C processes failed, including ,+C6, :6C17, and ,16. 7t has been found that there exist strong contingents within the international standards organizations that seem opposed to open standards that can be used and shared by all on a royalty4free basisD 7+, and 7?C standards committees are based around traditional industries and strongly support their industry5s needs and perspectives. There seems no representative of desktop publishing or home publishing industries within the 7+,H the traditional graphics arts concerns dominate this forum and have successfully published many excellent standards for this industry. .et this is still inappropriate to web publishing in many cases. 6lthough OT6: has been formed to consider some cross industry concerns, it has proven slow relative to the pace of industry needs and still limited to its representative industries of professional photography, graphics arts and paper. The C7? would seem to provide an ideal forum for these issues, but has shown no active interest in color management and has formal timelines that are at odds with color digital product development cycles. +till, this initiative has publicly committed to pursue the traditional standardization process after proving their validity in the open marketplace despite some resistance from the traditional standards community. The story behind the development of these initiatives provides some insight into the technical, legal and business barriers to revolving color reproduction issues in open systems. @y late 1(( , there were a number of proprietary, non4compatible color management solutions, including those offered by 6dobe, 6gfa, ?27, Lodak, Bewlett4 <ackard, Cinotype4Bell, <antone, Tektronix, and Ierox.
THE COLORSYNC CONSORTIUM

7n 1((!, 6pple Computer led an initiative to resolve this chaos that became known as the Color+ync Consortium. ,ver the next eighteen months, this initiative produced an open, cross4platform device color characterization profile format specification based on the 6pple Color+ync profile format and set the groundwork for unambiguous interaction among color devices and vendors in open systems.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 ,
7n late Oanuary 1((!, 6pple Computer introduced the first operating system4based color management system, Color+ync. :erry Murch of 6pple presented an overview of this product and architecture at the 2,:&6 +ymposium in the March 1((! in Munich. ,fficials of 2,:&6 convinced representatives from 6pple and 6dobe *and others/ to meet and discuss the possibility of providing open color management in the operating system based on 6pple Color+ync architecture. 6t +eybold *6pril 1((!/, +eybold officials, in collaboration with 2,:&6 officials, convened a second meeting with an attendance of around fifteen people. 6t this meeting 6pple and Lodak presented competing architectures for consideration as the foundation of a new, cross4platform, cross4vendor open color management solution. 7t was agreed by those present to support 6pple5s architecture, but to reserve final 8udgment until the end of summer and conditional upon 6pple5s commitment to put this work in the public domain. @y 6ugust, in con8unction with +7::&6<B, 6pple presented a complete architecture to an audience of about !) companies. 6pple restated that they were willing to put this work in the public domain. This work addressed most of the previous technical reservations of many developers. <rior to the next meeting in +eptember, after discussions with Lodak and Microsoft, 6pple agreed to concentrate initially on 8ust the profile format specification and not a cross4platform application programming interfaces *6<7/. 7n +eptember 1((!, 8ust prior to the +eybold Conference, 6pple presented a complete architecture, 6<7 and profile format specification to approximately $) imaging companies. 6s with previous meetings, this meeting did not exclude anyone from attending and encouraged everyone to provide feedback into the process and specifications. There was overwhelming support from those in attendance to continue this effort. 6t this point, it was felt that limiting the number of active participants would enhance the implementation of the actual profile format specification. 1uring his presentation the following week at +eybold, :erry Murch announced the formation of the Color+ync Consortium to create an industry standard profile format. @y this time, the profile format specification had been through twelve revisions. 1uring the next three months, 6pple worked closely with Lodak, and 6dobe, to insure a mutually acceptable specification was created. 7n Oanuary 1((", the first consortium meeting was held and the founding members were established. +ince this effort had been operating system centric, all of the ma8or operating system vendors were invited to participate. 6pple, +un, +:7 and Microsoft agreed to actively participate. 7@M, Bewlett4<ackard and 1igital declined to participate. 6dobe was invited to participate since <ost+cript was viewed as an application platform. 6gfa and Lodak were invited to attend to offer expert advice from the graphics arts standpoint since both had publicly
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 #
stated support for the new profile format. 2inally, a representative of 2,:&6 was invited to acknowledge their ongoing contributions to this effort. 2rom the beginning there existed strong disagreement over the scope of the consortium, but all agreed to attempt to implement the profile format specification as a starting point. The founding members strongly felt that a key missing concept from previous color management and communication solutions was a clear definition of all the variables involved in characterizing a device. The consortium defined a working concept called a profile connection space *<C+/. This concept explicitly describes all of the relevant parameters involved in describing colors in a color space. +imply because this effort originated in the graphics arts and desktop publishing field, a set of graphics arts standards was chosen as the basis of the profile <C+. The central importance of the <C+ is due to its role in profiles. 6ll input, display and output profiles connect between the canonical device space and the 7CC <C+. ,ne member strongly felt that a fixed, 1#), ideal print <C+ was the only possible choice that would work, despite strong opposition and proof to the contrary in successful product offerings from several other member. 6dobe brokered a compromise whereby the issue could be revisited at a future date. The fixed <C+ created the possibility of a color management module *CMM/ that simply interpolated between source and destination profile. Bowever, it also mandated a fixed color appearance model that has since been proven to be significantly flawed. 2rom its inception, a close association with 6dobe <ost+cript Cevel color processing pipeline has been an explicit guideline for the profile format development. 7n particular, the input and display models of Color+ync were combined with 6dobe <ost+cript Cevel printing model to form a robust profile format specification. 2inally, three4 dimensional support was added to the input and display profiles for future extendibility. @y Oune 1((", the twentieth revision was presented to the founding consortium members and deemed acceptable to base product development on.
FOUNDING
THE
ICC
The following year was spent transforming this informal consortium into the 7CC, a legal, formal, non4profit organization. The 7CC bylaws state thatD Gthe 7nternational Color Consortium was established in 1((! by eight industry vendors for the purpose of creating, promoting and encouraging the standardization and evolution of an open, vendor4 neutral, cross4platform color management system architecture and components. The 7CC has produced a specification thatD
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 6
Gdescribes the 7nternational Color Consortium <rofile 2ormat. The intent of this format is to provide a cross4platform device profile format. +uch device profiles can be used to translate color data created on one device into another device5s native color space. The acceptance of this format by operating system vendors allows end users to transparently move profiles and images with embedded profiles between different operating systems. 2or example, this allows a printer manufacturer to create a single profile for multiple operating systems. To accelerate the legal formation of the 7CC, the ,pen:C Consortium bylaws and membership agreement were taken as a guideline. 1espite this, the creation of a new consortium by competing companies proved to be an arduous and lengthy process that involved many revisions of the bylaws and membership agreement by each company5s legal team. The result was a formal organization led by some of the most power operating system and color management system vendors in the world. This clearly established the authority of the 7CC and membership was opened up to all that supported its goals and paid the nominal membership fees. Today, over fifty of the world5s premier color imaging companies have chosen to become members with even more companies choosing to support the 7CC profile format in their products. 7n parallel to this legal development, many minor technical corrections to the profile specification format were made.
ICC GOALS POLICY
AND
INTELLECTUAL PROPERTY
The next eighteen months were spent establishing clear goals for the consortium and struggling with intellectual property issues. +everal 7CC meetings were devoted almost completely to crafting clear and forward4 thinking goals for the consortium. The broad industry adoption of the 7CC profile format has provided a good first step into clarifying the unambiguous communication of color information in open systems that are now prevalent throughout industry. The 7CC goals provide a good indication of the future direction of this consortium. These goals were formally adopted at Ouly 1(($ meeting in <alo 6lto and are as followsD
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 '
1. The color management system should scale from automatic to sophisticated user intervened conditions. . The 7CC will incorporate new technology and standards in a timely manner. 7CC work will be offered to the relevant standards body for consideration as an 7nternational +tandard !. To create, promote, and encourage the standardization and evolution of an open, vendor4neutral, cross4platform color management system architecture and components. ". The 7CC will define a baseline CMM model and baseline CMM implementation that can be part of the specification that will provide the same results with the same numerical data. #. The 7CC will standardize on an appearance model*s/ to get from source to <C+ and <C+ to destination. $. Colorimetric reproductions *absolute and relative/ should be exact within the tolerance of the device within the device5s gamut.
CURRENT ICC STATUS
AND
DIRECTION
The last six months have seen the genesis of new work to inventory all known problems with the 7CC specification, create a set of reference implementation and establish guidelines for conformance testing. There are several other working groups within the 7CC to address specific technical issues. The <roblem 6ssessment 3orking :roup is led by Microsoft and chartered to list and prioritize all known problems with the 7CC specification and, if possible, suggest solutions to these problems. The &eference 7mplementation 3orking :roup is led by Bewlett4 <ackard and charted to create a reference CMM and profile creation implementation from the specification, taking input from the <rogram 6ssessment 3orking :roup for suggested solutions and priorities. The Conformance Testing 3orking :roup is led by <olaroid and chartered to create guidelines for conformance testing of profiles and investigate conformance testing for CMMs. :iven their very slow progress over the last seven years, it is doubtful that any of these or other 7CC working groups will be able to overcome most of the current problems and resistance to universal adoption by users and developers. 2ortunately, 7?C has provided a complementary color management solution to deal with this, 7?C $1($$4 41, the s&:@ standard default color space.
OPEN ISSUES WITHIN
THE
ICC
LIMITATIONS OF THE FI+ED PCS Cimitations of the 7CC5s fixed <C+ include the followingD
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 (
7t assumes color appearance models mix and match without 9uality degradation. @y convolving in human visual system data, it prevents the calibration or editing of third4party profiles, including monitor calibration. 7t propagates serious user interface problems by forcing new profiles for every parameter variation. 7t propagates storage issues due to the number of profiles. EM-EDDING ICC PROFILES IN IMAGE FILE FORMATS 7n addition to being stored as standalone files on a computer system, 7CC profiles can be embedded into some image file formats to maintain color fidelity data with the image data. Currently, embedding specification for Microsoft5s @M< format, 6dobe T722, ?<+, and 6pple <7CTs are specified. -USINESS GRAPHICS IN THE ICC PROFILE FORMAT @usiness graphics are a problem for the 7CC device4independent structure due to business graphics5 inherent device dependent implementations. Currently, the 7CC does not provide a clean solution despite implicit claims of the saturation rendering intent. 6 variable <C+ would help alleviate this issue, providing enough information to the CMM to create a pseudo4device dependent transform. E+TERNAL ACCESS OF ICC PROFILES The 7CC has not mandated that 7CC4capable output drivers provide externally accessible profiles, only that the drivers be able to read 7CC source profiles. This prevents third parties5 CMM access to these drivers. Conclusion 7t is important to remember what the 7CC is not the =holy grail> of color that will automatically manage the user5s color issues in and across computer systems. The 7CC profile format specification is not a comprehensive solution for all or even most common color management issues today. 6lthough there is an informal effort to align operating systems5 color management 6<7s across platforms, this is not a current goal of the 7CC. 3ithout strong support and adoption of application vendors, the 7CC is a weak color management solution, at best.
PROCESSING SE6UENCES
The processing se9uences take these algorithms and connect them together in appropriate se9uences to address particular devices and
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
3 9
situations. 6lthough large three4dimensional tables relating C7?I.J to some three4dimensional device space such as &:@ or CM. may seem simple to implement, this approach is fundamentally flawed if performed naively. 7t is sometimes assumed that one can simply take a large sample of target colors, measure them after processing them in a native device space, and then create a large three4dimensional lookup table to relate the two. This assumes a well4behaved deviceEwhich is rarely true of any raw digital4color deviceEand imposes significant conse9uences on other aspects of the workflow design. This is because a separate lookup table would have to be provided for every combination of device states, such as media type, dot gain, gamma value, and user preference. 6n alternative is to break down the device color reproduction model into a large number of individual steps and only combine them when practical. This also provides insight for device vendors to optimize other optical, electronic, chemical, and mechanical aspects of devices with greater confidence in why changes are necessary. Capture De ices Capture devices such as scanner and digital cameras have a difficult task in simulating precisely how the human visual system =sees> color imagery. This involves dynamic range control and white point ad8ustments, along with many other non4color related functions. 1igital cameras are physically limited in their ability to simulate the human visual system. This is often due to cost considerations that impact resolution and color processing. 7n any case, a typical digital camera processing se9uence looks like thisD 1. <review to estimate luminance level . Capture image in a mosaic CC1 array !. ?stimate white and black points ". Cinearize range #. Correct white and black points *dynamic range compensation/ $. 1e4mosaic image %. 6d8ust tone reproduction '. 6d8ust color balance (. 6d8ust saturation 1). ?ncode into a different color space
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, 7
Print
+canning a print is relatively the easiest task for capture devices in that the dynamic range in limited and the white point is usually present in the imagery. 6 typical processing se9uence is as followsD 1. <review to estimate white and black points, tone reproduction, and saturation . Cinearize range !. Capture image with corrected white and black points ". &emove media intent #. Correct tone reproduction $. 6d8ust color balance %. 6d8ust saturation '. ?ncode into a different color space
!ransparency +canning a positive transparency is involves reducing the dynamic range in an imperceptible manner. 6 typical processing se9uence is as followsD 1. <review to estimate white and black points, tone reproduction, and saturation . Cinearize range !. Capture image with corrected white and black points ". &emove media intent #. Correct tone reproduction $. 6d8ust color balance %. 6d8ust saturation '. ?ncode into a different color space
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
"egati es
+canning a negative transparency involves un4building the negative film characteristics and estimating what the original scene color characteristics were. 6 typical processing se9uence is as followsD 1. <review to estimate white and black points, tone reproduction, and saturation . ?stimate negative film characteristics !. Capture image with an optimized range and end points ". &emove negative film characteristics and convert to a positive image #. Correct tone reproduction $. 6d8ust color balance %. 6d8ust saturation '. ?ncode into a different color space (. <ossibly add a different media encoding
Displays 6 display pipeline is as followsD 1. ?stablish viewing parameters . Convert from nonlinear rgb values into linear &:@ values with a proper transfer curve !. Convert from linear &:@ values into I.J values ". 6d8ust for flare #. 3ith proper viewing parameters and color appearance model, convert to an BK+ space
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, )
Printers
6 printer pipeline looks like the following, assuming an BK+ space to start withD 1. Convert from an BK+ space into printer viewing environment . Convert to C7?C6@ values !. Make media rendering ad8ustments ". Make user preference ad8ustments #. Convert to CM. continuous tone device values $. Map out of gamut colors %. :enerate black color channel information '. Convert to device color space with half4toning or error diffusion (. <rovide any ink limiting necessary and other inkFmedia compensations 1). Make any arbitrary color ad8ustments for pure colors
FUNCTIONAL STRUCTURES
The functional structures provide the software support within the overall software environment to allow the data and software to communicate and function. This is the area of color reproduction systems that has remained most proprietary while open standards could provide tremendous benefit to users. The ,pen <repress 7nterface *,<7/ is one example in which a functional interface was standardized in a page description language to help the user easily reference source imagery in a network. This standard has proliferated across platforms and application vendors as the industry de facto standard for dealing with =for placement only> *2<,/ imagery. There are several different areas of functional structures, includingD Color management frameworks <rofile management methods Color management modules 7nterfaces with graphics libraries 1evice drivers Calibration utilities ,<7F2<, utilities
Color Management Framewor#s and APIs Color management frameworks *CM2s/ provide the set of 6<7s to manage color among devices and applications. This framework resides in the gray area between structures and architectures. Most frameworks today are a collection of structures and functionality with a very loose and vague underlying architecture. Currently, the 6<7s among the different operating system vendors all differ.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, 3
PROFILE MANAGEMENT METHODS

,ne of the ma8or services of CM2s is the ability to retrieve, store, modify, and 9uery collections of profiles. 3ith an abundance of profiles potentially available for any device and situation, it is usually unacceptable to place the entire burden of choosing the appropriate profile on either the end user or the application developer who might have little color expertise. <rofile retrieval is the primary responsibility for profile management methods. The method should be able to retrieve a specific profile and provide it to the re9uester. <rofile 9uerying is necessary to create a practical framework implementation. ,therwise, end users would be re9uired to choose from a potentially impracticable long list of available device profiles. <rofile 9uerying allows these available profiles to be reduced to only the reasonable profiles *for example, printer profiles/ or even a single preferred profile. <rofile storage is necessary to add profiles to the set provided by the framework. This is necessary to add new devices, edit device profiles, and calibrate devices. <rofile modification facilities allow profile calibration, characterization, and creation utilities to interact seamlessly within the framework.
COLOR MANAGEMENT METHODS

Color management modules profile the execution functionality of the color management frameworks to transform from one color space into another. These modules fre9uently provide additional utility functions to combine multiple profiles or invert profiles. The complexity can vary from simple interpolation engines to sophisticated color appearance and device modeling algorithms. Color management modules under the current implementations of 7CC and 7CC4compliant frameworks are basically simple three4dimensional interpolators. This forces some other aspect of the workflow design to handle numerous profiles that account for every combination of device states %. 6n alternative is an intelligent CMM in which the necessary device characterization parameters are provided in profiles along with the viewing parameters. The CMM can then translate direction from one color space into another without extra processing re9uired by a virtual device or extra profiles for all device states. The processing performed by such a CMM must be standardized, especially the color appearance model aspect.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, ,
COLOR CALI-RATION UTILITIES

There are numerous third4party color calibration services available. The robustness and 9uality of most remain untested. +canners 1isplays <rinters 1igital cameras
OPI7 FPO UTILITIES

6ldus created the ,pen <repress 7nterface protocol on top of <ost+cript to provide 2<, preview images in page layout programs to avoid transferring extra data until necessary. The manner in which the color information is transferred with these previews is also poorly defined. Page Description $anguages as Functional Structures This section describes page description languages as functional structures. 2or another viewpoint, see =<age 1escription Canguages as 1ata +tructures,> earlier in this paper.
CAPTURE DEVICE DRIVERS

6 typical capture driver processes the firmware output into a structure acceptable for the operating system. T367- drivers are representative. This is also the place where vendors often put last minute color ad8ustments.
DISPLAY DEVICE DRIVERS

1isplay or graphics adapter drivers maintain the video memory interface and, most important with respect to color, the gamma and any primary matrix controls. Many graphics adapter drivers are capable of changing the monitor color calibration, including gamma and white point ad8ustments.
GRAPHICS LI-RARIES
:raphics libraries, such as Microsoft :17 and :17Q, 6pple Ruick1raw and Ruickdraw:I, and ;nix I, provide the fundamental environment for all color functionality. This fact often imposes strict limitations on what is theoretically possible or wanted in color manipulation, and what can be implemented practically.
PRINTER DEVICE DRIVERS

<rinter drivers interface between the graphics library or <1Cs and the printer firmware, translating page layouts into intermediate color information the printer device understands. This can be either the <1Cs or some lower level interface.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, #
7n any event, there is often a color transformation before the colors are transferred into the printer firmware. The driver can be thought of generally in two parts, the host4based software that is an operating system extension and the printer4based firmware that is contained in silicon within the printer device.
COLOR IMAGING WOR8FLOWS

3orkflow is the common commercial term in several industries *for example, graphics arts, publishing, and photofinishing, among others/ to describe the se9uence of processing steps involved in creating an image. 3ithin Bewlett4<ackard, imaging or color pipeline is most often used to describe some algorithmic stages of this process. 6 typical workflow starts with designing the image layout or, more commonly, a general layout that contains text, graphics, and images. 6fter a design is done, the image is =ac9uired> or captured into a digital format. This may be by way of digital camera, photographic film scanner, network download, or a variety of other means. This digital image is then edited to correct any ob8ectionable artifacts, which could include layout problems, ac9uisition artifacts, and printing compensation, among many other possible problems. 2inally, the image is distributed, usually by way of hard4copy print or soft4copy formatted image. ?ach of these stages mandates a number of difficult decisions and compromises on image 9uality, which are described briefly here. 6s an image progresses through a workflow or imaging pipeline, a wide variety of degradations of color 9uality affect both the sub8ective and ob8ective color image 9uality characteristics described earlier. ?ach stage or section will discuss how 9uality characteristics are degraded or limited by that particular process.
F0ro"ide in later dra%t: Com0uter Science -or.%low H $e"ice Centric :ra0hics ArtsI -or.%low H *ress Centric *ress 4nly Scanner to *ress Color !lectronic *rinting Systems CC!*SD Ado2e *hotosho0 H $is0lay Centric Color Management Systems H 4S Centric ColorSmart H $e"ice Centric Csort o%D Inherent Arti%actsG
Design%$ayout The design and layout stage is where the concept of the image is thought out and planned. 7n commercial workflows, an artist or photographer usually does the design. ,ne of the primary rules is to =know your message, know you customer, and make sure they match.>
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, 6
&amifications of this rule include choice of sub8ect matter, overall layout, and composition. ,ther design and layout decisions include lighting balance of the sub8ect matter, gray balance, contrast and tone levels, colorfulness, and hue balance of colors in the scene. 7deally, these choices and layout designs will not be degraded by the rest of the workflow. Bowever, as we shall see, this is not the case. 7n fact, experienced designers take the workflow into consideration when choosing sub8ect matter and colors in their images in an effort to minimize unwanted image 9uality artifacts. Ac&uisition%Image Capture 7mage ac9uisition is the method of ac9uiring a digital image from a digital or analog source image. :enerally speaking, image ac9uisition from analog sources involves a series of physical processesD ,ptically focusing the image <artitioning the image into separate color channels using colored filters <artitioning the image into separate spatial pixels Ruantizing the image into a digital format 7n addition to the inherent artifacts of each of these processes, the process of manipulating and storing the image on a computer system produces some practical limitations that lead to other image 9uality artifacts. ?ach of these steps has a direct impact on image 9uality, which can vary from imperceptible to grossly unacceptable, depending on how well they are implemented. 2urthermore, in addition to these general artifacts, individual devices such as digital cameras, CC1 scanners, and <MT scanners each possess their own uni9ue problems. 7f images are not directly captured from the original scene, and secondary sources such as photographic prints, half4toned prints, and videotape are used, yet another set of artifacts are incorporated into the workflow process.
OPTICAL ARTIFACTS
The primary goal of the optics in any color image ac9uisition device is to focus the original scene or ob8ects accurately onto the sensor or image plane with a minimal optical artifact. <roblems with depth of field, depth of focus, flare, barrel and pincushion distortion, field curvature, diffraction, spherical aberration, astigmatism, coma, spherochromatism, and lateral color are among the most common technical difficulties facing optical designers. 6lthough some of these seem obscure, all are often present to some extent in most digital optical systems.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, '
Fadd in %uture dra%t: o0tical test 0attern imageG
COLOR FILTERING ARTIFACTS

6fter the color image has passed through the optical elements, it is usually separated into discrete color channels. The most common method is to use red, green, and blue filters simply either on top of the sensor elements *spatial color filtering/ or in multiple passes when ac9uiring the image, by using a different color filter for each pass *temporal color filtering/. +patial color filtering has lower resolution and encounters registration problems unless the red, green, and blue information are recorded at high enough resolutions. 2or example, to capture a !)) pixel4per4inch rgb image using spatial color filtering, the sensor needs to have at least ()) sensors per inch for the individual red, green, and blue elements to fit. 6lthough temporal color filtering overcomes the resolution problem, it often introduces different color registration problems when the sub8ect matter is not static. 6nother serious problem with color filtering is the lack of correlation to the human visual system. &ed, green, and blue filters used in image capture devices today do not =see> original scenes the same way humans do, and therefore do not record the same information. This has historically been a problem in photography *in particular with infrared4 emitting flowers/. ;nless the color filters are designed to be well correlated with the human visual system, there will always be some degree of color mismatch during the image ac9uisition phase. This can be most disturbing when trying to reproduce corporate colors in logos and products.
Fadd in %uture dra%t: illustrationG
SPATIAL PARTITIONING ARTIFACTS

6s briefly discussed earlier, sensors are positioned to partition the analog image spatially. This partitioning is the primary determinant of spatial addressability and spatial resolution. 2or example, if CC1 cells are placed too close to each other, the incident photon count =bleeds> from one cell to another, reducing spatial resolution without changing spatial addressability. 7f the final image print is to be at 1%# lines per inch on a printing press, it is recommended that the image have a spatial resolution of twice that. To produce an '.# x 114inch image at 1%# lines per inch by using a digital camera, re9uires having an rgb CC1 array of ',( # x 11,##) individual sensor elementE9uite an expensive proposition at current market prices.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, (
;nfortunately, unless the e9uivalent of such an array is used, there will be some loss of resolution during the ac9uisition process that cannot be recovered easily. CC1 pitch versus point4spread function of opticsD the electrons in CC1 do not automatically spill over to neighbor cells. ,nly at 1)S" overexposure does blooming happen today.
./0E'
DIGITAL 6UANITI,ATION ARTIFACTS

2inally, after this image has made its way through the optics and color filter distortions into the sensor cells, it must be converted into a digital signal. 6lthough '4bit 9uantization is still most prevalent, 1)4, 1 4, and 1$4bit 9uantizations are becoming more common. +imple linear 9uantization is still most common, but several nonlinear 9uantization methods have shown advantages in maintaining shadow and highlight details. This is because the bits captured are nonlinearly related to perceived lightness in the human visual system. <reliminary research has shown that '4bit 9uantization produces perceptible artifacts and at least 1) bits are desired. 3ith precision problems inherent in the imaging processing algorithms, up to 1$ bits might be re9uired to maintain imperceptible 9uantization artifacts. Ruantization artifacts show up most often in smooth gradations such as skin tones, computer graphics, and highlight areas. 7t is important to distinguish between bits for computational precision and bits of captured image data. ,ne is not likely to get above 1" bits of noise4free data from a digital camera or scanner.
./0E'
Color Space Encodings The color space chosen has a significant effect on the 9uantization error. This is due to the re9uirement of balancing the encoding to fill as much of the color space encoding as possible, and maintaining a perceptibly uniform space to minimize errors during color processing. 2or example, C7?I.J is a visually nonlinear and non4uniform space in which visible colors, when encoded using integers, occupies less than half of the encoding space *for example, its volume is less than #)T of the total gamut volume/. This is the worst of both worlds and re9uires special colored filters to match the xyz visual response functions. 6 common solution is a standard &:@ space such as CC7&4%)(Ewhich is non4uniform, but occupies a significant portion of the entire visual space. This provides an easy method to transform from the sensor color space into a device4 independent color space. Bowever, the colored filters for such standard spaces are not a linear transformation of these spaces. Therefore, some metamerism or color
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
, 9
mismatch will occur between how the device and the human visual system sees the original scene. Image Editing
7maging editing is the workflow stage in which all the separate image elements are integrated into a single document to compensate for past, present, and future image49uality artifacts. +imple image documents contain only a single pictorial image. Complex documents can contain a mixture of text, graphics, and pictorial images, as in advertisements and presentations. <ast artifacts are those inherited from the designFlayout and ac9uisition processes, described earlier in this paper. <resent artifacts are those intrinsic in editing on a color monitor, which is presently the most common today. This includes resolution, color space encoding, color depth, color gamut, and specific software application limitations. 2uture artifacts include compensating for the distribution and printing artifacts such as ink colorants, screens, ink bleed, and printer color gamut. +ome current software has the user manipulate the screen image to an ob8ectionable displayed appearance to print acceptably.
EDITING PAST ARTIFACTS

1esign and layout characteristics such as composition and cropping are often realized at this stage. 2or example, an image might be ac9uired that must be cropped to match the layout and design criteria. 2or convenience, this cropping is performed at the editing rather than the ac9uisition stage. This allows precise sub8ective decisions on composition with respect to the entire document. ?diting modifications such as unsharp masking and noise reduction are fre9uently performed to compensate for ac9uisition artifacts. +imple filters to correct spatial resolution and global noise *and even some spatial addressability/ artifacts are common to image editing applications. &ecently, filters to correct local noise such as dirt and scratches have become widely available. +ophisticated applications such as 6dobe <hotoshop and its third4party filters provide an extensive array of tools to minimize workflow artifacts on color image 9uality.
EDITING CURRENT ARTIFACTS

To edit a digital image on a color monitor, the image must be converted into the monitor5s color space at a displayable resolution. This means balancing compromises between the spatial and color characteristics and the user interface. Typically, "4bit rgb displays are used with a variable spatial resolution interface. This allows users to zoom in and work on particular details of
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# 7
an image or view the entire image *albeit at a reduced spatial resolution/ and make global changes. 6s briefly discussed with computer system limitations, using a color space of limited color addressability and resolution has a direct impact on image 9uality. ,ne of the most common difficulties in image editing is managing the color conversion to and from the monitor space. 6ny time an image is converted into a real device space, gamut4mapping support must be provided to map source image colors that are simply out of the device5s color gamut to colors within the device5s color gamut. This can cause serious degradations in image 9uality. To circumvent this problem, most image editing applications allow the user to edit the image in a printer device space *such as cmyk/. 6lthough an rgb image is displayed, it is converted on the fly to and from cmyk so that all editing changes are applied directly to the cmyk image. This limits the image and editing process to a particular printer, set of inks and often even a particular media.
EDITING FUTURE ARTIFACTS

The most common future artifacts compensated for by editing includeD :amut mapping +patial addressability differences Colorant characteristics ColorFmedia interactions
6s discussed earlier, gamut mapping is a ma8or area of current, unresolved research. 6 common artifact of poor gamut mapping, when attempting to resolve the dramatic differences between C&Ts and printed inks, is mapping bright yellows to white. ,ften, the spatial addressability of the edited image is very different from the distributed spatial resolution. 7f the resolution must be increased, then image data must be created. 7f done poorly, this can cause artifacts a loss of sharpness or pixelation *perceptibly large blocks of single picture elements/. @ecause there is a radical difference between the C&T and printed ink technology, some translation between the two must be realized. This includes the translation from specified &:@ phosphors to specified cmy*k/ inks. 7n addition to the change in color and possibly number of colorants *color addressability/, the complex, highly non4linear inkFmedia interactions must be compensated for. @y specifying various characteristicsEsuch as dot gains, under color removal, black generation, and ink primariesEa distribution device profile can be created. The device characterization profiles, specified during editing, go a long way to overcoming a variety of distribution and printing artifacts.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
Distri'ution%Printing 1istribution or printing an image involves several ma8or steps that can be viewed as a reflection of the ac9uisition stepsD 1. Color space conversion . 1igital 9uantization !. +patial partitioning ". Color partitioning #. 7nkFmedia interactions @ecause some of these processing steps are performed together, their order may be rearranged for implementation reasons. ?ach specific device and printing technology also has its own artifacts.
COLOR SPACE CONVERSION

Color space conversion resolves the color addressability mismatch between the source image and a pseudo distribution color space. The first step is to convert from the source image color space, usually a scanner or monitor rgb space or a device4independent color space into a pseudo printer color space. 6n example of this is converting from an rgb space into a cmy space. This is done with either simple algorithms *for example, c N 1 4 r/ or with a three4dimensional lookup table and interpolation techni9ues. 7n addition, gamut mapping is incorporated during this step. 1evice linearization takes the above cmy space and linearizes it with respect to the printer. Cinearization is performed separately from color space conversion to minimize interpolation artifacts. @lack generation is the process of converting from the above device linear cmy space into a color space with the same number of color channels as the number of device colorants. @lack generation, under color removal, gray component replacement, and under color addition all provide different methods to accomplish this.
DIGITAL 6UANTI,ATION
1igital 9uantization resolves the color resolution differences between the source and pseudo destination color spaces.
SPATIAL PARTITIONING
+patial partitioning resolves the differences in spatial addressability between the source and destination images. 6 common way to look at this is to convert a source image5s dots per inch into a destination image5s lines per inch.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# )
COLOR PARTITIONING
Color partitioning transforms the image from the pseudo distribution device space described earlier into the actual distribution device color space. This is most commonly done through error diffusion, dithering, or half4toningD ?rror diffusion takes a color resolution or a pixel in C<7, sets the device size pixel to the closest device color, and =diffuses> the color error to ad8acent pixels. 1ithering transforms a higher color resolution *and lower spatial addressability/ image into the distribution device color resolution and spatial addressability. This is commonly done by threshold matrices *or spot functions/ and screening angles. Balf4toning is a specific case of dithering where the spatial addressabilities are e9ual. 6lthough dithering or diffusion has been traditionally applied to bi4level processes, it is feasible to create a hybrid discrete, but not bi4level, combination of the two methods. +uch methods potentially allow for near4photographic49uality images for devices with significantly lower spatial addressability and color resolution.
IN8
AND
MEDIA INTERACTIONS
The actual process of applying colorants to the media is extremely technology dependent. @ut whichever technology is used, it is usually assumed to be precise and repeatableEgenerally speaking, a known, false simplification. &ecently, Kiggiano has shown a working spectral -eugebauer printer model that accurately models many printers to within C7?C6@ delta ? of #.).
PRINT DEVICE1SPECIFIC ARTIFACTS

<rinting is often divided into two basic technologiesD impact and non4 impact printing.
Impact Printing
7mpact printing applies colorants to a printing plate and then physically transfers the colorants from the plate to the media. -on4impact printing does not physically transfer the colorants from plate to media. Common artifacts to both technologies, but only for half4toned printers are moire patterns and rosettes, both artifacts from screening angles and halftone cell patterns. ,ther common dithering artifacts include contouring and speckled highlights. 7mpact printing, with the possible exception of silk screening, is a high4 volume process because of the complex custom setups for each image. This category includes gravure, offset, and letterpress printing press systems.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# 3
Common problems with impact printing are due mostly to their traditional analog heritage and such physical limitations as ink viscosity that cause bleeding, ink spectral properties that limit color gamuts, and press4speed variations that can cause dot smearing. 3ith an experienced press operator, the artifacts from these processes are minimized.
Fadd in %uture dra%t: illustrations %or gra"ure< o%%set and letter0ressG
"on(Impact Printing
-on4impact printing is the dominant form of digital printing today. 7nk 8et printers suffer from non4uniform dots that make color modeling complex, poor spatial resolution due to ink bleeding, and artifacts such as banding from mechanical artifacts. Color electrophotography has a number of specific artifacts, includingD plane4to4plane registration of multipass printers, prior4color plane4 dependent color density on single pass printers, a lack of uniform dot shape and placement, lack of consistency on a single page or page to page, and a serious dependence on humidity and temperature conditions. 1ye sublimation printers produce some of the best 9uality mass4market digital printers, but the transfer heater is often non4uniform, causing spatial uniformity problems. Thermal wax transfer printers have difficulty transferring very small dots. <hotographic digital printers, such as film recorders and laser printers, produce very high 9uality prints, yet both are dependent on the film emulsions and stable chemical processing, which is almost impossible to ensure. C&T film recorders often have cosine fall4off luminance characteristics due to poor optics, even on very expensive film recorders, and a lack of sophisticated electronics to compensate for the spatial non4uniformities of the C&T.
Fadd in %uture dra%t: illustrations o% technologiesG
+trategies to overcome inherent artifacts includeD 2ix the problem ?nhance other aspects to hide problem 6d8ust image after the fact 6d8ust other devices to compensate
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# ,
)or#*low Designs The workflow designs provide practical limitations that both the functional software and color reproduction results must abide by.
Single De ice Systems
This simply color reproduction system is comprised of a single device, such as a monitor, that does not interact with other devices directly. -eighborhood paint stores5 color matching systems and monitor centric computer graphics systems are representative of one such system. .et even today, the paint stores are attempting to cross4match the appearance of the paint to the monitor, and advanced computer graphics systems are concerned with integrating output with motion picture film and print output devices while maintaining consistent appearance.
Closed Se&uential Systems
These systems are exemplified by having a single scanner, display, and printer with possibly a press. ?ach device is tuned to a single reference, usually the press. The primary examples of these systems are custom designed military solutions and C?<+ solutions.
Closed Systems
These systems allow for multiple scanners and presses to interact, but each device is carefully tuned to a proprietary standard and cannot easily import or export color information out of the system.
+pen Application%De ice Systems
,pen applicationFdevice systems are represented by applications such as Microsoft <aint, or other systems that print into the :17 or graphics library and therefore into the user chosen printer driver.
+pen Application %Cross(De ice Systems
,pen applicationFcross4device systems are applications such as <hoto+hop or Corel1raw that allow users to specify device profiles.
+pen Systems with ,e*erence Media as a Common Denominator
The 7CC provides a clear example of an open system with a reference media used as a common denominator. Their reference is an ideal reflection print, even when no print media is in the workflow.
+pen Systems with Intelligent Parameter Protocols
,ne potential, but relatively unexplored extension of the 7CC work is to provide intelligent CMMs and an interchange space that is based on the human visual system instead of a virtual device.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# #
Color -uality
6s noted earlier, recent progress in standardizing digital color reproduction data structures includes 7?C standards or drafts for s&:@, sc&:@, and s.CC color spaces as well as color characterization standards for C&Ts, CC1s, plasma displays, digital pro8ectors, &:@ printers, scanners, and digital cameras. Microsoft has collaborated with leading vendors in each device market to develop color fidelity specifications related to these 7?C standards.
IMPORTANCE
OF
TESTING METHODS
,ne of the ongoing customer complaints against color management solutions over the last ten years is that they often simply don5t produce good visual matches in real world situations. :iven the number of recent books1U1#, articles1$,1%, web sites1'U !, tutorials, consortia "U!) and conferences!1U!# that address the communication and management of color information, this might seem somewhat surprising. The simple answer is that cross4technology and cross4viewing conditions color reproduction in an open network is a very complex problem. 6t its most basic level, this problem is an attempt to understand, model and simulate a large portion of the human visual cortex. 1espite this complexity, some isolated installations and carefully controlled demonstrations have indeed achieved good success, but seldom, if ever, do these reflect real world situations for mass markets in open networks. These ongoing complaints are due to a variety of real world problems and situations. +ome of these complaints are due to different software applications that display and print unacceptably different colors when representing <antone colors. 7t is still difficult to get scanners, displays, computers and printers from the same manufacturer *much less different manufacturers/ to produce acceptable color matches without additional and sophisticated intervention. 7t is also fre9uently difficult to get software applications from the same vendor to produce acceptable color matches while interacting with other applications from the same vendor. 2inally, it remains unclear where and how color management is performed and what device or software directs this performance in such a way to insure that multiple redundant or conflicting color transformations are not executed. This is by no means a complete list of the problems involved currently in color management =solutions,> but it does provide a representative
3i.e most con%erences< this ISJ1 Con%erence has annually s0onsored a num2er o% tutorials. Additionally< in de0th tutorials are a"aila2le %rom a "ariety o% indi"idual consultants.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# 6
sampling of some of the most problematic ob8ections. ;ntil these ob8ections are overcome, it will continue to be difficult to achieve adoption of color management solutions in mass markets. The creation of a viable color4imaging product is a balance of many, often conflicting product re9uirements including cost, performance, ease of use, size, flexibility, reliability, and color image 9uality. 3hile color image 9uality is one of the most complex re9uirements, it remains only one of many user re9uirements. 7t often consists of many underlying aspects such as resolution, color depth, spatial uniformity, color accuracy, color precision, and dynamic range to name a few. :iven this complexity, it is easier to focus on other product features as both a vendor and a reviewer. 1espite this, some reviewers have tried valiantly by showing a series of different print sample representations in various magazines. These reviews are limited by some of the very reproduction issues that are being reviewed and only show at most a limited number of the aspects that make up color image 9uality. 7t is the authors5 belief that unless and until a set of simple color fidelity test methods can be ubi9uitously established, little progress will be made in overcoming the very real and difficult problems that currently limit color management solutions. ,ne simple example of the potential impact of problems caused by these limitations is illustrated when one examines electronic commerce. Traditionally a product vendor creates a printed catalog of their inventory and distributes this catalog to a large number of potential customers. The printing and distribution process are very expensive and the actual customers represent a relatively small portion of the entire distribution list. 2inally, the number one customer complaint of catalog shopping is poor color matching between the catalog and the final product. 3eb4based catalog shopping is much more cost4effective for the product vendor in that it eliminates both the printing and the distribution of the catalog. .et to be successful in replacing traditional methods, it appears that it must be both more convenient and reliable than traditional methods. This means that the color complaints must be dramatically reduced. ;nfortunately, whereas a traditional catalog is created in a relatively closed color reproduction workflow, web4based catalogs exist in a very open and often ill4defined computer network. <roperly managing the unambiguous communication of color information in an open network, across application, utility, device and operating system vendors seems to be a fundamental re9uirement in resolving this problem. 6lthough there are clearly other factors involved in establishing successful electronic commerce solutions, a weak or limited color management solution will pose a significant hindrance to the process. This is 8ust one example of many that could have been posed.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# '
,ne of the roadblocks to overcoming these problems is that there is not a clear or simple method to show 9uantitatively that a color reproduction system isn5t working acceptably. 6s with any technical problem, it is much easier to fix it or improve it if one has some measure of whether and how much it is broken and in what way it is broken. The entire industry of statistical 9uality control rests upon this basic premise and has been well documented for decades. Currently there does not exist any widely accepted method or set of methods to ob8ectively and 9uantitatively evaluate color management solutions.
CONSIDERATIONS
IN
ESTA-LISHING TEST METHODS
To begin to establish test methods to address the problem described earlier, it is necessary to understand, consider and prioritize the many factors that affect color fidelity. 7t is common for a particular vendor or product reviewer to try and consider each device or application in isolation. 6lthough this approach might seem reasonable, it often masks the more difficult problems of integrating the device or application into a real world workflow. 6nother common approach that can provide misleading information is summarizing a device5s color fidelity with an average C7?C6@ delta ?P of some particular pictorial image. 6lthough this approach does provide some relevant information, it rarely provides a robust view of the color fidelity issues of an application or device. 7n addition, it compounds known weaknesses in C7?C6@, such as a perceptual blue4purple non4 linearity, directly into the testing method. This could cause devices that actually provide perceptually superior color fidelity to be at a disadvantage to devices that are optimized to C7?C6@. 6lthough no one single number can ever ade9uately provide a robust evaluation of color fidelity, some reporting processes must be extremely concise when evaluating many different aspects of a product. 7n circumstances where a single summary number is provided, it is recommended that a clear explanation of how this number was derived be available. ,ne goal of color fidelity test methods is to reflect and test the multiple software and hardware paths that underlie the most common use cases end users implement. Color reproduction systems today are combinations of hardware, software, utilities and operating systems. ?ach of these components provides a very wide variety of options that directly affect color reproduction. 6 small subset of these options includes different paper, resolution, interpolation, color space, device drivers, and profiles. 6lthough the actual number of options varies with each vendor product, it is clear that reporting, much less testing every combination of all of these options is simply impossible.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# (
7t is impractical to test all possible combinations of devices and software. Therefore it is recommended that the manufacturer5s recommended default setup be used in most cases. 6n exception would be in the case of evaluating a particular workflow such as CM.L +3,< prepress workflow. This also applies when choosing the non4 targeted aspects of the reproduction workflow. 2or example, when evaluating a printer, choose the most common components when constructing the rest of the system. @y choosing the most common use cases, applications or devices as reference standards, it becomes practical to establish a stable environment in which to test individual components. ,nce a test environment is established, try and keep it as stable as possible in order to provide some level of ob8ective comparison over time. +uch an environment allows one to test individual applications, CMMs, profiles, devices, and calibration utilities in an ob8ective manner. The one exception to this approach of establishing a stable default environment is when one is testing robustness against a wide variety of systems. 2or example, when developing new printer products, it is necessary to test against a wide range of software applications. @y maintaining as stable of an environment as possible, it becomes easier to isolate compatibility problems in the development process. 6n additional aspect of color fidelity test methods is to appropriately target these methods to their users. This paper describes four methods that trade off ease of use with informational accuracy and complexity. 7t is both futile and unreasonable to arbitrarily mandate that end users implement a full4blown psychophysical testing process. 7t is 8ust as problematic to assume that a simple visual assessment is ade9uate for device manufacturers in the product development process. 6lthough a simple visual assessment will not provide the robust information that a complex colorimetric assessment will, it may provide all information re9uired by a user. The series of methods, described later in this section, increase in complexity and resource re9uirements. This increased complexity and resource re9uirements is directly associated with an increased accuracy and depth of information. 2or example, the first method is a simple visual assessment that provides only passFfail information against a known tolerance. The complex colorimetric assessment provides a summary number, and also a breakdown of 9uantitative color differences by hue, chroma, and lightness in varies color 9uadrants. Therefore, it is critical that the test methods appropriately target the needs and resources of their users. Method .: Simple /isual Assessment The first method is a simple test document that exercises a number of software paths. +uch a document is described along with simple guidelines for acceptability. These guidelines include straightforward instructions such as =make sure red is not orange, blue is not purple, and
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
# 9
there are no cyan dots in yellow.> This method is ideal for 9uick visual tests that are a small aspect of larger test suites. +uch suites are common in application and device development schedules. The document should be a combination of text, colored text, charts, clip art, and raster images to ensure robust testing of processing paths through the system. 7t is common for applications, utilities, operating systems and devices to treat different ob8ect types differently. 7f care is not taken during the development process, it is easy for these different paths to produce unintentional differences. 2or example, a simple Microsoft 3ord or 6dobe <ageMaker document could be created to meet these re9uirements. 3hatever application or document format is chosen, it is critical that the color descriptions of the ob8ects in the document be well defined. There have been attempts to provide reasonably stable standard images and documents to base these assessments upon. 7+, TC1!) 3: and the Committee for +tandard Bigh <recision <ictures *+B7<</ !$ have both developed a set of standard pictorial images and measurement patches to enable color fidelity assessment of pictorial images. The 7+, +tandard Color 7mage 1ata *+C71/ targets the graphics arts market and provides CM.L data. The +B7<< images are based on 7T;4& @T.%)(4 and s&:@ data and contain device4independent color data. 2inally, more complex documents can be created to test a wide range of issues including color fidelity, performance, resolution and cost4per4page. +pencer and 6ssociates provides a good example of this type of work !%. @oth +B7<< and +C71 data sets attempt to unambiguously describe color information and how to use it. This unambiguous explanation is critical when attempting to integrate standard color ob8ects such as images into a test method. Three basic aspects of color are re9uired to unambiguously describe color in a manner useful for testD the reference device, viewing condition, and observer. 3ith each of these descriptions, it is necessary to also provide the underlying models. 1evice4independent color spaces are based on the human visual system, so the device and observer are identical. This does not eliminate the need to provide a clear description of the viewing conditions and the underlying appearance model to transform into and out of these viewing conditions. 2or example, when using C7?C6@ or C7?I.J, it is critical to provide descriptions of both what the viewing conditions are *surround, white point, luminance level, and so on/ and how these conditions are modeled. Many standards provide only the parameter levels and not the underlying models. This approach is misleading at best. 7t is comparable to speed limit signs that say =##> but don5t indicate if it is in miles4per4 hour or kilometers4per4hour. <roviding an ambient white point without describing the chromatic adaptation or color appearance model is a good example of this problem in the color field. 7t assumes that any chromatic adaptation or color appearance model is viable. This is simply untrue,
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 7
and many studies have proven that such models are significantly perceptibly different from each other , especially in saturated graphics such as corporate logos. 6nother common example of this problem is when a =gamma> is specified such as . . 3ithout providing an underlying model that might or might not include an offset, gain and flare component, it is unclear what the actual implementation of this parameter is. 2or example, 7T;4& @T.%)(4 has an exponential value of ." and this parameter is often cited as the =gamma> for B1TK, ignoring the significant offset component. 3hen the offset is factored in, a simple power function *with zero offset and a gain of one/ fits the exact same recommendation with a =gamma> of . . 3hich is correctM @oth are, within their own context. ,ne ob8ection raised to clearly specifying a color appearance model is that it is either proprietary or still under active research. This is a superfluous and misleading argument since the parameters provided are still based on some real4world implementation. 7n conclusion, a simple visual assessment document provides very high ease of use with a rough assessment of acceptability. Method 0: Complex /isual Assessment The second method is a more complex, but is still a reasonable visual test that uses a page of solid reference colors *some in gamut, some out of gamut/ and the code to produce this on a printer. 7t can also contain the information in the simple visual assessment document. This test provides direct, visual comparison with known tolerances also included in the test page. 7t is also easily extended to provide tests for scanners and display. +uch a test is ideally suited for product reviewers or test engineers or reviewers who do not have easy access to measurement e9uipment or have significant time constraints. The actual assessment consists of a series of reference colors and tolerances that make up a visual template. 6s with the =variations filter> in 6dobe <hotoshop, the target color is surrounded by patches with varying hues and chromaticities at a specified tolerance. 7n addition, patches of greater and lesser lightness are also provided. @y using a template with the target patch position cut out, the user can simply overlay the template onto the target color print and visually assess whether the target patch color is within the visual tolerances provided. @ecause simple patches are used, it is critical to ade9uately sample the device5s color gamut and to separate in4gamut colors from out4of4gamut colors. There is an implicit weighting of gamut information by the choice of target colors.
)
1he numerous articles in 0re"ious ISJ16SI$ Color Imaging Con%erences on color a00earance models and white 0oint ada0tation models clearly illustrate these di%%erences.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 similar method can be applied to scanners with the use of a digital template and a calibrated monitor. 6n alternative scanner method is to provide a digital template and an application utility that compares the scanned results against the target results. The advantage of this method over the simple visual assessment method is that an unambiguous acceptability tolerance is inherent in the evaluation and additional 9ualitative information on which part *or parts/ of the color space are acceptable, which are not, and in what direction. Method 1: Complex Colorimetric Assessment 6 third method uses a test target similar to that in the second method. 7t is of course possible to provide additional patch targets or use standard patch targets such as the 7+, TC1!) 3: targets or uniformly sampling the device color space or a device independent color space. Color differences are computed between the original target and the output. These are summarized using <ointer and Bunt5s color reproduction index methodology!',!(,"). This produces a summary average delta ?P, and also provides more detailed informationEsuch as differences in lightness, chroma, and hue, and the direction in which the printed output differs from the reference target in each of these dimensions. @y including both in4gamut and out of gamut colors, the gamut mapping issues can be summarized separately. @oth this test method and the previous can be modified to test different processing paths. 6dditional information can be derived and reported using this method. ,ne example is metrics concerning gamut volume and shape. The many issues involved in comparing color gamuts of different devicesEespecially among capture, display, and print technologiesE include complex issues such as having either similar viewing conditions or a standard method for compensating for viewing conditions. They also include issues that appear simple, but are in reality 9uite subtle and complex, such as what is white and what is black in the gamut comparison. 2inally, it is necessary to not only report a single number, such as gamut volume in a uniform color space and a typical viewing condition, but metrics that describe the shape of the gamut in a reasonable manner. This might include gamut circumference as various levels of gray or volume mismatch to standards gamuts such as +3,< or s&:@. :amut metrics are only one example of many other descriptive metrics that can be derived from this method. +uch a test is ideally suited for product color engineers who want to isolate and improve any color defects or artifacts in a product. 6lthough this method re9uires the use of a color measurement device, it provides much more detailed 9uantitative results about the color fidelity of the product being evaluated. 7n
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 )
addition, there are many automated measurement stages that simplify the time consuming task of measuring individual patches. Method 2: Psychophysical Assessment The final method of assessment is a full4blown psychophysical evaluation. 6 series of test documents are reproduced. 6 standard viewing condition environment is created that is representative of the target user market. 6 pool of color normal observers is then recruited. ;sing standard psychophysical methods, a variety of evaluation 9uestions can be submitted to the observers. +tatistical analysis of the results provides superior information that is independent of any limitations of color appearance, color space or color difference models. This method eliminates any dependence upon potential flaws in color appearance models in the previous method. 7t also provides detailed information on what aspects of the product are strong and weak with respect to color fidelity. 7t is not uncommon that the most valuable information is found in the general comments by the observers that are not part of core evaluation 9uestions. This method also provides the capability to determine the priority of color fidelity issues in comparison with other product re9uirements. ;nfortunately setting up, administering, and analyzing psychophysical experiments is a very intensive and time4consuming process has often has a significant amount of statistical variation. 3hile ideally this reflects the true end user, it can be difficult to extract clean conclusions. This brief summary of psychophysical testing cannot do 8ustice to this complex field. There are extensive references available for readers who are interested in this method "1,1),1 ," .
COMMON IMAGE CHARACTERISTICS

,b8ective characteristics can be determined by some physical means whereas sub8ective characteristics must rely on psychophysical means to derive their metrics. 7n summary, most images possess the following characteristics.
S)%1ective characteristics' Su2@ect matter 3ayout Com0osition 3ighting setu0 :ray 2alance 1onal control Color%ulness Hue 2alance Cogniti"ely color constant /%1ective characteristics' S0atial addressa2ility S0atial resolution Color addressa2ility Color resolution 9oise
1evice4independent color could be restated as an ob8ective characteristic of well4defined and unambiguous color definition. This means we know what =red> means *not orange and not magenta/, and so we know how
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 3
we should display the image for evaluation. +imilarly, the viewing conditions provide the target evaluation viewing condition in which to 8udge the images. ,therwise, we might be 8udging the images in untargeted conditions. Su'3ecti e Characteristics +ub8ectively, we would like our images to appear artistically excellent and appropriate for our message *for example, Ce 3ebCouvre/. 7n commercial applications, this is where the graphics artist or photographer design and lay out the image elements, including sub8ect matter, layout, composition, lighting setup, gray balance, tonal control, colorfulness, and hue balance. ?ach of these elements has a psychological impact on how the customer perceives the image and to some extent what =mood> is created by the image. 7t is important to match the sub8ective characteristics of the image to the intended audience. 2or most research presentations, this is a combination of the actual engineering and management audience, plus the final customer of the product or technology.
SU-9ECT MATTER
+ub8ect matter refers to the type of ob8ects in the image. ,nly the available ob8ects and the author5s creativity limit the variety of sub8ect matter. 6 common rule with respect to sub8ect matter is coherency. 6n example of this is to place children with toys and not with complex machinery, unless a particular message is desired. Choose the sub8ect matter carefully. 3ith respect to designing images for research presentations and reports, this might mean talking with the marketing people to find typical scenarios and situations that the final product or technology is planned for. 7 recently saw a nice example of this where the image design simulated a family newsletter to grandparents, thanking them for a gift. The images were consistent with this design and well illustrated the current status of the print 9uality in a real4world simulation. ?xamples such as this have much more impact than using only synthetic test charts or classical computer science images such as Cenna. They are also good reminders of who the customer is. Bowever, this does not mean choosing images that hide artifacts. 7 recommend two types of sub8ect matterD one that might be used by final customers, and a second set made of carefully chosen images that might rarely be used by final customers, but that illustrate the research or message more clearly than a typical image. :ood composition of images is a fundamental re9uirement. 7f images include faces that are cut off or are poorly cropped, the audience will be distracted by this and often feel the overall presentation or research is of poor 9uality.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 ,
add in %uture dra%t: a good and a 2ad eAam0le 0ictureG
LAYOUT
Cayout refers how the ob8ects are positioned with respect to each other. -aturalism is a common rule of layout. @y arranging the ob8ects in natural spatial relationships relative to each other, the image presents a natural, coherent message. 6s with coherency, one can place ob8ects in unnatural positions to relate a specific message.
Fadd in %uture dra%t: a good and a 2ad eAam0le 0ictureG
COMPOSITION
Composition refers to how the ob8ects are positioned with respect to the boundaries of the image. @asic guidelines of good composition includeD @alance white space. 3hite space is the portion or portions of the image that does not contain significant sub8ect matter. @y balancing the placement and amount of white space, images appear uncluttered and focused. Center sub8ect matter. @y centering the primary ob8ect or ob8ects in the image, a natural focus for the image is created. Maintain sub8ect matter integrity. Maintaining sub8ect matter integrity can be as simple as not cutting the heads off human sub8ects by the image border. Match a particular format, such as width and height constraints, as re9uired. The most common general formats include portrait and landscape, which by their names indicate composition preferences for particular sub8ect matter. Maintain a single format within a presentation.
LIGHTING SETUP
The term lighting setup or lighting balance is used to define how the illuminants in the original scene interact with the sub8ect matter. This includes not only things such as flood, fill, and back lighting, but also the luminance levels and color temperature of the illuminants. Typically, the sub8ect matter should be well lit to optimize the tone levels and gray balance.
GRAY -ALANCE
:ray balance is closely related to chromatic adaptation and white point compensation techni9ues in color science. &alph ?vans theorized that
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 #
the human visual system integrates most natural scenes to gray and based several film emulsions and all modern day automatic print enlargers on this theory. ,ne can explicitly choose to bias gray balance *such as in sunset or sunrise images/, but for most situations, neutrals in the original scene should remain neutral in the final image.
Fadd in %uture dra%t: note on grey 2alance o% shadows onlyG Fadd in %uture dra%t: a good and a 2ad eAam0le 0ictureG
TONAL CONTROL
6nsel 6dams wrote several classical works on contrast and tone control. Bis =zone control theory> divides the dynamic range of scenes into a few e9ually spaced regions. The basic premise is to design images with these zones in mind to insure sub8ectively good dynamic ranges, shadow detail, highlights, and depth perception. Maintaining the original tone levels throughout the reproduction workflow poses an enormous challenge.
Fadd in %uture dra%t: a good and a 2ad eAam0le 0icture Co"erly color%ul as well as a 0ale imageDG
COLORFULNESS
Colorfulness has been defined by 1r. &obert Bunt asD V6nW attribute of a visual sensation according to which the perceived color of an area appears to exhibit more or less hue" 6 wide variety of other terms are commonly used and misused to describe this, such as mutedness, vividness, color brightness, saturation, bounce, and many others. @y purposely designing the amount of colorfulness desired in the image, one can manipulate the customer experience to a degree. This is why very bright primaries are used in many commercial packages in grocery stores, and is seriously considered by fashion and automotive designers when choosing colors for their products.
HUE -ALANCE
Bue balance describes the colors of the sub8ect matter in the scene, such as green grass and blue sky. 6n ocean scene may have an overall bluish hue balance and still maintain a neutral gray balance. @y biasing hues of the sub8ect matter, the =feel> of the image is controlled. @rownish dull images =feel> drab, colorful =warm> primaries *orange, yellow, red, and so on/ appear lively, and clean, natural blues and greens often =feel> more soothing than some other choices. 7t is important that the choice of hues in the sub8ect matter not distract from the primary message of the image.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 6
COGNITIVE COLOR CONSTANCY

,ne implicit assumption made by the designer is that the image color will be cognitively constant throughout the workflow processing. This means that the color of the image will =appear> the same no matter what device it is displayed or printed on, regardless of the environment where it is viewed. This is impossible with current technology, but for the ideal image, it is included as one of the desired sub8ective characteristics. 6 number of new color management systems, either in development or on the market, are attempting to handle this problem effectively. +'3ecti e Characteristics ,b8ectively, we would like the ideal image to have no physical constraints. ;nlike sub8ective characteristics, ob8ective characteristics are not usually specified in the design and layout, but are implicitly limited by the device limitations in the other workflow processes. This would include infinite spatial addressability, infinite spatial resolution, infinite color addressability, infinite color resolution, and freedom from noise.
SPATIAL ADDRESSA-ILITY
+patial addressability is the ability to address or locate individual pixels. This is commonly characterized as dots4per4inch *dpi/ or lines4per4inch *C<7/. +patial addressability has a dramatic impact on both performance and image 9uality.
SPATIAL RESOLUTION
+patial resolution is inversely related to the size of each pixel, so that the ideal image has an infinite amount of infinitely small pixels. This characteristic describes the graininess or resolving power of the image. 7deally, the spatial addressability and spatial resolution should match. 7f the resolution is greater than the addressability, the image is blurred. 7f the resolution is less than the addressability, the image is grainy.
COLOR ADDRESSA-ILITY
Color addressability is the number of color channels that can be addressed. 2or example, rgb images have three color channels and therefore possess a color addressability of three. +pectral images typically have color channels for every ), 1), #, or 1 nanometers in a range *again, typically/ from ")) to %)) nanometers *inclusive/. This would provide color addressabilities of 1$, !1, $1, and !)1.
Fadd in %uture dra%t: discussion on how and why three color systems wor. in generalG
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 '
COLOR RESOLUTION
Color resolution describes the precision of each of these channels. 2or example, common rgb images have a color resolution of ' bits *or #$ levels/ per color channel. The ideal image would have an infinite number of color channels with infinite levels per channel. Color resolutions of 1), 1 , and 1$ bits are becoming more common. The combination of the color space in which the data is encoded and the color resolution have a dramatic impact on 9uality. 3hereas "4bit images can often be ade9uately encoded in the C7?C6@ color space, a similar image is rarely satisfactory if encoded in the C7?I.J color space.
NOISE
2inally, the ideal image would possess no noise. -oise is fre9uently divided into global noise *such as white or blue noise/ and local noise *such as dirt and scratches/. 3hite noise is uniformly distributed random noise, blue noise is more heavily distributed in high fre9uencies, and pink noise is more heavily distributed in low fre9uencies. These types are often due to systematic process problems. Cocal noise is often due to some physical defect or damage, such as mechanical misalignment.
COLOR 6UALITY METRICS

Color image 9uality and color metrics are fundamental in 8udging the effectiveness of digital color reproduction systems. There are many aspects of image 9uality, including print 9ualityFimage 9uality, algorithmic testing, mechanical testing, user preferences, and market alignment. 6 color 9uality metric is a numeric description of the perceptual or acceptable difference between two colors or images. This difference is usually described by a ?uclidean metric in a 9uasi4perceptible uniform color space such as C7?C6@. 7n one sense, color 9uality metrics are an attempt to summarize and simplify psychophysical models and results. Considering the complexity involved in the field of psychophysics, the importance and accuracy of color 9uality metrics must always be carefully considered. <sychophysics is the science of investigating and understanding the relationship between our senses and the physical stimulus that creates them.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 (
Color metrics such as delta ?P have been around for many years. ;nfortunately, most users of these metrics ignore their underlying assumptions such as a standard C illuminant source, the limited application of comparing on two single patches of color for matching, the medium gray surround, and the reflective media. There have been some attempts to extend this simple descriptor with weighting and spatial considerations '. The ma8or advances in color metrics have been made in the field of paint colorants, especially automotive paints. This field has performed extensive psychophysical experiments 9uantifying color differences. This work has resulted in modifying the C7? color difference metric (. ;nfortunately, few publications have picked up this or previous color metric work and applied it to modern work, and many publications still use the flawed 1(%$ C7? color difference e9uations when these are no longer recommended. 6nother common, seldom4investigated assumption in practice is the use of arithmetic averages of color errors. 7t has been pointed out previously that this is statistically flawed and that a geometric average should be used instead. The arithmetic average assumes a :aussian distribution of error values, although this is extremely uncommon when average errors are less than 1) delta ?P. 7n such cases, basic statistics recommend a geometric average as more accurate. 6lthough published, this is seldom used in modern research. To properly model user preferences, it is necessary to detect color artifacts. The general detection of image artifacts has few practical broad procedures. ,ne such procedure is presented here based on practical experience of the author over a period of 1# years teaching color technicians to 8udge and correct color images, and also by viewing and correcting over !)),))) images.
Fadd in %uture dra%t: summaryG
1espite these limitations, the concept of a topological metric based on a visually uniform color space seems sound. 3hether or not this is a ?uclidean metric is open for debate. +ome researchers have suggested &iemannian or city4block color difference metrics instead of ?uclidean metrics. 4ser Pre*erence Methods ;ser preference methods attempt to model particular customer preferences and biases generically with respect to image 9uality. These preferences are often cultural or application oriented, such as kids preferring brighter colors. +uch broad stereotyping must be used cautiously, as with all stereotyping. The goal of color reproduction can vary. 2or most practical situations, the goal is preferred color reproductions.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
6 9
There are several aspects of users preferencesD Content4based preferences, such as company logos 7mage type preferences, such as presentation graphics <erceived accuracy preferences, such as skin tones, error tolerances, and direction 7t has been shown that image content does not provide a significant influence on the perceived color differences in reproductions and only slight influence on the acceptability!). 6lthough some studies have argued contrary to this, most of these arguments lack a statistical basis. This is born out of the legacy of automatic color correction in the photofinishing industry, where simple color corrections have long had a history of success to a high degree of acceptability. 6lthough this is not a complete solution and in some cases content does impact acceptability choices, it is a secondary factor. ?ven in photofinishing and automatic exposure controls, content estimation5s primary result is to balance exposure control. ?ssentially, it is where to compromise in the overall lightness balance of the scene, especially in the example of scenes that are poorly balanced to begin with, such as back4lit portraits. 7t has been shown that image type does make a significant difference in user5s acceptability criteria. 7n particular, simple targets such as the Mac@eth Color Checker provide neither reasonable nor accurate representations on how users 8udge color reproduction acceptability. This is a very unfortunate circumstance and one that has great impact on common practice. Currently, many vendors base their color reproduction algorithms on how well they reproduce simple charts using the C7?C6@ delta ? metric. This has the double flaw of basing an algorithm on a target known not to be representative of the end4user criteria, and basing an algorithm on an error metric known to be flawed and acknowledged as obsolete. .et this still seems to be the common practice. 6 better solution would be to have a set of representative images consistent with the end4user market and 8udge this set of five to nine images on a regular basis. This is consistent with the +C71FC:6T+ recommendations and the digital images !). 6t best, the targets and error metrics should be used for primary estimates, followed up by reference images, and finally a larger set of images to test any boundary conditions. ,ften, small errors in algorithms are only found after several hundred images are reproductions. This process is consistent with the history of photographic film research and development. Combinations of balancing photographic areas of the gamut with highly chromatic areas provide another example of implications of user preferences. 2or presentation graphics, users prefer pure device colors and not colorimetric matches that incur high spatial artifacts.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' 7
<erceived accuracy preferences are relevant for so4called memory colors, such as green grass, blue sky, and skin tones ". 7t has been found that most prefer a significantly different reproduction than reality. 2or several years, film vendors positioned their professional films as more saturated or more realistic. ?rror tolerances and direction preferences provide the allowable variance around the perceived accurate or =ideal> reproduction color. This ellipsoid is often not centered on this aim point.
Cultural differences provide excellent examples of this case. 2or example, most users tolerate some variance in color reproduction, but the direction of variance can be critical. Two cultures that typify this difference in tolerance are -ative 6mericans and 7ndians from the subcontinent of southern 6sia, with respect to skin tone reproduction. -ative 6mericans will tolerate significant hue shifts in skin tone reproduction providing the error is more red than the original. 7ndians from the subcontinent of southern 6sia will tolerate significant hue shifts in skin tone reproduction providing the error is more green than the original. ?ither culture will generally ob8ect and maintain the opposite hue shift produces as unacceptable reproduction of skin tone. De ice Modeling 1evice models, including simply three4dimensional lookup tables, attempt to correlate some aspect of the device color space with the human visual system. ,ften this is removed from the fundamental device color space. Most device models translate between some well4behaved device color space, such as monitor &:@ and a human visual system4based color space such as C7?I.J. The two basic approaches to device modeling are analytical and empirical. 6nalytical device models attempt to provide ad8ustments and functionality that simulate the physical, optical, chemical, electrical, and mechanical properties of the device. The crux of this approach is to break the model into small, modular pieces, each of which represents some intuitive control over a single part of the physical device. ?mpirical models treat the device as a black box and attempt to model the device statistically. This is often done by creating and interpolating three4dimensional lookup tables or by fitting parametric e9uations such as polynomial e9uations. +tatistical fits are employed in both approaches to provide reasonable predictions in production devices. Bowever, too often the underlying statistical assumptions for normal distributions are not verified or valid, and this leads to false results and
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
'
conclusions based on flawed statistical assumptions. These errors are difficult to uncover and 9uantify after the fact. Analytical Models 6nalytical models are device models that provide ad8ustments and functionality to simulate the physical, optical, chemical, or mechanical properties of the device. 2or example, a monitor model consisting of a !x! matrix and a transfer function simulates the mixing of phosphor light off the faceplate and the electron gun conversion between voltage and luminance. 1etailed processing se9uences in the can provide overviews of typical analytical device functions. 6ccurate analytical models can lead to breakthroughs in improvements in 9uality, cost, and performance. More complex analytical models involve modeling ink on paper interactions and how light scatters in these situations. 6nalytical models are rarely used in actual production codeH instead, runtime optimizations are recommended, including the 8udicious use of three4dimensional tables. This maintains a reasonable balance between flexibility, intuitive controls, and performance issues. 6nother consideration in analytical models is when to use floating point, fixed integer, or simple integer computations. 6lthough floating4point computations have performance limitations, they provide easier implementation of optimal precision. 3hen used in combinations with runtime interpolation tables, these performance issues can be insignificant. 2ixed4point computations allow a compromise with precision and performance, at the cost of ease of development and testing. Cittle research has been published on the impact of computational precision with respect to image 9uality!1. Empirical Models ?mpirical models provide little or no insight into the underlying device functionality. They are simply a black4box algorithm to transform between a viewing4condition4dependent human visual space such as C7?I.J or C7?C6@ and a well4behaved pseudo4device space such as CM.. Common examples of such models are three4dimensional lookup tables, neural networks, fuzzy logic, and some higher4order polynomial regression models. 7n some cases, it is prudentEgiven resource and time constraintsEto create an empirical model, especially if improving or calibrating the device are not concerns. 6n additional case for some empirical models is performance. Typically, three4dimensional interpolation is more efficient than complex analytical models. 7n this case, the empirical model should be built at run4time
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' )
from the analytical model to provide maximum flexibility and calibration facilities. This method of building run4time empirical models from analytical models provides the most efficient use of profile management by optimizing the number of profiles necessary and maintaining the fastest performance. The disadvantage is that the color management module must be able to understand or execute the analytical model to create the empirical model. This re9uires either intelligent CMMs with a common base on analytical models or custom CMMs for each custom analytical model. ,ne possibility is to provide a framework with the capability to create empirical models from custom CMMs on demand any time a profile is exported. 6 key consideration when implementing three4dimensional lookup tables is percentage of occupied volume. This is the percentage of the three4dimensional volume occupied by the color gamut of the device. This volume depends not only on the color space chosen to interpolate in, but also the encoding dimensions of the color space. 2or example, interpolating in C7?I.J space results in a poor use of interpolation volume.
,ne method to reduce the artifacts that occur as a result of poor volume usage is to increase the number of vertices. The unfortunate side effect is to increase the size of the data by a power of three. 7n some cases this can also cause a dramatic impact of performance. 7t is also possible to have dimensions that are une9ual. This can be used to weight the lightness dimension more than the chroma dimensions if interpolating in a lightness4chroma space. -eural -ets and 2uzzy Cogic have become fashionable in recent years in color research. ?xcept as a heuristic short cut to implementing products, these solutions provide no fundamental new knowledge or control of how devices work, and tend to lead researchers away from areas in need of research toward implementation solutions. Flare Compensation Methods 2lare compensation methods model the effects of flare in the capture optics *lens/, off the faceplate *monitors/, or off the surface of the media. ?ven when flare is properly measured on monitors, it is not always advisable to transform in color imagery into this space because significant data loss will often occur *s&:@/. Currently, the best way to handle the viewing flare in monitor environments is still ambiguous. The problem is that the flare is typically so high *# to 1) percent/ that the black levels are unacceptable.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' 3
To transform data into this space would not be a problem, but to transform it out of this space would re9uire that there not be true blacks. 7deally, a perceptual rendering intent will resolve this issue. ;nfortunately, if this is not performed in floating point, there will be data loss in the shadows. 7n summary, even this case should follow that standard se9uence. 7nternal optical flare in lens systems is a well4documented phenomenon. 2lare off the media surface is more commonly called gloss, or texture effects, and much has been written on this. 6lthough there is much documentation on this issue with respect to the painting industry, little has been addressed with respect to a methodical treatment in digital color reproduction. 7n fact, the issue of gloss is usually confounded with other issues by simply stating that gloss increases perceived saturation and dynamic range. 2urthermore, most measurement standards in the photographic and graphics arts industries recommend )F"# degree measurement geometries, which actually hide the effects of gloss and therefore create misleading measurements. 6 sample comparison of measurements with )F"# and )Fdiffuse degree geometries on glossy media illustrates this point.
Fadd in %uture dra%t: gra0hG
Sensor Compensation Methods +ensor compensation methods attempt to limit the impact of the difference between the device sensors, human cones, and sensor systems. Metamerism is a primary concern when attempting to compensate for device sensor artifacts. The sensitivity of the sensors is rarely a linear transformation of the human visual system. The illuminants in most scanners do not match the original viewing illuminants. These two facts make metameric problems significant concerns in that the image appears one way under one set of lights to the user, and completely different to the scanning device. These limitations are due to practical cost and availability issues imposed by and on device vendors *1F6 linear or nonlinear/. 1igital camera sensor arrays are often in mosaic patterns, re9uiring processing to reconstruct a =full> three4color image. +ome cameras use a scanning tri4linear array, improving resolution at the cost of temporal artifacts while maintaining a low cost. CC1 arrays are based on filling an electron well from the photon input energy, and regularly draining and counting this well. Many problems
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' ,
are encountered in sensitivity, contamination with nearby wells, and trade4offs in performance and noise. +ome scanners use a single linear array with three passes under different lights or filters. There can be shifts in the original image artifacts, but the scanning times are much slower. ,ther scanners use a tri4linear array with one pass, which results in increased optical path complexity and potential alignment problems. "egati e Film Scanning Methods -egative scanning methods model negative films to varying degrees of sophistication and provide transforms between the negative film and a positive representation. 2irst, unbuild the negative film, convert to a positive format, and then possibly rebuild the positive film look. 6 simple matrix ignores the entire photographic development history and assumes incorrectly that film cross4talk is linear. 6 simple one4 dimensional lookup table further confounds these issues. Automatic )hite Point Estimation 6utomatic white point estimation provides estimates of the scene white point for capture devices, crudely modeling the human visual system5s ability to adapt to white in most circumstances. To transform from a real4world scene into a device space, an estimate of the white point must be derived to use in even the most simplistic color appearance models. Bistogram stretching is the simplest and most error4prone method, assuming no knowledge of reproduction systems or physics. ?vans proposed the gray world assumption over four decades ago ! and it is still a fundamental method of white4point estimation. Cand and McCann proposed that white4point estimation be based spatially on independent cone channels. Automatic Exposure Control 6utomatic exposure control methods provide dynamic range compression or expansion for capture devices to match the input source with the optimal range of the sensors. This models the human visual system5s ability to adapt to a wide range of scene luminances. Bistogram stretching is the most rudimentary and flawed approach to this process.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' #
Dynamic ,ange Compensation 1ynamic range compensation methods transform the dynamic range of one device or viewing condition into another. The most common dynamic range compensation methods involve transforming from the 1)))D1 or greater dynamic range of real4world scenes into the 1))D1 or less dynamic range of photographic film or many CC1 capture devices. There have been several traditional approaches to this problemD linear compression, empirical photographic curves, and &etinex processing. &etinex and homomorphic filtering models compensate for spatially local luminance adaptation and therefore lighten shadow areas and darken highlight areas without losing high fre9uency information. Cinear compression is commonly performed as histogram stretching. 6 histogram is taken of the image, and some percentage point is stretched to black and a second percentage *(#/ is stretched to white. The obvious problems with this approach are overcome in the following two methods. ?ssentially, this approach ignores all of the visual system. The photographic and graphics arts industries have incorporated a classical film curve into this compression ! . This has several advantages over the previous method. Bighlight and shadow detail are maintained perceptually, skin tones have an appropriate contrast, and the mid4tones retain detailed information.
Fadd in %uture dra%t: :ra0hG
Color Enhancement Methods Color enhancement methods attempt to correct for previous color reproduction system errors such as poor exposure or color balance. The four basic types of color enhancement methods areD ?xposure Tone Color balance +aturation
The three basic approaches to these methods are the followingD ?mpirical statistics <rior system processing knowledge ;ser preference4based solutions
+ome of these are automatic, and some re9uire manual or default implementations. +ome of these approaches are image dependent and some are not. ,bviously, image4dependent approaches tend to suffer performance problems.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' 6
E+POSURE ENHANCEMENT
+imply setting the maximum image values to white and minimum to black is the most rudimentary approach to this problem. Bistogram stretching is the second most rudimentary approach to this process and also confounds the results with tone enhancement.
TONE ENHANCEMENT
Tone enhancement methods attempt to provide reasonable contrast results.
COLOR -ALANCE ENHANCEMENT

Color balance enhancement attempts to eliminate color casts caused by previous processing methods. Bistogram e9ualization is the most rudimentary and flawed approach to this process.
SATURATION ENHANCEMENT
+aturation enhancement methods provide saturation compression or expansion for devices to match the references or user preferences. Media Intent Methods Media intent methods model the =look> of various media in an independent manner. ,ther devices or media can be made to simulate this =look> or it can be subtracted out when inappropriate, such as normalizing and aligning the mediaFdevice neutral axis. 5amut Mapping Methods :amut mapping methods compensate for the limited gamut of physical devices and try to sacrifice some colors with the least ob8ections. Typically, there are two approaches to gamut mappingD Mapping all the colors from an BK+ into the device gamut Mapping two gamuts together
7n either approach, traditionally hue has been weighted more important than lightness or chroma. This is a misconception. The idea is to hide the color artifacts in an imperceptible manner. +ometimes this involves significant hue shifts. :amut mapping methods should either be performed in an BK+ or use an BK+4based metric. This is to ensure that no accidental color shifts occur. 2or example, if one simply converts a monitor &:@ space into BC+ using the classic computer graphics e9uations, then when mapping highly chromatic colors to less chromatic color along the same hue and lightness vector, a visual hue and lightness shift will often occur. +ometimes this shift is dramatic.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' '
There are two basic approaches to gamut mapping. The first is to prioritize colors that at within both the source and destination gamuts and compromise of color that are not in the union of the source and destination gamuts. This allows for methods that can be modularized, such as the 7CC, but at potential 9uality loss. The second maps the source gamut colors into the destination gamut colors, optimizing volume usage. The difficulty in performing gamut mapping in an BK+ is that one must transform the device gamut into the BK+ also. 7n some devices, one wants to map saturated colors to more pure device states, which re9uires prior knowledge of the final device states and preferences in gamut mapping methods. @y implication, this re9uires the gamut mapping methods to be dependent on the final device states, which also makes them dependent on every transformation between gamut mapping and the final device color transformation. 2or example, with ink 8et printers, this includes black generation, half4toning, ink limiting, and possibly mechanical compensations such as shingling. +ome gamut mapping methods depend on image type such as presentation graphics, which is due more to half4toning artifacts than anything else. This also brings up an interesting issue of gamut mapping techni9ues that depend on spatial characteristics of raw device color space artifacts.
MEDIA LINEARI,ATION
Media linearization algorithms such as dot gain algorithms attempt to account for combinations of device problems.
CHANNEL GENERATION ALGORITHMS

Channel generation algorithms either generate additional color channels or planes or reduce them. This is most commonly found in black generation algorithms. @lack generation is the most common of these methods and is primarily used to save money!!. 3ith the advent of digital color reproduction, greater than four4color reproductions have become more popular, with the ability to predict output more accurately.
CONTINUOUS:DISCRETE ALGORITHMS
ContinuousF1iscrete algorithms convert device space color between continuous or greater discrete levels and lesser discrete levels including binary levels of color. This is often done in con8unction with spatial expansion of the colors, as in half4toning!".
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' (
These methods implicitly incorporate visual tradeoffs in spatial versus color resolutions. De ice Mechanism Compensation Methods 1evice mechanism compensation methods compensate for real4world limitations of devices that affect the final color reproduction 9uality and must be compensated for in a manner consistent with hiding these artifacts from the human visual system. ?xamples of such artifacts include banding, spatial non4uniformity, bi4directional printing, press speeds, and aliasing. In#%Media Compensation Issues 7nkFMedia compensation issues provide solutions to limitations found in the ink or media when previous color reproduction methods re9uest inappropriate combinations. 7nk limiting is a fundamental method in ink 8et printers to overcome inkFmedia limitations. 2or example, it might be physically possible to place four drops of each of four colors on a page at any position for a particular printer, but any more than three drops of any color will saturate the media. 7f dealing with combinations of pigment and dye inks, these heuristic algorithms become even more convoluted and complex.
A PROCESS
FOR
EVALUATING IMAGE 6UALITY
This section describes how to evaluate the 9uality of a color image. The following method evolved over a period of years in the photofinishing industry and has proven very effective in the past. These steps present a =bottom up> approach to image evaluation. That is, the small details are evaluated first and the overall, more general characteristics are evaluated last. This approach typically makes it easier to document image artifacts without being distracted by other artifacts. 7t is often easier to ignore the details you know about, and look at the overall problems, than knowingly ignore the general problems and examine the details. 6nother way to look at this method is as a progression from small spatial details to overall color issues. 3hen making corrections to images, the opposite approach can be taken to correct the most blatant defects first. This is not the only method, but it has been proven effective in real4 world situations for large production processes. 3ith this approach in mind, the following steps are takenD ?xamine the image for physical defects *scratches or chemical stains/ Cook for systematic artifacts *banding, noise, error diffusion patterns/
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
' 9
?valuate the focus or sharpness of the image ?xamine the tonal balance *including shadow and highlight details/ ?valuate the overall color balance *too greenM too redM/ ?valuate the colorfulness of the image *too vibrantM too paleM/ ?ach of these steps is discussed in the following sections.
IMAGE 6UALITY LIMITATIONS
OF
COMPUTER SYSTEMS
Computer4generated imagery provides a clear example of digital images that are theoretically limited only by the computer system. 3ith computer4generated imagery *using the right creation tools and applications/, the user has tremendous control of almost all sub8ective characteristics described earlier. The only exception is cognitive color constancy. There is no explicit color device associated with computer4generated imageryH so spatial resolution is not inherently limited except by the size of the software variable types *! 4bit integers, and so on/. -oise, almost by definition, is not a factor in computer4generated imagery. 6 strong argument could be made that computer4processing systems are intrinsically independent of device colors. This can be accomplished by encoding the colors in spectral radiances *or reflectances/, some C7? color space based on the human visual system, or clearly encoding the transformations between the image rgb color space and a C7? color space. This last method can be accomplished by using a number of standard rgb color spaces such as CC7&4%)(, CC7&4$)1, +M<T?, or -T+C. This assumes that the color space is somehow encoded in such a way that the receiver will be able to decode it appropriately. 6ll of these spaces are well defined and have transformations to and from the C7?I.J color space. .et, it is still more common to avoid using these methods, so the lack of device4independent color is listed as an inherent artifact. 6 weaker argument could be made that computer4based imagery can possess viewing4condition4independent color. This is only true if one of two methods is used. The first method includes the following stepsD ?ncode the color space and illuminant of the image into a device4 independent space ?ncode the viewing conditions into the image Bave an accurate color appearance model that can process this information along with the destination viewing conditions parameters @ecause there currently exists no accurate color appearance model, it is only possible to achieve an approximation.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( 7
6 second method is to mandate that the image be viewed on identical e9uipment in identical viewing conditions. This is often impractical except in laboratory settings. +patial addressability is limited by the amount of available storage memory. 6lthough this might not sound like much of a limitation, for "4bit rgb images, an '.# x 11 inch !)) pixel4per4inch image takes over " M@ of storage. Most commercial presses take =four4up> or =eight4 up> sheets, which multiplies this amount by a factor of " or '. 7t is not uncommon for publishing and reproduction houses to be handling !)) M@ images. 1ata throughput alone imposes some practical limitations on spatial addressability. ,ne goal is to make the spatial addressability of the ac9uired image be twice the maximum likely spatial addressability of the destination device. Color addressability and resolution are also limited only by available storage memory. Typically, "4bit rgb or ! 4bit cmyk images are used today, although "4bit C7?C6@ images are becoming more common. Bowever, with computer4generated digital imagery, very deep spectral data is possible. De ice Speci*ic Arti*acts The output of digital cameras possess three ma8or inherent artifactsD ,ptical flare Cimited spatial addressability &esolution and very limited color resolution
1igital cameras of interest typically use two4dimensional CC1 arrays as sensors. The rgb colored filters covering these arrays are either regularly distributed or distributed in a mosaic pattern to increase the greenFluminance image content. 3ith O<?: and other common compression techni9ues, mosaic arrays cause increase chromatic artifacts. This is due to the lack of processes to make the sensor image conform to the assumptions of the compression algorithms. The color resolution of digital cameras is also a serious constraint, usually in the range of $ to ' useable bits. This has a direct impact on image 9uality by reducing or eliminating shadow and highlight details. The spatial characteristics are limited by the size and cost of the sensorH !))) x ))) pixel sensors exist, but are very expensive. More common sensors are in the range of $") x "'), which produces a poor 9uality full4page image, considering that most high49uality prints are at least 1#) pixels per inch in resolution. CC1 flatbed scanners typically use a one4dimensional CC1 array *or three one4dimensional CC1 arrays/. +ome scanners use spatial color filtering techni9ues while others use temporal color filtering techni9ues. 6s with digital cameras, CC1 scanners also possess a limited dynamic
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
range, usually from ' to 1) useable bits. +patial addressability is usually limited to !)) or $)) pixels per inch with some higher performances using interpolation. <MT drum scanners use a single set of photo4multiplier tubes *a total of three <MTsEone each for red, green, and blue/ and move the negative, transparency, or print input image in two dimensions past the <MT set. <MT scanners are used primarily for the increased dynamic range *' to 1" bits/ and high spatial addressability and resolution. @y having a single pixel sensor and the appropriate optics, scanning resolutions of "))) pixels per inch are common. 4sing Secondary Sources ;sing secondary sourcesEsuch as photographic prints, half4toned prints, or videotape with ac9uisition devicesEcreates an additional level of image 9uality artifacts. 1ownloading digital images from the network also introduces additional image 9uality concerns. <hotographic prints are already limited by the color space of the emulsion dyes, granularity, MT2 of the enlarger optics, and noise. +canning an image with the same hues using an ?ktachrome, 6gfachrome, and 2u8ichrome print will produce three visually different images. This is due to the mismatch of color filters with the human visual system. @ecause photographic prints *and transparencies/ already have their tonal characteristics chemically encoded for viewing, it is difficult to maintain good tonal characteristics, such as shadow and highlight details. 2ingerprints and dirt are ma8or problems with using photographic negatives as secondary sources *in addition to the problems of modeling and converting the film negative color space into a visually relevant color space/. 7f you scan an image from a book, magazine, or poster, you are most likely using a half4toned print as a source *not to mention possibly violating copyright laws/. The most common problems occur when the scanning resolution is mismatched to the screening fre9uencies and moire patterns are enhanced.
Fadd in %uture dra%t: 0ractical guidelines to a"oid moire CStd screen angles and s0ot %unctionDG
Most videotape formats use interlacing and analog signals. This re9uires that some merging occur between the two temporally separated frames and the signal 9uantization. 7n addition, videotape usually has low spatial resolution and a large noise component. 4se a Standard Set o* Images ,ften, a standard set of test images are used to examine particular processes. The choice of images is very important to this process. The
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( )
test images should be of high 9uality for all the image characteristics described in this paper, to ensure that defects in a process are =real> and not inherent within the image data. ;sing a small number of standard images is recommended for these reasonsD These images are usually publicly available and have no royalty re9uirement provisions. Most images that we encounter in our everyday lives have some type of copyright or royalty restrictions *or both/ that legally inhibit their use and distribution. These royalties can be very expensive. +tandard image sets such as the 7T' or :6T2 sets have already been reviewed by panels of experts to verify their 9uality. The consistent use of a small number of images provides a familiarity that makes evaluating particular artifacts or new products much easier. The use of a common set of images among physically distributed sites significantly enhances communications when dealing with particular artifacts and can 9uickly resolve =simple> problems such as operator error or invalid setup parameters. 4se the Same /iewing En ironment Consistently 6 preliminary but important step in 8udging image 9uality is to set up a viewing environment. 1epending on the situation, this might be as simple as making sure there is a relatively uncluttered space in your cubicle, well lit and devoid of other images or vibrant colors. 6 more complex environment might include a Macbeth multi4illuminant viewing booth in a 9uiet room with a comfortable chair and controlled ambient lighting. 6lmost anywhere between these two extremes is a reasonable environment. 6gain, let your particular situation dictate the solution. 7f very high 9uality, consistent results for marketing and statistical analysis are re9uired, a more complex environment is appropriate to minimize any unwanted *or undetected/ influences or distractions of the observers. 2or routine investigations of prototype printing algorithms and artifacts, a simple environment is usually more convenient and effective. 7n general, elements to be aware of when setting up your viewing environment include the followingD 2ll)!inant intensit* levels. Bow bright is the lightM 7s it comfortable for viewing imagesM 2ll)!inant color. 1o you want to simulate printing standards *1#)/, photographic standards *1$#/, or a more typical office environment *2 /M .o #istracting i!ages or vi%rant colors. 1istracting images easily distort 8udgment, and vibrant colors distort illuminant adaptation within the viewing environment.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( 3
A relativel* co!forta%le setting. ;ncomfortable or distracted observers produce erratic results. Consistent )se of the vie3ing environ!ent. The more consistent you are, the easier it is to correlate your results ob8ectively with others. 6 magnifying glass or loupe is also 9uite useful in examining small image structures or details. 6 1)I or ")I loupe is usually most helpful. Examine the Image *or Physical De*ects <hysical defectsEsuch as scratches, water spots, CC1 detector dropout, or chemical stainsEare often the most apparent and ob8ectionable image defects. 3hen looking at the entire image, however, a small scratch might go unnoticed until the image is enlarged for the customer5s final print. ,ne easy way to inspect for physical damage is to use a one4half4inch to one4inch =viewing area> and scan the image by using a continuous snake4like visual movement from the bottom left to the upper right. This provides a consistent, easy path to follow visually while avoiding the viewing pattern, which provides an unconscious, false sense of familiarity that leads to overlooking defects. 6nother approach is to rotate the image 1') degrees to avoid ob8ect recognition. This approach is especially useful if the image contains text. ,ur amazing ability to recognize ob8ects or words and =fill in the blanks> is a definite hindrance when trying to locate and evaluate the existence of those very same =blanks> or defects. Therefore, explicit steps need to be taken to the work around that possibility. 6fter you scan the image for physical damage *and document where appropriate/, it becomes fairly straightforward to ignore these defects mentally and continue with the rest of the evaluation. $oo# *or Systematic Arti*acts +ystematic structural and spatial artifacts include things such as banding, granularity noise, streaking, process scratches, dithering and half4toning patterns, print headFpen dropout, compression, and sharpening artifacts. The physical viewing method is similar to that used for finding physical damageEthe same snake4like viewing patternEalthough a ")I magnifying loupe is often very helpful in examining small details in prints, such as half4toning patterns. ,ccasionally, by staring at the entire image and not focusing on any particular section, subtle patterns will become more apparent. The key to this step is the ability to pick out patterns *often very faint ones or those =hidden> by the amount of informational content or other image noise/. ;sually this kind of inspection is performed when looking for specific defects, such as research weighing the impact of different dithering
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( ,
algorithms. 7n this case, the observer already has a clear idea of what they are looking for and what it looks like. 7n the situation where one is not familiar with what defects are in the image, this process can be iterated, starting with the entire image viewed as a whole and repeating the viewing pattern with progressively smaller =viewing areas> until a very small =viewing area> is used. 6lthough somewhat tedious and time consuming, this iterative techni9ue often provides numerous insights into many aspects of the image structure. 6gain, you should let your particular situation dictate how extensive a process to use and which artifacts to concentrate on. E aluate the Focus or Sharpness o* the Image The next step evaluates the sharpness *or focus/ of an image. 6 general look at the overall image will provide a good =feel> for how sharp the image is *1oes it feel =crisp> or =fuzzy> or too =harsh>M/ -ext, locate the more finely structured ob8ects in the image, such as backlit hair, fine tree limbs, and gear threads. ?xactly what elements you choose depend upon the scene content in your particular image. Try to find fine, dark structures on a light, neutral background. 6fter you find them, use the magnifying loupe to examine them in detail. 7f you have control over the imaging process, it is possible to use resolution charts to measure the sharpness of that particular process or system ob8ectively. ,ne example would be to print dots and lines at the device resolution and measure their placement and size. 2or example, a printer manufacturer might advertise % ) dots per inch *dpi/ resolution but only be able to achieve $)) dpi accurately by using !)) dpi dots. ;nless the image or scene is explicitly designed and created otherwise, we usually want an infinite amount of sharpness or resolution. &emember that sharpness and contrast are different, so you can have very high resolution, sharp images that contain soft ob8ects. @ecause infinite sharpness is physically impossible, some 8udgment must be made about =what is sharp enough.> This 8udgment ideally would involve a close correlation with the intended audience for the images. 7f the audience or customer *or market/ is unknown, as in the case of images on the 3orld 3ide 3eb, try to determine which market the sharpness is acceptable for. 2or example, is the image sharp enough for professional photographers to acceptM Bowever, this re9uires an in4depth knowledge of the 9uality re9uirements for specific industries. Examine the !onal 6alance The previous steps examined artifacts that could be compared to some ideal measure, such as infinite resolution or a complete lack of physical damage and structural artifacts. 7t is significantly more difficult to
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( #
determine what is =good enough> when evaluating tonal balance, color balance, and colorfulness where there is no such standard. 7nstead, a series of indirect or comparative 8udgments are used, which re9uire the observer to determine what is =right> or =natural> *for example, correct skin color/. 6lthough this is known to be inaccurate with respect to instrumental measurements, it correlates well with how customers want their images to appear. ?valuating tonal balance involves evaluating the following four image aspectsD Tonal range Cocal contrast :lobal contrast ,verall =exposure>
To evaluate tonal range, pick the lightest and darkest portions of the imageD The lightest portion should be a white that matches the raw media, such as the paper stock or the minimum density of the film transparency. The darkest portion should appear very distinctly black relative to the rest of the image, as opposed to very dark gray. 7n some cases, the artist or designer explicitly designs an image without whites or blacks, but this is relatively rare. Cocal contrast determines the amount of information or detail in the =local> portion of the tonal range. <ick three gray levelsEdark gray, light gray, and a medium grayEand identify portions of the image that contain each one. 2or each portion, examine the amount of detail or information contained in that areaD 6re the fine details distinguishable in the shadow areasM 7n the highlight areasM MidtonesM 7f not, this might indicate ob8ectionably poor local contrast. 2or gray4scale images or very demanding markets, you might also pick two additional gray levels to 8udge, bringing the total to five. -ow, mentally and visually move back from the image and look at the overall contrastD 1oes it appear too muddyM Too harsh or starkM 1oes the contrast match the messageM 2or example, high contrast often enhances aggressive messages, whereas lower contrast can enhance more subtle images. 2inally, evaluate the overall image =exposure>D
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( 6
6re the primary elements in the image easily visibleM 6re they comfortably viewableM 1o the natural ob8ects *like skin tone/ have the appropriate densityM ;sually the primary elements of attention in an image are within the middle of the tonal range. E aluate the + erall Color 6alance Color balance evaluation is very similar to the previous evaluation of tonal balance. 7t consists of evaluating local and global color balance. ;sing the same gray portions that you used for evaluating local tonal contrast, determine if these areas are neutral *or =correct>/ or if they contain some unwanted *or =unnatural>/ colorcastD 6re the shadows too greenM 6re the highlights too pinkM *Clouds are a good element to study highlight color balance./ ,ften such inappropriate colorcasts are complementaryEfor example, yellowFblue or greenFredEbecause of some inherent process limitations of certain technologies. ?xamine recognizable elements in the image for correct color balanceD 1o the skin tones appear =right> or are they too yellowM Too redM Bow about the blue sky regionsM CloudsM :rassM 7f not, determine in what way they appear =off.> 2inally, look at the entire image as a whole. 1oes it appear well color4 balanced or is there some overall colorcast to itM E aluate the Color*ulness o* the Image The final step is to 8udge the overall colorfulness of the imageD 1oes it appear too vibrant or too paleM 6re there contours near the saturated or neutral colored areasM Bave the areas of saturated color lost informational detailM 6ll these 9uestions, plus the 9uestions from the previous steps, indicate various artifacts of different image 9uality characteristics. To provide better support for these evaluations, a set of image characteristics are described in the following section.
PRACTICAL SUGGESTIONS
2or color reproduction on the 3orld 3ide 3ebD use s&:@, embed 7CC profiles, use 2lash<ix or <-:, and calibrate all devices.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( '
;se application4based color management solutions Calibrate devices and use standard viewing conditions, and embed profiles or use s&:@. ;se operating system4based color management frameworks
Fadd in %uture dra%t: more 0ractical in%ormation and suggestionsG
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( (
IEC Color Standards

7t has long been foreseen that faithful production and reproduction of color information would be re9uired in a global digital network environment, because the 3eb is an open system. This means that anyone can produce color images anytime to be received by anonymous people who will display, make prints, or store the images for future use. @ecause of the lack of an alternative information infrastructure for global communication with color information, the 3eb4based communication should be increased, provided that bandwidth of communication supports the increase. 7n this context, one needs a management scheme in terms of faithful production and reproduction of color information that will closely relate to color information products and the systems available in the worldwide market place. The color management scheme will provide a target for design for manufacturers and evaluation criteria for test houses and consumers. To meet the needs of markets, consensus in international level is necessary in the areas ofD Color measurement and characterization of relevant e9uipment such as color scanners and digital cameras, color displays and pro8ectors, and color printers, 6 standard color representation by digital data, in other words digital color space used for e9uipment4independent color communication, Methods to tune e9uipment4dependent characteristics to the standard digital color space, or color management, and finally, ,b8ective methods to evaluate a degree of attainment of color management.
APPROACH
To realize an open color communication system, Technical Committee 1)) of the 7nternational ?lectrotechnical Commission *7?C/ decided to establish a <ro8ect Team on color measurement and management in multimedia systems and e9uipment in 1resden in +eptember 1(($. 7?CFTC 1)) asked the <ro8ect Team to develop a series of 7nternational +tandards in 7?C $1($$ series in parallel by means of a new process of standard development in use of modern way working within the <ro8ect Team. The team has made use of =virtual meetings> in which every interested expert is invited to participate to contribute hisFher idea in use of e4mail, ftp and the 3eb. There were no physical limitation of apparent budgets, meeting places, rooms, and occasion for the virtual meetings. &e9uired are the ability to share the time and knowledge of participants who aim at the common targetD 7nternational +tandards.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
( 9
The virtual meetings are possible with clear and strong support of its parent Technical Committee and Chief ?xecutive ,fficer of the 7?C, together with actual face4to4face meetings *physical meetings in contrast with virtual meetings/.
POSSI-LE ARCHITECTURE SYSTEMS
FOR
COLOR MANAGEMENT
IN
OPEN
+pecific aspects of open systems like the 3orld 3ide 3eb is that effectively anonymous creators and users are involved, and that ob8ectives of color communication are not specifiedD some are for remote learningFeducation, some are for e4commerceFe4business, and some for remote diagnostic of patientsH they will range from entertainment to professional applications. The open color communication system will be composed of e9uipment of ac9uisition of color information such as color scanners and digital stillFmovie cameras by which information is digitized and put to the 7nternetH additional e9uipment of reproduction of color information retrieved or received from the 7nternet such as color image displays and pro8ectors manufactured by varieties of technologies and color printers for hardcopies. The combination of these two classes of e9uipment via the 7nternet will not known in advanceH however, creator and consumer of color images and electronic documents that contain color image always wish they are appropriately used, viewed, and reproduced. 6 model of the open color communication system is shown in 2igure #, where any combination of input and output e9uipment is possible.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 7
+ig)re 5. Mo#el of an open color co!!)nication s*ste!
7nternational digital network is not always =international,> but it could be regional, national, or local area network. The important aspect of this model is that a source site does not know a destination site where a particular kind of e9uipment is in useD e9uipment model, technology used, manufacturer, and date of manufacturing are unknown. -evertheless, both parties are apt to expect faithful color reproduction. Bow can we realize a color management system to answer their expectationsM ,ne answer will be based on communication with a header of description of color information coding such as the 7CC <rofile format. 7n this case, every creator of color information in digital form should prepare the header to be attached to the images. Consumer side, the header should be appropriately interpreted and processed in reproduction. 7t will cause increase of the size of color information and computational complexity to handle the information contained in the header, normally pixel4by4pixel interpretation. The other answer is to introduce default method of digitizing color information based on an internationally standardized method. 7n this case, there will be no header attached. -o additional pixel4by4pixel computation will be re9uired, if e9uipment for color information are designed to accept the standardized color encoding scheme. 7t will provide benefit for consumers, manufactures, and 7nternet service providers from the point of view of reduced amount of information transfer between two parties.
RESULTS
OF
INTERNATIONAL STANDARDI,ATION
IN
IEC:TC 133
6s the result of international standardization activities by Technical Area 2: Colour measurement and management of 7?CFTC 1)) based on the second model in the last section E that is, introduction of a default &:@ color space for color communication in open system E the following 7nternational +tandards in 7?C $1($$ series have been established. 7?C $1($$4 41D 1(((, Multimedia systems and e9uipment U Colour measurement and management U <art 41D Colour management U 1efault &:@ colour space U s&:@ 7?C $1($$4!D ))), Multimedia systems and e9uipment U Colour measurement and management U <art !D ?9uipment using cathode ray tubes 7?C $1($$4"D ))), Multimedia systems and e9uipment U Colour measurement and management U <art "D ?9uipment using li9uid crystal display panels 7?C $1($$4#D ))), Multimedia systems and e9uipment U Colour measurement and management U <art #D ?9uipment using plasma display panels
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7?C $1($$4(D ))), Multimedia systems and e9uipment U Colour measurement and management U <art (D 1igital cameras There are active <ro8ect Teams for 7?C $1($$4 4) *Colour management/, 7?C $1($$4 4 *?xtended &:@ colour space U s&:@$"/, 7?C $1($$4 4! *1efault .CC colour space U s.CC/, 7?C $1($$4% *Colour printers/, 7?C $1($$4' *Multimedia colour scanners/ under Technical 6rea *T6 /. +tandardization on pro8ectors *7?C $1($$4$/, based on re9uests from the industry, is expected to be started. 6ll information of the activities of 7?C TC 1))FT6 will be available from the 7?C 3eb site maintained by 7?CFTC 1)) at httpDFFwww.iec.chFtc1))FtxtF1))strucXta .htm and detained additional web site maintained by T64manager at httpDFFwww.map.tu.chiba4 u.ac.8pF7?CF1))FT6 .
HOW COLOR MANAGEMENT -E REALI,ED

The color management is based on characterization of relevant e9uipment, which will have potentiality be used for color communication in the open system, in other words, to be connected to the 7nternet in one or the other methodsH and the incorporation to the standard &:@ color space. 6 scheme to make use of date obtained in the process of characterization of a specific color production and reproduction e9uipment will be modeled as shown in 2igure # and 2igure %.
S$)#;)r; RGC l ur !')"e Normalization ? T E<$e#;e; 3 " l ur !')"e , & L
Digital cameras/scanners
Cight ,b8ect
,pt4electric
7nputFoutput characteristics
Characteristics
+ig)re 4. Color !anage!ent sche!e for color i!age pro#)ction
7n 2igure $, opto4electric conversion is the core function of color image production whose input *light or ob8ect/ to output *data/ characteristics should be ac9uired by the common methods specified by 7?C $1($$4' for digital cameras and by 7?C $1($$4' for multimedia color scanners.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 )
S$)#;)r; RGC l ur !')"e ? T 3 , & L
1isplaysF<rinters 1ata4to4light ob8ect conversion
Normalization
E<$e#;e; RG" l ur !')"e
;sers
7nputF,utput characterization
Characteristics
+ig)re 7. Color !anage!ent sche!e for color i!age repro#)ction
7n 2igure %, data4to4lightFob8ect conversion is the core function of color image reproduction whose input *data/ to output *light or ob8ect/ characteristics should be ac9uired by the common methods specified by 7?C $1($$4! *C&T display/, 7?C $1($$4" *CC1 display/, 7?C $1($$4# *<1< display/, 7?C $1($$4$ *<ro8ection system/, or 7?C $1($$4% *color printer/, depending on the actual type of e9uipment. The activity of 7nternational +tandardization was initiated by the Oapanese -ational Committee of 7?CFTC 1)) in 1((%, but now it is one of the ma8or pro8ects of 7?CFTC 1)). 7n March ))), the framework of the parent technical committee was changed to purely pro8ect based work by a small pro8ect team corresponding to each approved pro8ect. Bowever, existing pro8ects on color measurement and management have been allocated under a newly established Technical 6rea supervised by Technical 6rea Manager *T6M/ and Technical +ecretary *T+/. 7?CFTC 1)) is now disbanding all 3orking :roups and +ub4committees and restructuring to a new organization. This restructuring will be successful in terms of even more speedy and market relevant 7nternational +tandardization for not only color measurement and management, but also other areas of multimedia systems and e9uipment.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 3
!erminology
The field of color management has been getting an increasing amount of media attention during the past decade or so in addition to focused development efforts by a number of leading imaging companies. This attention has made it the focus of recent addresses at +eybold +eminars by +teve Oobs of 6pple Computer!# and @ill :ates of Microsoft Corporation!$. 6 9uick web search turns out thousands of articles and web pages about color management !%. +till, there seems to be no one completely satisfied with the results to this point. This is evident when companies like 6pple Computer !', Canon 7nformation +ystems, and ?astman Lodak!( evangelize alternative color management 6<7s to the native operating system4based 6<7 suite. 7t is also evident from numerous articles by industry pundits pointing out the complexities") or shortcomings"1 of current solutions along with their fears of any complementary technology" . The 7CC itself is struggling to investigate solutions to known problems "!. Importance o* !erminology :iven the diversity of core technologies contributing to color management solutions, it is extremely important that they communication clearly. Bowever, even such basic terms and =white,> =black,> =linear,> =gamma,> =rgb,> and =cmyk> are typically miscommunicated across industries. The C7? and 7?C 8ointly publish the foundation reference on color terminology, The International Lighting Vocabulary. Bowever, this reference is seldom used effectively and needs continuous updating to keep up with scientific progress in the field, such as the recent adoption of a single color appearance model by the C7? "". De*ining )hite and 6lac# 1r. &obert Bunt has provided an excellent review of the importance of well4defined terms for =white> and =black.> 7n summary, one must include the viewing conditions that dramatically affect the appearance of the colors. 7n additional the common viewing condition parameters such as surround, and chromatic adaptation, one must also be careful to describe whether it the white refers to a perfect reflecting diffuserH the targeted media white point or even a self4emissive device. +imilar problems exist for describing black points. 3hile flare or veiling glare is a common physical effect, it is rarely considered in color management systems. 3hen flare is considered, it is rarely with the additional effects of surround and simultaneous contrast, which dramatically impact the appearance of the black point "#. The 7+, !$$" international draft standard on illumination conditions provides some additional progress in specifying viewing conditions for
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 ,
the graphics arts and professional photographers. Bowever, it explicitly excludes guidance for cross4media comparisons or viewing conditions that reflect practical home or office conditions. 7n summary, seldom do industry practices effectively incorporate enough information to clearly communicate the meanings of =black> and =white> such that they are reproducible by a third party. $inear )ith ,espect !o7 6nother difficult term to understand is the color =gray.> 7n addition to the viewing4condition problems described earlier, there is an additional problem describing gray. 7t is common industry practice to refer to color spaces as linear or non4linear. Bowever, there is little description of what this linearity refers to. <oynton aptly points out the degeneration of terminology in the television industry when a single prime indicated the difference%. .et, there is still a fundamental misconception among many engineers on the existence and need for two classes of color spaces and thus two types of linearity. These two broad classes of color spaces are used by image creatorsD we term them =intensity> and =intuitive.> The intensity space is =linear with respect to intensity> and is also commonly known as luminance. <hysically based modeling is a good example of the first classH computer paint programs are good examples of the second. The intuitive space is =linear with respect to perception> and is also commonly known as lightness. These two classes applications both work on &:@ displays, but they use the frame buffer in different ways. 6ll !1 software algorithms *for example, polygonal rendering, ray tracing, and radiosity/ work in intensity space, because they all attempt to model the physics of the real world. The differences between the algorithms are in the accuracy of the modeling and, conse9uently, the performance of the algorithms. The most accurate applications work in a luminanceFchromaFchroma space and do not convert to an &:@ space until the image is ready to be displayed. Bowever, such algorithms are very slow and rarely used. Most algorithms that can be accelerated by hardware work in an &:@ space. The frame buffer is often used to store both partial and final results. 6lthough interpolation in &:@ space yields results that are only rough approximations of what is physically correct, these results are suitable for many applications. 7n an &:@ intensity space, a mid4gray would be represented as approximately *).1', ).1', ).1'/. This value is then scaled by the size of the frame buffer and stored as an integer approximation. 2or example, in a !)4bit frame buffer, that mid4level gray would be represented as *%!%, %!%, %!%/. <hysically based modeling tends to need deeper frame buffers, so that there are enough bits to represent shading in the darker tones. 7n an '4bit frame buffer mid4gray would be only *"$,
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 #
"$, "$/ leaving only "# gray levels. Contouring artifacts are occasionally visible in such conditions. <aint and other programs that work in an intuitive space can use the frame buffer more efficiently. 7n intuitive space, the mid4level gray is represented as *).#, ).#, ).#/. ?ven on an '4bit frame buffer this provides 1 ' shadow tones, which is 9uite ade9uate for the display technology in typical viewing environments. This is not the only reason that such applications work best with mid4 gray as the middle digital count. <aint programs often provide color pickers that allow users to directly select &:@ values, either through numeric entry or through linear sliders. 7n both cases, the user will expect the middle intensity to be in the middle of the range of values. The failure to recognize the difference between these intensity and intuitive spaces has led to great confusion. -one of the color standards we examined actually represented their mid4gray as precisely *).#, ).#, ).#/. -one of the commercial systems sets mid4gray at *).1', ).1', ).1'/ either. This is one of the reasons that &:@ color standards fall short of users5 expectations. 7n summary, seldom do industry practices effectively incorporate enough information to clearly communicate the meanings of =gray> or =linear.> 5amma ,ne attempt to resolve the =linear> terminology problem is with the term =gamma.> Bowever, the term is not used consistently. This inconsistency greatly adds to the level of confusion. Latoh and 1eguchi"$ published an excellent review on C&T characteristics in which they defined seven different but useful gamma terms 8ust within C&T devices. 7n this same paper, they pointed out how gamma is commonly used across incompatible modeling e9uations. This misuse has lead to many common myths about the default =gamma> of C&Ts being ." or .# or .' or 1.'. 7n summary, seldom do industry practices effectively incorporate enough information to clearly communicate the meaning of =gamma.> 5amma
DEFINITIONS
OF
GAMMA
3e start this discussion by defining the separate aspects of gamma. Kiewing gamma. This is the overall system gamma that we want to obtain, and is typically computed by multiplying the camera gamma by the display gamma as shown here.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 6
display gamma = CRT gamma x LUT gamma
(0.1)
Camera gamma. This is the characteristic of the image sensor or video camera standard transfer function. C&T gamma. This is the gamma of the physical C&T. C;T gamma. This is the gamma of the frame buffer lookup table 1isplay gamma. This is the =display system> gamma downstream of the frame buffer, which is typically computed by multiplying the C&T gamma by the C;T gamma as followsD
viewing gamma = camera gamma x display gamma (0.2)
These definitions do not describe the individual gamma parameter in e9uation ).". 7nstead, they describe the resulting power parameter of the appropriate transfer function when fit by a power function. 7t is extremely important to keep this distinction clear. ,therwise, one assumes that e9uations )." and ).# are e9uivalent, the system black level is truly ).), and the system gain is 1.).
VIEWING GAMMA
The reason that a viewing gamma of 1.1 # is used instead of 1.) is to compensate for the viewing environment conditions, including ambient illumination and flare. Bistorically, viewing gammas of 1.# have been used for viewing pro8ected slides in a dark room, and viewing gammas of 1. # have been used for viewing monitors in a dim room. This dim room value of 1. # has been used extensively in television systems and assumes an ambient luminance level of approximately 1# lux. The current s&:@ standard assumes an encoding ambient luminance level of $" lux, which is more representative of a dim room in viewing computer4generated imagery. +uch a system assumes a viewing gamma of 1.1 # and is thus consistent with the %)( standard described later in this paper. 6lthough we believe that the typical office or home viewing environment actually has an ambient luminance level around )) lux, we found it impractical to attempt to account for the resulting large levels of flare that resulted. 7n addition, recent work by the 7+, OT6: standards committee supports the ambient luminance level of $" lux. 7f the viewing condition is different from the standard, then the decoding process must compensate. This can be done by modifying the gamma values in e9uation ). by the appropriate factor. 7f one does modify the gamma values in e9uation 1. , extreme care must be taken to avoid 9uantization errors when working with "4bit images and high viewing flare levels. The 7T;4& @T.%)( transfer function in combination with its target monitor is attempting to achieve a viewing gamma of 1.1 # by incorrectly assuming a C&T gamma of .# and an C;T gamma of 1.)F . as shown in the following e9uation. The 8ustification of a
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 '
viewing gamma value of 1.1 # is described in the section on viewing environment compensation.
(0.3)
;sing the actual power function fit value for the %)( transfer function of 1.)F1.(#$ and maintaining the display gamma of 1.1 #, we can solve for the ideal target monitor gamma of . . This is consistent with the C&T gamma value and the 7?C $1($$4 41 s&:@ color space.
CAMERA GAMMA
The camera gamma 1.)F . was the standard for television camera encoding before the advent of color televisions, and was formalized in 1(#! with the -T+C broadcast television standards. More recently, 7T;4 & @T.%)( has been adopted internationally and contains camera gamma of 1.)F1.(#$. The actual exponent factor in the %)( transfer function is 1.)F . . 1espite the fact that the exponent of the %)( function is 1.)F . , the actual %)( encoding transfer function is closer to a C&T gamma of 1.)F1.(#$ than 1.)F . . This is due to the large offset of ).)(( in the transfer function e9uation. This is well matched to the eye5s own non4 linearity and it helps minimize transmission noise in the dark areas. @roadcast television camera gamma standards, and the 7T;4& @T.%)( standard in particular, define the transformation of real4world C7?I.J tristimulus values into a target &:@ monitor space. This is essentially a composite of two transformations. The resulting image is not an exact appearance match of the original scene, but instead is a preferred reproduction of the original scene that is consistent with the limitations of a monitor. @ecause all television sets have to display content generated with this encoding, it was very important for all C&T gamma designs to conform to it. ,nly recently has the computer monitor market become as large as the television market. 6s a result, most computer monitors still perform optimally with imagery using with a camera gamma value of approximately 1.)F1.(#$.
CRT GAMMA
The non4linearity of the electro4optical radiation transfer function of C&Ts is often expressed by a mathematical power function exponent parameter called gamma. This transfer function describes how much visible radiant energy *cdFm / results from voltages applied to the C&T electron gun. @ecause most of the other characteristics of C&T4based computer monitors are linear *including 16Cs and video amplifiers/, the resulting
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 (
transfer function has the same gamma value determining its non4 linearityD
2 5 A6k1D 7 k28ga!!a 60.-8
3hereD
k1 D A 2
and k2 are the system gain and offset
is the normalized pixel value is the maximum luminance of the C&T
is the resulting luminance
This e9uation and a thorough analysis of the C&T characteristics and history are well described ". The key point that we want to convey is that gamma component of the C&T gamma is dependent only on the electron gun design, and the vast ma8ority of monitors and TK sets of today are based on designs that result, on average, in the value . for gamma component of the C&T gamma and a . overall C&T gamma value when typical system gain and offsets are optimally set. Most of the variations between computer monitors and television sets are due to the differences in system gain and offsets *k1 and k /, which are partially under user control in the form of contrast and brightness knobs. Bowever, the actual setup is often not known, but the best C&T performance occurs when the system offset puts the dark parts of the images at the C&T cut4off, for example, the black *pixel value )/ parts of the C&T image are 8ust about to emit light. ;nder these conditions, e9uation )." becomesD
2 5 ADga!!a 60.58
and the monitor has the widest4dynamic range. This is, however, not the common condition. The simplified form in e9uation ).# is what is typically found in the computer literature. There are significant variations, with widest variations being in the set4 up and screen reflectivity *older and less expensive display can reflect up to ) percent of the ambient light/. These factors typically cannot be characterized ahead of time because they might change in the course of the day *ambient light/ and at the whim of the user *contrast and brightness modifications/. Bowever, in practice, the process tends to be self4regulatory, with users looking for darker places to set their monitors and modifying the controls to re4establish the expected display appearance. ?xhaustive testing carried on at Bewlett4<ackard on K:6 computer monitors from many brands has shown the average C&T gamma to be indeed . , with a standard deviation of about ). .
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
9 9
LUT GAMMA
Two special circumstances will lead computer systems to deviate systematically from the . C&T gamma and the 1.) C;T gammaD color dithering for 1$4color systems and system4imposed gamma correction through C;T. The first was common until a few years ago. ;ntil about 1((!, most 3indows4based <Cs were well described by a display gamma of 1.' because despite having . C&T gamma systems, the colors were dithered into the "4bit frame buffers, resulting in a flattening of the system transfer function. This happens because screen dithering mixes colors linearly in the eye, making it less dependent on the C&T non4 linearity. @ecause most 3indows4based <Cs currently support 1$4 or "4bit color modes, . C&T gamma is now the average. The second systematic deviation happens when the graphics system in the computer hardware or software imposes its own gamma correction. This is done for a variety of reasons, but is usually an attempt to compromise between image display and graphicsFimage processing performance. Most computer graphics rendering assumes linear radiation space, that is, transparency operations, and so does image processing, that is, scaling and filtering. The gamma correcting of image data can be described by applying an exponent to the image data. 2or the Macintosh, the display gamma is around 1.#%1 using a C;T gamma of 1." * . F1." N 1.#%1/. 2or +:7 workstations, the display gamma is around 1. (" using a C;T gamma of 1.% * . F1.% N 1. ("/. There has been significant confusion derived from assuming the C&T gamma value is identical to the exponent in e9uation ).". This has led to many claims of C&T gamma values of .# for video, 1.' for the 6pple Macintosh, and 1." for +:7 monitors. Bowever, it has been our experience that this misconception is not well founded in the actual physics of the displays and solid measurements. De*ining ,56 7t is common to communicate colors as &:@ values and think that this is meaningful. 2or example, a value of ##,),) is supposed to represent red, but what =red> is thisM 7s it an orangish4red, a dark4blood red, a pinkish red, or whatM :iven the current &:@ standards and incompatible &:@ devices, it is impossible to answer this 9uestion. 6 representative, but by no means exhaustive, list is included at the end of this section. 7n addition to the incompatibilities between the standards used in different individual technologies, there are several separate technological areas for which multiple incompatible &:@ standards are defined *broadcast television, vision research, robotics, and operating system defaults/. Oust within the field of broadcast television standards,
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7 7
there is often ambiguity within a single &:@ space as to whether it is gamma4corrected or not. 2inally, many standards *:L+, K&MC, :72, and so on/ do not provide any definition of &:@ at all, but assume that there is one ubi9uitous and unambiguous &:@ space, which further exacerbates the problem. 6 file in one of these standard formats could be in any of the color spaces in the table at the end of this sectionEand these represent only a small sample of the &:@ color spaces in common practice across industries. This ambiguity in the controlling color space can potentially be costly and even life threatening. 7f that sounds extreme, consider the effects of misinterpretation of color information in the fields of web4based catalog shopping and telemedicine. 6 customer ordering the wrong color 8acket and shipping it back can cost the vendor money. 6 doctor misdiagnosing a problem could cost the patient his life. 7n color as in all things, miscommunication can be expense and fatal. 7f one considers a CM. space to be a simple inversion of an &:@ space, which is done often in practice, this greatly increases the number of conflicting definitions by adding the numerous incompatible densitometry standards *6, M, T, to name only a few/.
SM*1! &* ,#,' !8? 3) 3,9 Microso%t#) IS461C 37## 8F$#( Smith and *or.ony6 K*!: 91SC 9#3#7 A00le#3 Sony#6 I1?H& 81.67 #9 CI! 93 s&:8,( V&M3# S:I#, C:A1S#' I1?H& 81.'7967 CI! 96,6)
7n summary, seldom do industry practices effectively incorporate enough information to clearly communicate the meaning of =rgb.> De*ining CM89 6 very similar problem to that described earlier exists when communicating color with CM.L. 7t is interesting to note that the term =cyan> is a relatively recent one for the graphics arts and was adopted explicitly to avoid confusion with =blue.>$! ?ven if one indicated that they are using +3,< cmyk, there is still not enough information for reproducible communication. +ome of the critical missing pieces include typical viewing condition information, color appearance models and the underlying models or e9uations of some of the parameters. Mc1owell has published an excellent review of the progress of current CM.L standards $". ,ne of the more provocative proposals is the concept of reference printing conditions. This proposal is conceptual similar to the s&:@ standard for &:@ devices, but instead provides a series of standard CM.L color spaces. 7n summary, seldom do industry practices effectively incorporate enough information to clearly communicate the meaning of =cmyk.>
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
)ho Decides )hat is :,ight;< 3e examined how various common color terms are confusing and miscommunicated. 7t is reasonable to ask why aren5t there already well4 defined standards would provide definitions for all of this. ;ntil recently, it was difficult to see how any of the current standards organizations could solve this problem. Currently, the organizational structures of the ma8or standards bodies seem to be almost incompatible with each other by design. +ome seem based on horizontal technologies across systems *C7?/, some seem based on vertical technologies *7+,/, some seem to have a matrix structure *OTC1/ and some are in transition between structures *7?C/. These different approaches while all potentially valid and useful, lead to inevitable overlap of charters and responsibilities. Cooperation via the liaison procedures is often only a matter of exchanging documents, not true 8oint efforts. +ome 8oint efforts are tactical collaborations for closed4loop color reproduction systems. Bowever, their impact can be felt far beyond their written scope, again leading to confusion in the industry on which standard to follow. This approach is at odds with producing a cross4 technology standard that is compatible with most industry practicesH a strong re9uirement of the computer based industries. The 7?C has until recently been focused on hardware, in particular connectors and not the firmware that is an integral part of the hardware design in electronics. 3ith the advent of 7?CFTC1)), this focus seems to be shifting to a much broader systems approach, but again, cooperation with other relevant bodies is still to be determined, although a strong emphasis on industry consortia is being considered. The 7+, is organized by vertical technologies such as photography *TC" /, cinemaphotography *TC4!$/, graphics arts *TC41!)/, and does not have any display4centric TCs. This vertical organization precludes the development of broad, cross4industry solutions. Traditionally, these TCs are focused on high4end, professional standards, sometimes to the detriment of the consumer market needs. The 7T; functions as a giant technical committee focused on telecommunications *7T;4T/ and receivers *7T;4&/, in particular, broadcast television. 7t is actually a part of the ;nited -ations and structurally independent of 7+,, C7? and 7?C. Their focus has the same weakness in lack of cooperation and broad cross4technology vision, as does the 7+,. 2ocusing on one specific field is a good way to make progress within that industry, but does not provide the broad vision needed to solve cross4industry and cross4technology problems. The OTC1, seemingly designed for =8oint technical> efforts, has historically lacked a firm connection to industry needs. This was dramatically illustrated by the ,16 and :6C17 architectures, which were developed completely independently of the computer industry that would have to implement this massive endeavor. 6lthough there was
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7 )
industry involvement in these efforts, operating system vendors, who drive architectural changes in computer systems, were not heavily involved. The C7? would appear to be the ideal body for broad color standards, but its organizational structure and bylaws are completely incompatible with modern technological developments. The C7? is organized to move at a much more deliberate pace. That pace is suitable to standardization in industries where change happens slowly. @ut in the technology4driven fields of today products cycles from development to obsolescence can occur in less than three years. 7ndustry consortia or collaborations have recently made significant impact in color management standards as witnessed by the broad adoption of the 7CC and s&:@ efforts. .et, narrow proprietary concerns and 9uestionable intellectual property issues plague these relatively informal bodies. 2inally, individual companies with a ma8ority market share occasionally establish de facto standards in sheer frustration with the entire standards processes described here. These efforts are often conceived with little input from other experts in the field and suffer from significant technical flaws that must be painfully addressed in the open market. ?ach of these bodies and their technical committees depends on the volunteer efforts of experts in the industry. Bowever, with color management, this expertise is rarely available for such unselfishness without some assurance of reasonable progress. 2inally, many of the experts in narrow technical committees are committed to advancing only the needs of their particular field and have little experience in other markets. .et, some solution must be found or all of the markets will be limited by their inability to unambiguously communicate color across technological boundaries. &ecently the Ooint Technology 6dvisory @oard of 7+, and 7?CEwhich resolves all policy and scope disputes across 7+,, 7?C and OTC1E mandated that 7?C TC1)) would administer color management standards for these organizations. This should help provide some help in resolving terminology issues. 3hether this occurs will depend upon the cooperation of individual technical committees, especially those of 7+, and OTC1.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7 3
Suggestions
The following list provides suggestions that have been found useful in resolving the terminology problems discussed in this paper. 1. 1o not provide parameter values without the underlying e9uations or models. 3ithout including the underlying e9uations or models, even by reference, parameter values are misleading at best. . To help clarify the meanings of white and black, the appropriate viewing conditions should be included and include all of the re9uired parameters in the C7?C6M(%5s viewing conditions such as surround, luminance level, and so on. The amount and methods for flare and simultaneous contrast compensation is also re9uired. !. 1o not use the term =linear> without completing the phrase =linear with respect toG> ". 1o not use the term =gamma> without clarify ad8ectives. ,ne suggestion is to use more generic terms such as =simple exponential curve gamma> and =complex exponential curve gamma> to describe differences in some C&T modeling e9uations. #. 1o not refer to &:@ colors without referring to a very well defined &:@ standard such as s&:@1( that includes a reference viewing condition, observer, device and color appearance model. $. 1o not refer to CM.L colors without referring to a very well defined CM.L standard that includes a reference viewing condition, observer, device and color appearance model. %. ?ncourage that all imaging application color standards be developed within the C7? in the future and encourage your national standards bodies to support this transition. This all might sound overwhelming, but there does exist a mechanism to make it much simpler. This would be to refer to either standard or device 7CC profiles when describing many of these terms. 3hile not yet perfect, the 7CC is attempting to provide clear, unambiguous communication standards for color management. 7t would also be beneficial if the C7? provided guidelines that either adopted these suggestions or improved upon them as a part of their new imaging division.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7 ,
A'out the Author

75ve been trying to understand how to reproduction color images almost my entire life. 7 remember watching in fascination as the color animation cells were captured one frame at a time into motion pictures. 7 copied my first slide and processed my first roll of color duplicating film when 7 was twelve. ,ver the next twenty years, 7 had the uni9ue and fortunate experience of participating in a family business that grew into the largest slide reproduction and collection image digitization laboratory in the ;nited +tates. +lide reproduction is a hybrid field that spans consumer and professional photography, motion pictures technology, presentation graphics, and computer graphics. 1igitization of large image collections is another hybrid field that spans television technology, professional photography, motion picture technology, library science, and computer programming. 1uring this time, 7 learned every technical and production management aspect of these fields as well as 7 couldEfrom motion picture film processing, to teaching color image evaluation and correction, to optical and contact printing, to designing a closed4loop calibration narrow4band additive light source. This time provided me with the uni9ue opportunity to be a craftsman in a number of traditional color reproduction industries. 3hile 7 understood the practical aspects of each of these fields, 7 still didn5t understand some of the fundamental science. 7 knew what colors and tones were possible to reproduce with various techni9ues, and what wasn5t possible, but 7 didn5t understand why. 7 returned to school to obtain a Master of +cience degree in Color +cience from Munsell Color +cience Caboratory at &ochester 7nstitute of Technology, to understand why colors reproduced as they did. 6fter several years of school, 7 was amazed that modern4day color reproduction works as well as it doesY The complexity of color science built into photographic film is truly awesome. +till, 7 didn5t understand many aspects of modern4day color reproduction. 7 spent eighteen months at 6pple Computer as the color engineer on Color+ync .), and as the founding chairman of the 7nternational Color Consortium. 7 wrote my first production color printer processing pipeline, and more followed. 7 learned half4toning, desktop publishing, and computer system architecture re9uirements. 1uring the next three years, 7 worked at Bewlett4<ackard Caboratories, one of the world5s foremost industrial research laboratories in color reproduction. 7 had the opportunity to learn a variety of different color technologies, from fundamental modeling standpoint, including ink8et printer, electrophotographic printing, CC1 scanning, and digital photography. 7 have collaborated with a number of leading companies in most fields of color reproduction, including Microsoft, 6gfa, Beidelberg4Cinotype4Bell, and 6dobe. 7 have also been able to again
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7 #
serve as the chairman of the 7CC to co4author a growing standard with s&:@, 7?C $1($$4 41. 1uring the following two years, 7 worked in Bewlett4<ackard5s CaserOet 1ivision, coordinating color management and color standards issues across the company. 7 initiated the development of the sc&:@ standards, 7?C $1($$4 4 C1, and co4developed Microsoft !"L Color "uality #pecifications for $rinter %&Ms and other color test kits. +ince ,ctober 1(((, 75ve worked at Microsoft Corporation as the corporate Color 6rchitect, and as <rogram Manager in their advanced graphics section. 7 have led efforts to continue development in color standards, and expanded the 3BRC color specifications to C&Ts, CC1s, and other devices. Throughout these years, 7 have had the wonderful opportunity to learn color reproduction from the world5s leading experts. 7 will be forever in their debt for providing me with this ongoing education and life experience. My continuing struggles to reproduce a particular burgundy or cerulean have been a delight and frustration in my life. 3ith this paper, 7 hope to share some of what 75ve learned along the way in the fascinating field of digital color reproduction.
C 4 3 4 &
M A 9 A : ! M ! 9 1
C 4 9 C ! * 1 S
7 6
Endnotes
) 3 , # 6 ' ( 9 7
) 3 ,
# 6
' (
9 )7 )
))
)3 ),
)#
)6
)'
)( )9 37 3 3) 33 3,
3# 36 3' 3( 39
,7
, ,)
Margulis< $an< Ma%eready& A 'repress (eso)rce< MIS:*ress< 9ew Lor.< 996 Lule< L.A.C.< *rinci0les o% Color &e0roduction< Kohn -iley J Sons< 96' !"ans< &.M.< An Introduction to Color< Kohn -iley J Sons< 9ew Lor.< 9,( Hunt< &. -. :.< *he (eprod)ction of Colo)r in 'hotography 'rinting + *elevision< Fountain *ress< Hert%ordshire< 99# Hall< &oy< Ill),ination and Color in Co,p)ter !enerated I,agery< S0ringerHVerlag< 8erlin< 9(9 Mac$onald< 3indsay< Com0uter :enerated Color: A *ractical :uide to *resentation and $is0lay< Kohn -iley< 99, *oynton< Charles< A 1echnical Introduction to $igital Video< Kohn -iley< 996 Fairchild< Mar. $.< Color Appearance Models< AddisonH-esley< &eading< 99( 9ic.erson< $.< 4SA uni%orm color scales sam0les M A uni+ue set< Color &es. A00l< 6< 'H33 C 9( D 8erns< &. S.< 8illmeyer and Salt5manEs *rinci0les o% Color 1echnology< Kohn -iley J Sons< 9ew Lor.< )777 =ing< K. C.< 1he Color !ngineer< <N 0roceedings o% the First Color Imaging Con%erence< ISJ16SI$< Scottsdale< 993< 0.99H 77 8oynton< &.M.< Human Color Vision< 40tical Society o% America< 99) Cornsweet< 1.9.< Visual *erce0tion< 9ew Lor.< Academic *ress< 9'7 Hunter< &ichard S. and Harold< &ichard -.< *he Meas)re,ent of Appearance< )nd !dition< Kohn -iley J Sons< 9ew Lor.< 9(' Hunt< &.-.:.< Measuring Colour< Fountain *ress< !ngland 99( 8erns< &.S.< 1he Im0ortance o% Color A00earance Models in Color Management Systems< *roceedings o% the (th Congress o% the International Colour Association< Volume < 99'< 0. 7 CI!< OColorimetry<N )nd ed. CI! *u2l. #.)< CI!< 9(6< htt0:66www.cie.co.at6cie6home.html Kones< 3.A. and Condit< H.&.< 1he 8rightness Scale o% !Aterior Scenes and the Com0utation o% Correct *hotogra0hic !A0osure< K. 40t. Soc. Am.< 3 < 6# H6'( C 9, D MacAdam< $.3.< /uality o% Color &e0roduction< K. Soc. Mot. *ic. 1el. !ng.< #6< ,('H# ) C 9# D :iorgianni< !dward K. and Madden< 1homas !.< -igital Color Manage,ent< AddisonH-esley< &eading< 99( 8reneman< !.K.< 1he !%%ect o% 3e"el o% Illuminance and &elati"e Surround 3uminance on the A00earance o% 8lac.HandH -hite *hotogra0hs< *ho. Sci. !ng.< 6< ')H '9 C 96)D Hunt< &.-.:< O-hy is 8lac.HandH-hite so Im0ortant in ColorPN< *roceedings o% the Fourth Color Imaging Con%erence< ISJ16SI$< Scottsdale< 996< 0.#, 91SC< *anel )< 1he 91SC Color 1ele"ision Standards< *roc. I.&.!.< ,6H,( C 9#,D Motta< &.< OAn Analytical Model %or the Colorimetric Characteri5ation o% Color C&1s<N &ochester Institute o% 1echnology< 99 8erns< &.S.< :or5yns.i< M.!.< and Motta< &.K.< C&1 Colorimetry. *art II: Metrology< Color &es. A00l.< "ol. (< #< 0 3 # C 993D OIn%ormation 1echnology M $igital Com0ression and coding o% continuousHtone still images< *art : &e+uirements and guidelines<N I1?H1 ( < IS46I!C 79 (H < I1?< :ene"a< 993 $is0oto< :.< Sto.es< M.< 3imitations in Communicating Color A00earance with the ICC *ro%ile Format< *roceedings o% the 1hird Color Imaging Con%erence< ISJ16SI$< Scottsdale< 99#< 0. ## Qhang< Farrell< J -andell C 99'D A00lication o% a s0atial eAtension to CI!3A8< ISJ16S*I! !lectronic Imaging 9'. Fairchild< Mar.< CI! 1C H3, dra%t re0ort< htt0:66www.cis.rit.edu60eo0le6%aculty6%airchild6CAM.html A9SI6C:A1S< O:ra0hic 1echnology M 1hree Com0onent Color $ata $e%initions<N A9SI6C:A1S< 996 Sto.es< M.< Colorimetric 1olerances o% $igital Images< &I1 C 99 D Kones< 3.A.< *hotogra0hic &e0roduction o% 1one< K. 40t. Soc. Am.< #< )3)H)#( C 9) D Lule< L.A.C.< *rinci0les o% Color &e0roduction< Kohn -iley J Sons< 9ew Lor. C 96'D ?lichney< &.< RA &e"iew o% Hal%toning 1echni+uesR< Color Imaging: $e"iceHInde0endent Color< Color Hardco0y< and :ra0hic Arts V< *roc. S*I! "ol. 3963< Kan.< )777. Sey2old Seminars< htt0:66www.sey2oldseminars.com69ews6s%9'6Ko2sS=eynote.html< San Francisco< 99' Sey2old Seminars< htt0:66www.sey2oldseminars.com69ews6s%9'6:atesS=eynote.html< San Francisco< 99' Search on htt0:66www.search.com6In%osee.6 < 3#<7<7)77.htmlP/?!&LTU))colorVmanagementU))JC433T-A00le Com0uter< htt0:66www.a00le.com60r6li2rary6 99(6mar6 'ado2e.html< Cu0ertino< 99( htt0:66www..oda..com6cgiH2in6we2Catalog.0lP sectionTJccT?SJlcTenJ0roductT=4$A=V$igitalVScienceVColorVManagementVSolutions =ieren< Michael< htt0:66www.intero0.s2%orums.com60lanner6ignoreP MI"alTsessionsS2yidJ!"entTSS9L9(JSu0erty0eT!ducationalV*rogramsJI1!MSC3ASSTSessionSidJI1!MSC3ASSSVA 3?!TS737)JsearchSgrou0TSessions< Sey2old Seminars< 99( Fraser< 8ruce< htt0:66www.5dnet.com65dnn6content6macw6 376macw77)#.html< Mac-ee.< 996 =ieren< Michael< htt0:66www.sey2oldseminars.com6!"ents6s%9'61ranscri0ts63#.html< Sey2old Seminars< 99'
,3
,, ,# ,6
,' ,(
,9 #7
# #) #3
#, ##
#6
#' #( #9 67
6) 63 6,
Michael Sto.es< O1he History o% the ICC<N 0roceedings o% the Fi%th Color Imaging Con%erence< ISJ16SI$< Scottsdale< 99'< 0.)66 Fairchild< Mar.< CI! 1C H3, dra%t re0ort< htt0:66www.cis.rit.edu60eo0le6%aculty6%airchild6CAM.html Sto.es< Michael< htt0:66www.srg2.com6FA/s&:8'796s&:8'79com0ati2ility.htm =atoh< 9aoya and $eguchi< 1atsuya< O&econsideration o% C&1 Monitor Characteritics<N 0roceedings o% the Fi%th Color Imaging Con%erence< ISJ16SI$< Scottsdale< 99'< 0.33 SM*1! &* ,#H 99,< OSM*1! C Color Monitor Colorimetry<N SM*1!< 99,. I!C 6 966H)H !d. .7: 999< Multimedia systems and e+ui0ment M Colour measurement and management M *art )H : Colour management M $e%ault &:8 colour s0ace M s&:8 C 999H 7D. !8? 1ech. 3) 3< OChromaticity 1olerances %or Studio Monitors<N 9( 91SC 9#3< OAn Analytical Model %or the Colorimetric Characteri5ation o% Color C&1s<N Motta< &icardo< &ochester Institute o% 1echnology< 99 < 099 O1he Virtual &eality Modeling 3anguage S0eci%ication< Version ).7<N V&M3 Consortium< August 996< htt0:66www."rml.org Microso%t< Inc.< 0ri"ate communication with Matthew Anderson< May 99' A00le Com0uter< Inc.< 2ased on A00leEs de%ault system ICC 0ro%ile o% a 3 inch 1rinitron monitor< this 0ro%ile shi0s with ColorSync< htt0:66www.a00le.com Silicon :ra0hics< Inc.< 0ri"ate communication with 1odd 9ewman< August 99' IS4 1C,)661C 37< O:ra0hic 1echnology and *hotogra0hy M Standard monitor &:8 s0eci%ications<N IS4< :ene"a< A0ril 99' =atoh< 9aoya and $eguchi< 1atsuya< O&econsideration o% C&1 Monitor Characteristics<N 0roceedings o% the Fi%th Color Imaging Con%erence< ISJ16SI$< Scottsdale< 99'< 0.33 A9SI6C:A1S< O:ra0hic 1echnology M 1hree Com0onent Color $ata $e%initions<N A9SI6C:A1S< 996 3am< =.M.< OMetamerism and colour constancy<N ?ni"ersity o% 8rad%ord ?ni"ersity C8F$D< 9(# I1?H& 81.67 < htt0:66www.itu.ch I1?H& 81.'796)< O8asic *arameter Values %or the H$1V Standard %or Studio and %or International *rogramme !Achange<N I1?< :ene"a< 997< htt0:66www.itu.ch Smith< V.C.< and *or.orny< K.< OS0ectral Sensiti"ity o% the Fo"eal Cone *hoto0igments 2etween ,77 and #77 nm<N Vision &es.< #< 9'# CI!< OColorimetry<N )nd ed. CI! *u2l. #.)< CI!< 9(6< htt0:66www.cie.co.at6cie6home.html Lule< L.A.C.< *rinci0les o% Color &e0roduction< Kohn -iley J Sons< 96'< 0 6 Mc$owell< $a"id< O:ra0hics arts color standards u0date M 99(<N *roceedings o% S*I!< Color Imaging< Volume 3377< 0. #6< 99(

Color MGMT Concepts

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Color MGMT Concepts

Uploaded by

Copyright:

Available Formats

Color Management Concepts

Michael Stokes Microsoft Corporation Draft v. 0.7, March 5, 2001 Contents

WHO IS THIS PAPER FOR?

INTRODUCTION TO DIGITAL COLOR REPRODUCTION

Color Science: Models and Algorithms

HUMAN VISUAL SYSTEM

VIEWING ENVIRONMENT MODELS

STANDARD REFERENCE MEDIA DESCRIPTIONS

INTRODUCTION 5 -EFORE COLORSYNC

THE COLORSYNC CONSORTIUM

ICC GOALS POLICY

CURRENT ICC STATUS

OPEN ISSUES WITHIN

PROFILE MANAGEMENT METHODS

COLOR MANAGEMENT METHODS

COLOR CALI-RATION UTILITIES

OPI7 FPO UTILITIES

CAPTURE DEVICE DRIVERS

DISPLAY DEVICE DRIVERS

PRINTER DEVICE DRIVERS

COLOR IMAGING WOR8FLOWS

Fadd in %uture dra%t: o0tical test 0attern imageG

COLOR FILTERING ARTIFACTS

SPATIAL PARTITIONING ARTIFACTS

DIGITAL 6UANITI,ATION ARTIFACTS

EDITING PAST ARTIFACTS

EDITING CURRENT ARTIFACTS

EDITING FUTURE ARTIFACTS

COLOR SPACE CONVERSION

PRINT DEVICE1SPECIFIC ARTIFACTS

ESTA-LISHING TEST METHODS

COMMON IMAGE CHARACTERISTICS

add in %uture dra%t: a good and a 2ad eAam0le 0ictureG

Fadd in %uture dra%t: a good and a 2ad eAam0le 0ictureG

COGNITIVE COLOR CONSTANCY

Fadd in %uture dra%t: a good and a 2ad eAam0le 0ictureG

COLOR 6UALITY METRICS

COLOR -ALANCE ENHANCEMENT

CHANNEL GENERATION ALGORITHMS

EVALUATING IMAGE 6UALITY

IMAGE 6UALITY LIMITATIONS

IEC Color Standards

POSSI-LE ARCHITECTURE SYSTEMS

+ig)re 5. Mo#el of an open color co!!)nication s*ste!

HOW COLOR MANAGEMENT -E REALI,ED

+ig)re 4. Color !anage!ent sche!e for color i!age pro#)ction

S$)#;)r; RGC l ur !')"e ? T 3 , & L

1isplaysF<rinters 1ata4to4light ob8ect conversion

+ig)re 7. Color !anage!ent sche!e for color i!age repro#)ction

display gamma = CRT gamma x LUT gamma

and k2 are the system gain and offset

is the normalized pixel value is the maximum luminance of the C&T

is the resulting luminance

A'out the Author

You might also like