You are on page 1of 16

Machine Visi n

Lecture # 1: Introduction

Muhammad Rzi Abbas Department of Mechatronics and Control Engineering


muhammadrziabbas@uet.edu.pk
Lecturer, Mechatronics Dept.
University of Engineering and Technology, Lahore
Vision
• Five Senses: vision, hearing, smell, taste, and touch
• Sense of Vision provides most of the data the human brain works on
• The speed of interpretation is enormous
• We approximately need tenth of a second to understand and interpret a
scene
• ‘Optical illusions’ sometimes challenges our vision system
Vision
• Human Vision System is not a simple system, it just has evolved over
millions of years.
• Humans wished that machines should do their work, but capability of
vision limited the machines to certain tasks only.
• Hence Machine Vision was thought of and worked on.
• The more the researchers worked on vision systems, the more they
realized that how far they are from Human Vision System.
Vision
• Human brain processes 1010 cells (or neurons), where each neuron is
acting like a microcontroller and just to add more difficulty, this
processing is happening concurrently.
• No human made computer can handle such amount of processing
Machine Vision

• “It is the study of methods and techniques whereby artificial vision


systems can be constructed and usefully employed in practical
applications”
Machine Vision

One picture is worth more than ten thousand words


(Anonymous)

Do you think colors exist in this world? What is the


reality of colors that we see?
Human (Biological) vs Machine Vision
• Qualitative vs Quantitative
• People can rely on inference systems and assumptions for
comprehension
• Computers do not ‘see’ in the same way that human beings are able
to
• Examine individual pixels of images, processing them and attempting
to develop conclusions in a consistent manner
• Spectral limitations of human vision
• Passive vs Active (we rely on external illumination)
Machine Vision System
• Why Machine Vision System is a difficult thing?
• Consider an example of Character Recognition
Machine Vision System
• Test Examples

• A simple vision algorithm can identify the test examples even though
there is a certain amount of noise present
Machine Vision System
• There are two kinds of situations:
• Noise and/or distortions are present, hence misclassification occurs
• Noise and/or distortions are not present and still misinterpretation occurs
• The later is more dangerous situation since it indicates an unexpected
limitation of the technique
• Such situations may arise due to disorientation or displacement of the test
examples relative to the training set.
• This will lead us to another serious problem of insufficient training set
examples.
Machine Vision System
• Increasing the training is an obvious solution but this solution is
accompanied with further problems
• Problem of storage space
• Problem of generation of more training examples
• Problem of searching when examples are tested
• These problems are further exacerbated as we move closer to real-
life situation
• This is called ‘Combinatorial Explosion’
Machine Vision System
• Just calculate the possible binary patterns which can be formed using
a 20x20 grid.
• Hence such techniques are not only impractical but also theoretically
impossible.

We need some other algorithms to tackle this


recognition problem.
Machine Vision System
• Impose few realizable limitations
• Fix the center of an object using major axis
• Use the knowledge that no pattern can be made up of random disconnected
locations
• All characters are made of roughly constant width, hence no information is
present in the width property
Machine Vision System
• Sometimes making generalizations is the only goal, whereas at
others, discrimination is required.
• For example: An ‘A’ is generalized as ‘A’ whether it is computer typed or hand
written, bold or italic, large sized or small sized etc.
• In some cases discrimination becomes important because ‘n’ is not ‘h’.
• So certain generalizations have to be removed before going on to
recognition.
• At early stages of Machine Vision, researchers divided the Machine
Vision problem into a two step problem of ‘Preprocessing’ and
‘Recognition’.
Machine Vision System
• Preprocessing was referred to as tackling all generalizations
• Recognition was done using Statistical Pattern Recognition
techniques
• Intermediate techniques, such as Hough Transforms, were never
explored.
References
• Image Processing, Analysis and Machine Vision by Milan Sonka,
Vaclav Hlavac and Roger Boyle, 3rd Edition, 2008.
• Chapter 1 (Sections 1.2 and 1.3)
• Machine Vision by David Vernon, Published in 1991
• Chapter 1 (Section 1.1)
• Computer and Machine Vision – Theory, Algorithms, Practicalities by
E.R.Davies, 4th Edition ELSEVIER, 2012
• Chapter 1 (Sections 1.1, 1.2.1 and 1.2.2)

You might also like