CHAPTER 1

INTRODUCTION

1
1.1 GENERAL REMARKS

Retinal image provides a snapshot of what is happening inside the human

body. The state of retinal vessel has been shown to reflect the cardiovascular
conditions of the body. The wrong identification of vessels may result in a large
variation of these measurements, leading to a wrong clinical diagnosis.
Examination of blood vessels in the eye allows detection of eye diseases such as
glaucoma and diabetic retinopathy. Diabetes is a disease that occurs when the
pancreas does not secrete enough insulin or the body is unable to process it
properly. The affects of diabetes on the eye is called diabetic retinopathy. It affects
the circulatory system of retina.
1.2 DIABETIC RETINOPATHY

Fig.1.1 Retinal image shows diabetic retinopathy

Diabetic retinopathy, is retinopathy (damage to the retina) caused by

complications of diabetes, which can eventually lead to blindness. It is an ocular
manifestation of diabetes, a systemic disease, which affects up to 80 percent of all
patients who have had diabetes for 10 years or more. Despite these intimidating
statistics, research indicates that at least 90% of these new cases could be reduced
if there was proper and vigilant treatment and monitoring of the eyes. The longer a
person has diabetes, the higher his or her chances of developing diabetic
retinopathy. Diabetic retinopathy often has no early warning signs.

2
1.3 RISK FACTORS

All people with diabetes mellitus (Type I diabetes and Type II diabetes) are
at risk. The longer a person has diabetes, the higher the risk of developing some
ocular problem. After 20 years of diabetes, nearly all patients with Type I diabetes
and greater than 60% of patients with Type II diabetes have some degree of
retinopathy. It has been shown that the widely accepted WHO and American
Diabetes Association diagnostic cutoff for diabetes of a fasting plasma glucose
7.0 mmol/l (126 mg/dl) does not accurately identify diabetic retinopathy among
patients.

1.4 TYPES OF DR

1st Phase (BDR) Background diabetic retinopathy: In this case arteries in

the retina become weakened & leak forming small dot like haemorrhages.
These leaking vessels lead to swelling or oedema in the retina and decrease
vision.
2nd Phase (PDR) Proliferative diabetic retinopathy: In this case,
circulation problems cause in the areas of retina become ischemic. New
blood vessels developed to maintain adequate oxygen level with in the
retina, it is called neovascularisation. These vessels bleed easily and blood
leak in to the retina and vitreous causing spots or floaters along with
decreased vision.
1.5 SYMPTOMS OF DR

Diabetic retinopathy often has no early warning signs.

Blurred vision

Vision get worse during the day

3
 Macular edema

Retinal ischemia (lack of blood flow)

1.6 DIAGNOSIS

Visual acuity test

Pupil dilation

Ophthalmoscopy or fundus photography

Fundus Fluorescein angiography (FFA)

Optical coherence tomography (OCT)

Digital Retinal Screening Programs

Slit Lamp Bio-microscopy Retinal Screening Programs

1.7 DETECTION OF DR

Diabetic retinopathy is the cause for blindness in the human society.

Diabetic retinopathy is a highly specific vascular complication and the prevalence
of retinopathy is strongly related to the duration of diabetes. It is beneficial to have
regular cost-effective eye screening for diabetes. Early detection of it prevents
blindness. Image processing techniques can reduce the work of ophthalmologists
and the tools used automatically locate the exudates. Early detection helps the
patients to aware of the seriousness of the disease. Detailed visualization of micro-
vascular changes in the human retina is clinically limited. An automated screening

4
system can analyze digital color retinal images for important features of diabetic
retinopathy. These are used to detect the landmarks of retinal image such as optic
disc, blood vessels etc.

CHAPTER 2
LITERATURE SURVEY

5
1.Yannis A. Tolias, and Stavros M. Panas, A Fuzzy Vessel Tracking
Algorithm for Retinal Images Based on Fuzzy Clustering IEEE
TRANSACTIONS ON MEDICAL IMAGING, APRIL 1998
This paper present a new unsupervised fuzzy algorithm for vessel tracking
that is applied to the detection of the ocular fundus vessels. This method
overcomes the problems of initialization and vessel profile modeling that are
encountered in the literature and automatically tracks fundus vessels using
linguistic descriptions like vessel and nonvessel. The main tool for
determining vessel and nonvessel regions along a vessel profile is the fuzzy C-
means clustering algorithm that is fed with properly preprocessed data. Additional
procedures for checking the validity of the detected vessels and handling junctions
and forks are also presented. This paper presented a novel scheme that
automatically tracks vessels in fundus images without the need of user
intervention.
2.M. Elena Martinez-Perez, Alun D. Hughes, Alice V. Stanton, Simon A.
Thom, Neil Chapman, Anil A. Bharath, and Kim H. Parker Retinal
Vascular Tree Morphology: A Semi-Automatic Quantification, IEEE
TRANSACTIONS ON BIOMEDICAL ENGINEERING, AUGUST
2002.

6
 In this paper a semi-automatic method to measure and quantify
geometrical and topological properties of continuous vascular trees in clinical
fundus images is described. Measurements are made from binary images obtained
from segmentation process. The skeletons of the segmented trees are produced by
thinning off branch and crossing points are identified and segments of the trees are
labeled and stored as a chain code. The operator selects a tree to be measured and
decides if it is an arterial or venous tree. An automatic process then measures the
lengths, areas and angles of the individual segments of the tree. Geometrical data
and the connectivity information between branches from continuous retinal vessel
trees are tabulated. A number of geometrical properties and topological indexes are
derived.

3. Enrico Grisan, Alessandro Pesce, Alfredo Giani1, Marco Foracchia,

and Alfredo Ruggeri , A new tracking system for the robust
extraction of retinal vessel structure, Proceedings of the 26th Annual
International Conference of the IEEE EMBS San Francisco, CA, USA,
September2004.
Identification and measurement of blood vessels in retinal images could
allow quantitative evaluation of clinical features, which may allow early diagnosis
and effective monitoring of therapies in retinopathy. A new system is proposed for
the automatic extraction of the vascular structure in retinal images, based on a
sparse tracking technique. After processing pixels on a grid of rows and columns to
determine a set of starting points (seeds), the tracking procedure starts. It moves
along the vessel by analyzing subsequent vessel cross sections (lines perpendicular
to the vessel direction), and extracting the vessel center, caliber and direction.
Vessel points in a cross section are found by means of a fuzzy c-means classifier.

7
The system proposed here is the first that integrates a reliable tracking technique
with bifurcations and crossing identification, showing the possibility of devising a
fully automatic system for vascular morphology and lesion analysis in retinal
images.

4.Yi Yin, Mouloud Adel, Mireille Guillaume, and Salah Bourennane, A

PROBABILISTIC BASED METHOD FOR TRACKING VESSELS
IN RETINAL IMAGES, IEEE 17th International Conference on
Image Processing, September 2010.
Vessel detection is an important process in many medical imaging
applications. In this paper, an edge tracking scheme is proposed for the detection of
blood vessels in retinal images. This method detects edge points iteratively based
on a Bayesian approach using local grey levels statistics and continuity properties
of blood vessels. Combining the grey level profile and vessel geometric properties
improves the accuracy and robustness of the tracking process. Experiments on both
synthetic and real retinal images show promising results. The difficulty in this
method is to calculate the a priori probability of different configurations. So, a
more advanced a priori model should be adopted to improve the performance of
this method. Besides, a deeper evaluation on retinal images is needed to make the
proposed method widely usable for vessel detection technique.

5.Saurabh Garg, Jayanthi Sivaswamy, Siva Chandra,

UNSUPERVISED CURVATURE-BASED RETINAL VESSEL

8
SEGMENTATION, CVIT, International Institute of Information
Technology, Hyderabad, India, 2007.
Unsupervised methods for automatic vessel segmentation from retinal
images are attractive when only small datasets, with associated ground truth
markings, are available. This paper presents an unsupervised, curvature-based
method for segmenting the complete vessel tree from color retinal images. The
vessels are modeled as trenches and the medial lines of the trenches are extracted
using the curvature information derived from a novel curvature estimate. The
complete vessel structure is then extracted using a modified region growing
method. Obtaining large datasets with ground truth is essential in developing
robust solutions for vessel segmentation that can be used in mass programs. It is
difficult to achieve as the ground truth generation is a tedious process that demands
patience. This signals the challenges in creating consistent quality and large size
datasets for testing and benchmarking segmentation algorithms. Unsupervised
techniques for segmentation are hence quite attractive in this scenario. In
registration applications, an added strength of this method is it can be simplified by
removing the segmentation stage as the medial axis map provides sufficient corner
(control) points.

6. Hind Azegrouz, Emanuele Trucco, MAX-MIN CENTRAL VEIN

DETECTION IN RETINAL FUNDUS IMAGES, School of
Engineering and Physical Sciences Department of Electrical Electronic
and Computer engineering, Heriot Watt University,2006.
This paper describes a new framework for the automated tracking of the
central retinal vein in retinal images. The procedure first computes a binary image

9
of the retinal vasculature, and then obtains the skeleton (medial axis) of the
vascular network. Terminal and branching points of the network are then located,
and the network converted into a graph representation including length and
thickness information for all vessels. Finally, a MaxMin approach is used to locate
the central vein: The candidates central vein is the minimal paths from the optic
disk to all terminal nodes found using Dijkstra algorithm. The actual central vein is
selected among the all candidates by maximizing a merit function estimating the
total vessel area in the image. This paper presents a new automatic method to track
the central vein from fundus retinal images. The method uses connectivity and
anatomy information to determine the minimal path in the vessel graph that
maximizes a cost function. For the cost function described here used only length,
connectivity and thickness of the graph arcs.

7.V.Vijaya Kumari, N.SuriyaNarayanan, Diabetic Retinopathy-Early

Detection Using Image Processing Techniques, (IJCSE) International
Journal on Computer Science and Engineering, Vol. 02, No. 02, 2010,
357-361.
Diabetic retinopathy is the cause for blindness in the human society.
Early detection of it prevents blindness. Image processing techniques can reduce
the work of ophthalmologists and the tools used automatically locate the exudates.
Early detection helps the patients to aware of the seriousness of the disease. This
paper presents a method which is automatic and involves two steps: optic disk
detection and exudates detection. The extraction of optic disk is done using
propagation through radii method. Exudates detection is done using feature
extraction, template matching and enhanced MDD classifiers and the methods are
compared. The position of optic disk is determined using the Principal Component

10
Analysis method with certain modifications which makes the method to be fast and
robust. This is useful when ophthalmologist visits an eye camp, he can get the
retinal images of many patients using fundus camera and once these images are fed
to system, the abnormalities can be easily detected. This reduces the analysis time
and improves the efficiency.

8.Neera Singh, Ramesh Chandra Tripathi, Automated Early Detection

of Diabetic Retinopathy Using Image Analysis Techniques ,
International Journal of Computer Applications (0975 8887), Volume
8 No.2, October 2010.
Diabetic retinopathy (DR) is a common retinal complication associated
with diabetes. It is a major cause of blindness in middle as well as older age
groups. Therefore early detection through regular screening and timely intervention
will be highly beneficial in effectively controlling the progress of the disease.
Since the ratio of people afflicted with the disease to the number of eye specialist
who can screen these patients is very high, there is a need of automated diagnostic
system for diabetic retinopathy changes in the eye so that only diseased persons
can be referred to the specialist for further intervention and treatment. The spatial
distribution of exudates and microaneurysms and hemorrhages, especially in
relation to the fovea can be used to determine the severity of diabetic retinopathy.
Image analysis tools can be used for automated detection of these various features
and stages of Diabetes Retinopathy and can be referred to the specialist
accordingly for intervention, thus making it a very effective tool for effective
screening of Diabetic Retinopathy patients

11
9.Huiqi Li, Wynne Hsu, Mong Li Lee*, and Tien Yin Wong,
Automatic Grading of Retinal Vessel Caliber, IEEE
TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO.
7, JULY 2005.
New clinical studies suggest that narrowing of the retinal blood vessels
may be an early indicator of cardiovascular diseases. One measure to quantify the
severity of retinal arteriolar narrowing is the arteriolar-to-venular diameter ratio
(AVR). The manual computation of AVR is a tedious process involving repeated
measurements of the diameters of all arterioles and venules in the retinal images by
human graders. Consistency and reproducibility are concerns. To facilitate large-
scale clinical use in the general population, it is essential to have a precise, efficient
and automatic system to compute this AVR. In this study the potential of wavelet
analysis for amplitude estimation of perfusion signals in vitro and in vivo was
explored. It was found that FFT and wavelet analysis gave very similar results on
periodic perfusion signals present in the in vitro study.

10.Akara Sopharak, Bunyarit Uyyanonvara, Automatic exudates

detection from diabetic retinopathy retinal images using Fuzzy C-
means and morphological methods,
Exudates are the primary signs of diabetic retinopathy which are mainly
cause of blindness and could be prevented with an early screening process. Pupil
dilation is required in the normal screening process but this affects patients vision.
This paper investigated and proposed automatic methods of exudates detection on
low-contrast images taken from non-dilated pupils. The process has two main
segmentation steps which are coarse segmentation using Fuzzy C-Means clustering
12
and fine segmentation using morphological reconstruction. Four features, namely
intensity, standard deviation on intensity, hue and adapted edge, were selected for
coarse segmentation. The detection results are validated by comparing with expert
ophthalmologists hand-drawn ground-truth.

CHAPTER 3

IMAGE PROCESSING

13
3.1 DIGITAL IMAGE PROCESSING

An image may be defined as a two dimensional function, f(x, y), where x

and y are spatial (plane) coordinates, and the amplitude of f at any pair of
coordinates(x, y) is called the intensity or gray level of the image at that point.
When x, y, and the amplitude values of f are all finite, discrete quantities is call the
image a digital image.

3.2 FUNDAMENTAL STEPS IN IMAGE PROCESSING

14
 Wavelets and Multi Resolution Processing

Color Image Processing Compression Morphological

Processing

Image Restoration

Segmentation

Image Enhancement
Knowledge Base

Representation & description

Image
Acquisition
Object Recognition

Figure 3.1 Fundamental steps in image processing

3.2.1 Image acquisition:

The image acquisition stage involves preprocessing, such as scaling. It is

used to acquire a digital image. The first process gave some hints regarding the
origin of digital images.

3.2.2 Image restoration:

15
 It is an area that deals with improving the appearance of an image.
Image enhancement is subjective; image restoration is objective, in the sense that
restoration techniques tend to be based on mathematical or probabilistic models of
image degradation.

3.2.3 Color image processing:

It is an area that has been gaining in importance because of the
significant increase in the use of digital images over the Internet.

3.2.4 Compression:
It deals with techniques for reducing the storage required to save an
image, or the bandwidth required to transmit it. Though storage technology has
improved significantly over the past decade, the same cannot be said for
transmission capacity. This is true particularly in uses of the internet, which are
characterized by significant pictorial content. Image compression familiar uses
of computers in the form of image file extensions such as jpg file extension
used in the JPEG (joint photographic experts group) compression standard.

3.2.5 Segmentation:
It procedures partition an image into its constituent parts or objects. In
general, autonomous segmentation is one of the most difficult tasks in digital
image processing. A rugged segmentation procedure brings the process a long
way toward successful solution of imaging problems that required objects to be
identified individually.

3.2.6 Representation and description:

It follows the output of a segmentation stage, which usually is raw pixel
data, constituting either the boundary of a region (i.e., the set of pixels
separating one image region from another) or all the points in the region itself.

16
 Description, also called feature selection, deals with extracting attributes that
result in some quantitative information of interest or are basic for differentiating
one class of objects from another.

3.2.7 Recognition:
It is the process that assigns a label (e.g., vehicle) to an object based on
its descriptors.

3.2.8 Knowledge base:

Knowledge about a problem domain is coded into an image processing
system in the form of a knowledge database. This knowledge may be as simple
as detailing regions of an image where the information of interest is known to
be located, thus limiting the search that has to be conducted in seeking that
information. Knowledge about a problem domain is coded into an image
processing system in the form of a knowledge database.

3.2.9 Morphological processing:

It deals with tools for extracting image components hat are useful in
the representation and description of shape. The identification of objects within an
image can be a very difficult task. One way to simplify the problem is to change
the grayscale image into a binary image, in which each pixel is restricted to a value
of either 0 or 1. The techniques used on these binary images go by such names as:
blob analysis, connectivity analysis, and morphological image processing (from the
Greek word morphe, meaning shape or form). The foundation of morphological
processing is in the mathematically rigorous field of set theory; however, this level
of sophistication is seldom needed.

3.2.10 Image Enhancement:

17
 Image Enhancement is among the simplest and most appealing areas of
digital image processing. Basically, the idea behind enhancement techniques is
to bring out detail that is obscured, or simply to highlight certain features of
interest in an image. A familiar example of enhancement is when its increase
the contrast of an image because it looks better.

3.2.11 Image restoration:

Image Restoration is an area that also deals with improving the
appearance of an image. However, unlike enhancement, which is subjective, image
restoration is objective, in the sense that restoration techniques tend to be based on
mathematical or probabilistic models of image degradation. Enhancement, on the
other hand, is based on human subjective preferences regarding what constitutes a
good enhancement result.

3.3 IDENTIFICATION OF FUNDAL LANDMARK

Retinal images taken by a fundus camera are widely used by the medical
community for diagnosing human diseases. Inspection of the retinal vasculature
can reveal hypertension, diabetes, arteriosclerosis, cardiovascular disease and
stroke. Furthermore each person has a unique blood vessels tree and its
segmentation may be useful for the biometric person identification and for medical
registration methods. In all cases, proper blood vessels detection is crucial.
The Diabetic Retinopathy (DR) is a complication of diabetes that is
caused by changes in the blood vessels of retina. The symptoms can distort or blur
the patients vision and are the main causes of blindness. The identification of
fundal landmark features such as optic nerve head, macula and the retinal vessels
as reference co-ordinates, is a pre requisite before systems can achieve more
complex tasks of identifying pathological entities from DR images. Position of
optic nerve head with respect to macula is used to differentiate left eye and right

18
eye. When optic nerve head is swollen and lesions particularly exudates, are very
near to optic nerve head and if the size and brightness of exudates are same as that
of optic nerve head in that case there is possibility of false detection of optic disc.
The location of optic nerve head is important in retinal image analysis, to
locate anatomical components in retinal images, for vessel tracking, as a reference
length for measuring distances in the retinal images and for registering changes
within the optic disc region due to disease.

3.4. FEATURES OF OPTIC NERVE HEAD AND OPTIC CUP

Normal structures of the retina are optic nerve head, macula, and blood
vessels. The small area of the retina where optic nerve leaves the eye is optic nerve
head. This is the bright yellow part in the normal fundus image that can be seen as
round or vertically oval. The optic cup which is embedded in optic nerve head is
the brightest, cup-like area in the optic nerve head. About 0.5cm to the nasal side
of macula all nerve fibers of the retina converge to form the optic nerve.

Figure 3.2: Retinal images of Optic Nerve Head and Optic Cup

3.5. FEATURES OF MACULA

Macula is an oval shaped highly pigmented, located roughly in the
center of the retina, temporal to the optic nerve. It is a small and highly sensitive
part of the retina responsible for detailed central vision. The fovea is at the centre
of the macula. Care has to be taken while giving laser treatment in treating DR
since macula is responsible for central vision. It is located at a distance of

19
approximately twice the diameter of the optic nerve head just below the horizontal
axis of the optic nerve head.

Figure 3.3: Retinal images of Macula

3.6. FEATURES OF BLOOD VESSELS

An important aspect of DR is the microvascular changes that cause
detectable changes in the appearance of the retinal blood vessel. In diabetic
retinopathy, the fine vessels on the surface of the retina become damaged, a
condition that can be treated by laser surgery. The blood vessels have lower
reflectance compared to other retinal surfaces.

Figure 3.4: Retinal images of Blood Vessels

3.7. FEATURES OF EXUDATES

These are one of the main characteristics of DR and can vary in size
from tiny specks to large patches. Hard exudates are bright yellow color with well
defined margins. Weakening of retinal capillary walls and loss of pericyte support,
give rise to microaneurysm formation. Serum lipoproteins leak from
microaneurysms and are deposited as exudates.

20
 Figure 3.5: Retinal images of Exudates

These intra retinal deposits are intimately associated with various retinal
pathologies and detection of exudates is quite important for the early diagnosis of
DR.

21
 CHAPTER 4
PROBLEM DEFINITION AND
METHODOLOGY

4.1. Support Vector Machines (SVM)

In machine learning, support vector machines (SVMs, also support vector

networks) are supervised learning models with associated learning algorithms that
analyze data and recognize patterns, used for classification and regression analysis.
The basic SVM takes a set of input data and predicts, for each given input, which
of two possible classes forms the output, making it a non-probabilistic binary linear
classifier. Given a set of training examples, each marked as belonging to one of

22
two categories, an SVM training algorithm builds a model that assigns new
examples into one category or the other. An SVM model is a representation of the
examples as points in space, mapped so that the examples of the separate
categories are divided by a clear gap that is as wide as possible. New examples are
then mapped into that same space and predicted to belong to a category based on
which side of the gap they fall on. In addition to performing linear classification,
SVMs can efficiently perform a non-linear classification using what is called
the kernel trick, implicitly mapping their inputs into high-dimensional feature
spaces.

SVMs can be used to solve various real world problems:

SVMs are helpful in text and hypertext categorization as their application

can significantly reduce the need for labeled training instances in both the
standard inductive and transductive settings.

Classification of images can also be performed using SVMs. Experimental

results show that SVMs achieve significantly higher search accuracy than
traditional query refinement schemes after just three to four rounds of relevance
feedback.

SVMs are also useful in medical science to classify proteins with up to 90%
of the compounds classified correctly.

Hand-written characters can be recognized using SVMs..

4.2. Contrast Limited Adaptive Histogram Equalization (CLAHE)

Contrast Limited AHE (CLAHE) differs from ordinary adaptive histogram

equalization in its contrast limiting. This feature can also be applied to global

23
histogram equalization, giving rise to contrast limited histogram equalization
(CLHE), which is rarely used in practice. CLAHE was developed to prevent the
over amplification of noise that adaptive histogram equalization can give rise to.
CLAHE operates on small regions in the image, called tiles, rather than the entire
image. Each tile's contrast is enhanced, so that the histogram of the output region
approximately. The neighboring tiles are then combined using bilinear
interpolation to eliminate artificially induced boundaries. The contrast, especially
in homogeneous areas, can be limited to avoid amplifying any noise that might be
present in the image. CLAHE limits the amplification by clipping the histogram at
a predefined value before computing the CDF.

4.3. Canny edge detector

The Canny edge detector is an edge detection operator that uses a multi-
stage algorithm to detect a wide range of edges in images. Canny's aim was to
discover the optimal edge detection algorithm. In this situation, an "optimal" edge
detector means:

Good detection the algorithm should mark as many real edges in the image
as possible.

Good localization edges marked should be as close as possible to the edge

in the real image.

Minimal response a given edge in the image should only be marked once,
and where possible, image noise should not create false edges.

Because the Canny edge detector is susceptible to noise present in raw

unprocessed image data, it uses a filter. An edge in an image may point in a variety

24
of directions, so the Canny algorithm uses four filters to detect horizontal, vertical
and diagonal edges in the blurred image. The canny edge detector first smoothes
the image to eliminate and noise. It then finds the image gradient to highlight
regions with high spatial derivatives. The algorithm then tracks along these regions
and suppresses any pixel that is not at the maximum.

4.4. Crossover in retinal image

i. The proposed algorithm uses Discrete Wavelet Transform to transfer the

spatial domain information to the frequency domain information.
ii. The frequency domain provides the features not only the pixel intensity, but
also it provides phase information and the high frequency (Sudden changes
in the images) information. The phase information and the high and low
frequency information from the image used to calculate the crossover
information from the input image dataset.
iii. The above crossover information and optimal forest (Set of vessel trees) of
the image gives the exact extraction of the blood vessel.
4.5. Vessel tree extraction
The main technical novelty of the object extraction approach presented in
this project is that we formulate segmentation energy over binary variables whose
values only indicate whether the pixel is inside or outside the object of interest. In
contrast to the earlier path-based combinatorial methods this can be seen as a
region-based approach to encoding image segments. In fact, the difference
between path-based and region-based representations of segments on a discrete
graph is analogous to the difference between explicit contour representation, e.g.
snakes, and the implicit level-sets approach.

25
 CHAPTER 5
TESTING

5.1 INTRODUCTION

26
 Nowadays there are many different methods how to segment the retinal
vasculature from the fundus images. For example, blood vessel segmentation using
wavelet transform, region growing or adaptive filtering have been used. However,
due to the unique properties of each technique, a single generally accepted vessel
detection algorithm does not exist. The main goal of presented method is a fast
blood-vessels segmentation based on two-dimensional discrete wavelet transform
(2D DWT). In this paper I will introduce a method which is composed from four
main parts.

i. Preprocessing
ii. Decomposition by 2D DWT
iii. Thresholding procedure

iv. Reconstruction with summation

Figure 5.1 Block Diagram for Blood Vessels Extraction

27
 Figure 5.2 Flow Chart for detection of Vasculature

5.2 MODIFIED HISTOGRAM EQUALIZATION TECHNIQUE

In Preprocessing (image enhancement) - The retinal images has been

taken in a RGB mode by fundus camera Canon CF-60UDi with a digital camera
Canon D20. However, the best vasculature information is in the green channel. So
the first step is to separate this channel to a new image. Before applying the vessel
segmentation algorithm it is necessary to denois the image. This is realized by a
filter based on anisotropic diffusion.The filter iteratively uses diffusion equation in
combination with information about the edges. As a consequence, the homogenic
(but noisy) areas are blurred and the edges are preserved.

Index t denotes the time (iterations).Original RGB retinal image on the left,
G channel separated from the original RGB image in the middle, G channel image
filtered by an anisotropic diffusion filter on the right. Histogram equalization

28
technique increases the dynamic range of the histogram of an image. It assigns the
intensity values of pixels in the input image such that the output image contains a
uniform distribution of intensities. The algorithm developed uses a morphological
operation to smooth the background and, as a result, veins, hemorrhages and micro
aneurysms can be seen clearly.

Two types of structuring elements (SE):

(i) Diamond -shaped,

A diamond-shaped structuring element shown is used to make veins
clearer.

(ii) Disk - shaped are used in this work.

A disk-shaped structuring element shown in is used to

Remove noise from the background.

5.3 DECOMPOSITION BY 2D DWT

In blood vessel extraction the 2D discrete wavelet transform is used based on

a bank of filters, which corresponds to specific type of a wavelet. This algorithm
used the Reversed Bi orthogonal wavelet (RBIO) as denoted in Matlab. The shape
of this wavelet almost correspondents to the shape of blood vessels in the retinal
image. The wavelet transform decompose the image to levels, where each level
represents specific frequency band of the wavelet. Its choose 3 levels 2D DWT,
which is sufficient for detecting the retinal vasculature. Each level is then
decomposed in three directions: vertical, horizontal and diagonal and Bi orthogonal
wavelet: rbio on the left, shape of a blood vessel on the right One decomposition
level of 2D DWT.

29
 Figure 5.3 One decomposition level of 2D DWT

5.3.1 Thresholding

The next step is to threshold within each direction in each level. The main
task of thresholding is to highlight high values of wavelet coefficients which
almost correspond to the blood-vessels and suppress small values which
correspond to noise or unimportant structures in the image. The key parameter in
this process is the choice of the threshold value. A good way how to get this value
is to use the histogram of the image. 88 % of the pixels in the wavelet coefficient
image are noise or unimportant structures and only 12 %belongs to the blood-
vessels (determined as a result of the experiments). The threshold value has been
set to brightness value 30, because 88% of pixels are below this value.

30
 Figure 5.4The Thresholding procedure

To binarize the image, a threshold should be carefully chosen. Too small a

threshold will produce an image that has edges linked together. However, a big
threshold will produce edge segments that form curves. It obtained good results by
setting the threshold at 25% of the gray intensities contained into the image.

5.3.2 Reconstruction with summation

Now it is necessary to reconstruct the final binary image. The first step is to
logically add all thresholded directions in each level (Fig 5.6). So three images
which reflect the segmented blood vessels. Due to the 2D DWT, before adding
those images in a final image, it is necessary to interpolate them to the same size
by bilinear interpolation. However the final image has to be binary. Furthermore
each image includes some noise, which is necessary to remove.

5.4 MORPHOLOGICAL IMAGE PROCESSING

In Exudates Detection Hemorrhages and micro aneurysms contribute to

defects in a retina with diabetic retinopathy. All stages of diabetic retinopathy show
such defects. Therefore, it is important to distinguish them from the noisy
background of the retina image. The algorithm developed uses a morphological
31
operation to smooth the background and, as a result, veins, hemorrhages and micro
aneurysms can be seen clearly. The identification of objects within an image can be
a very difficult task. One way to simplify the problem is to change the grayscale
image into a binary image, in which each pixel is restricted to a value of either 0
or 1. The techniques used on these binary images go by such names as: blob
analysis, connectivity analysis, and morphological image processing. The
foundation of morphological processing is in the mathematically rigorous field of
set theory; however, this level of sophistication is seldom needed. Most
morphological algorithms are simple logic operations and very ad hoc. In other
words, each application requires a custom solution developed by trial-and-error.
This is usually more of an art than a science. A bag of tricks is used rather than
standard algorithms and formal mathematical properties.

The two most common morphological operations erosion and dilation. In

erosion, every object pixel that is touching a background pixel is changed into a
background pixel. In dilation, every background pixel that is touching an object
pixel is changed into an object pixel. Erosion makes the objects smaller, and can
break a single object into multiple objects. Dilation makes the objects larger, and
can merge multiple objects into one.

Morphological operation Exudates appear as bright lesions in retinopathic

images and have sharp edges and high contrast with the background. Most of the
standard edge detectors like Sobel edge detection and Canny edge detection add a
lot of noise and miss out key edges. When used for extracting exudates edges and
hence are not suitable for this application. So boundary detection for exudates
using morphological operations. A binary image is one having only two gray
levels, 0 and 1. It refer to pixels with gray level 0 as background pixels and gray
level 1 pixels as interior pixels. A connected set of interior pixels forms an
32
object in the binary image. Some of these objects will correspond to physical
structures on the microscope slide. Others may be due to artifacts or noise. Image
segmentation is the process of delineating the actual structures in the image, and
binary morphological processing can be very useful in that endeavor.

The structuring elements are determined using prior heuristic knowledge of

all the images. The image undergoes several morphological openings using each of
the structuring elements. A short explanation of the function used and the
mathematics involved is as follows.

The morphological opening operation erodes an image and then dilates the eroded
image using the same structuring element for both operations with a disc SE where
R = 10. As a result, the objects completely destroyed by the erosion are not
recovered. This behavior is the very basis of the filtering properties of the opening
operator. The image structures are selectively filtered out depending on the
selection of the shape and size of SE. This means, all foreground image structures
that do not contain the structuring element are removed by the opening. The shape
and size of SE are set according to image structures to be extracted.

The opened set is the union of all SEs fitting the set:

(5.1)

where B = structuring element, X = set of pixels that make up the image, cB =

opening of set using structuring element B. Alternatively, the opened set can be
represented as

(5.2)

33
 The image is further processed with a disc SE of R = 18, followed by a
diamond SE of R = 3, each time, and hence the image has its intensity adjusted
before the SE morphs the image. This increases the contrast of the output image J
as it spreads pixel intensities more evenly over the intensity range. The perimeter
and area of the features can be easily extracted from these pre-processed images.

5.4.1 Binary erosion and dialation

The basic fitting operation of morphology is the erosion of an image by

a structuring element. Erosion is done by scanning the image with the structuring
element. When the structuring element fits completely inside the object, the probe
position is marked. The erosion result consists of all scanning locations where the
structuring element fits inside the object. The eroded image is usually a shrunken
version of the image, and the shrinking effect is controlled by the structuring
element size and shape. The erosion of set A by set B is defined by

(5.3)

A binary image consists of foreground and background pixels. In

morphology, for every operator that changes the foreground in a particular way,
there is a dual operator that changes the background in the same way. The
complement of an image is formed by reversing the foreground and background
pixels. The dual of the erosion operator is the dilation operator.

34
 Figure 5.5: Morphological Operations are used in the Processing of Binary
Images.

Figures (b) and (c) show how the image is changed by the two most
common morphological operations, erosion and dilation. In erosion, every object
pixel that is touching a background pixel is changed into a background pixel. In
dilation, every background pixel that is touching an object pixel is changed into an
object pixel. Erosion makes the objects smaller, and can break a single object into
multiple objects. Dilation makes the objects larger, and can merge multiple objects
into one.

As shown in (d), opening is defined as erosion followed by a dilation. Figure

(e) shows the opposite operation of closing, defined as dilation followed by an
erosion. As illustrated by these examples, opening removes small islands and thin
filaments of object pixels. Likewise, closing removes islands and thin filaments of
background pixels. These techniques are useful for handling noisy images where

35
some pixels have the wrong binary value. For instance, it might be known that an
object cannot contain a "hole", or that the object's border must be smooth.

5.5 ADVANTAGES OF PROPOSED ALGORITHM

5.5.1 Preprocessing (image enhancement)

Modified Histogram Equalization Technique to improve the quality (clarity)

of images for human viewing. Removing blurring and noise, increasing contrast,
and revealing details are examples of enhancement operations. The increased
quality of the image can be proved by calculating the Peak Signal to Noise Ratio as
greater than 23dB.

5.5.2 Blood vessel extraction

The proposed algorithm uses Discrete Wavelet Transform to transfer the

spatial domain information to the frequency domain information.
The frequency domain provides the features not only the pixel intensity, but
also it provides phase information and the high frequency (Sudden changes
in the images) information.
The above features of the image give the exact extraction of the blood
vessel.
5.5.3 Exudates detection

The morphological operation is applied to the entire structure of the

input image. So accurate detection of exudates wherever exudates presents
is possible.

36
CHAPTER 7
RESULT

37
 Fig.7.1. Input Retinal (RGB) image

Pre-processing stage can be regarded as the bedrock of this work.The aim

of pre-processing is to attenuate the noise, to improve the contrast and to correct
the non-uniform illumination.

38
 Fig.7.2. Intensity Gray Image

In digital image processing, images are either indexed images or RGB

(Red, Green, Blue) images. In the RGB images the green channel exhibits the best
contrast between the vessels and background while the red and blue ones tends to
be more noisy. Hence intensity conversion of image is done using green channel,
as the retinal blood vessels appears darker in gray image.

39
 Fig.7.3. Median Filtered Retinal Image

Filtering is used to suppress the unwanted noise which gets added into
the fundus image. Here median filtering is used as it is very robust and has the
capability to filter any outliers and is an excellent choice for removal of salt and
pepper noise.

40
 Fig.7.4. CLAHE Enhanced Retinal Image

Then contrast-limited adaptive histogram equalization (CLAHE) is used

to enhance contrast of small regions in the image. CLAHE operates on small
regions in the image, called tiles, rather than the entire image. Each tile's contrast is
enhanced, so that the histogram of the output region approximately matches the
histogram specified by the 'Distribution' parameter. The neighboring tiles are then
combined using bilinear interpolation to eliminate artificially induced boundaries.
The contrast, especially in homogeneous areas, can be limited to avoid amplifying
any noise that might be present in the image.

41
 Fig.7.5. Image Dilation after Thresholding (image A)

By selecting suitable threshold, histogram equalized image is

thresholded so that only dark regions (including blood vessels) are visible. This
threshold value is selected by trial and error method. Dilation operation is
performed to achieve continuity in detected blood vessels.

42
 Fig.7.6. Canny Edge Detected Image (image B)

By selecting suitable threshold, histogram equalized image is thresholded so

that only dark regions (including blood vessels) are visible. This threshold value is
selected by trial and error method. Dilation operation is performed to achieve
continuity in detected blood vessels. Edge detection is performed on the green
component image using canny filter with suitable threshold.

43
 Fig.7.7. Logical AND between image A and image B

Logical AND operation on two images, one being a threshold image

and another being an edge detected image. The thresholding is done on an adaptive
histogram equalized image. Edge detection is done using canny edge detector.

44
 Fig.7.8. Blood Vessels Detection

The vessel extraction algorithm yields the skeleton of the thicker blood
vessels. These thicker vessels include the main blood vessel and other smaller but
thick vessels emanating from the optic disk.

45
 Fig.7.9. Noisy Blood Vessel

Exudates have the potential of occurring as noise in the vessel

segmentation because closing and subtracting the irregularly shaped exudates using
regular structuring elements is not an exact operation.

46
 Fig.7.10. Red Layer Extraction

The result of the exudates detection algorithm is dilated and subtracted

from the blood vessel result. This removes any noise due to exudates in the
extraction of the main blood vessel. The resulting image after exudates subtraction
is then processed to obtain the optic disk center.

47
 Fig.7.11. Green Layer Extraction

The results of the vessel extraction algorithm on an image having no

exudates. Figure 8.11 showing the entire vessel network. Figure 8.10 having only
the thicker vessels arising from the optic disk.

48
 CHAPTER 8
CONCLUTION AND FUTUR
ENHANCEMENT

8.1 CONCLUTION

In this algorithm investigated and proposed a method based on anatomical

structural details and retinal image information. This system intends to help the
ophthalmologists not only in DR screening process but any other eye related

49
abnormality which is based on retinal photography. It is not a final result
application but it can be a preliminary diagnosis tool or a decision support system
for ophthalmologists. Human ophthalmologists are still needed for the cases where
detection results are not very obvious. This type of presentation will enable
clinicians to identify retinal landmarks more quickly and will also help to take
decision while treating the abnormality, particularly retinopathy.

8.2 FUTURE ENCHANCEMENT

Diabetes affects slowly the circulatory system including that of the retina. So
the vision of a patient may start to deteriorate and lead to diabetic retinopathy. This
work proposes an algorithm for the detection of various stages of diabetic
retinopathy and the degree of blindness from diabetic retinopathy (DR) images of
both left and right eye. This algorithm uses color fundus images obtained from
mydriatic camera. The quantitative performance is evaluated by calculating sensitivity,
specificity and predictive value. Overall sensitivity (Se), specificity (Sp) and predictive
value (PV) obtained in detecting optic nerve head from normal images and from
abnormal images.

50
 CHAPTER 9
REFERENCE

1. Yannis A. Tolias, and Stavros M. Panas, A Fuzzy Vessel Tracking Algorithm

for Retinal Images Based on Fuzzy Clustering IEEE TRANSACTIONS ON
MEDICAL IMAGING, APRIL 1998

51
2. M. Elena Martinez-Perez, Alun D. Hughes, Alice V. Stanton, Simon A. Thom,
Neil Chapman, Anil A. Bharath, and Kim H. Parker Retinal Vascular Tree
Morphology: A Semi-Automatic Quantification, IEEE TRANSACTIONS ON
BIOMEDICAL ENGINEERING, AUGUST 2002.
3. Enrico Grisan, Alessandro Pesce, Alfredo Giani1, Marco Foracchia, and Alfredo
Ruggeri A new tracking system for the robust extraction of retinal vessel
structure, Proceedings of the 26th Annual International Conference of the
IEEE EMBS San Francisco, CA, USA, September2004.
4. Yi Yin, Mouloud Adel, Mireille Guillaume, and Salah Bourennane, A
PROBABILISTIC BASED METHOD FOR TRACKING VESSELS IN
RETINAL IMAGES, IEEE 17th International Conference on Image
Processing, September 2010,
5. Saurabh Garg, Jayanthi Sivaswamy, Siva Chandra, UNSUPERVISED
CURVATURE-BASED RETINAL VESSEL SEGMENTATION, CVIT,
International Institute of Information Technology, Hyderabad, India, 2007
6. Hind Azegrouz, Emanuele Trucco, MAX-MIN CENTRAL VEIN
DETECTION IN RETINAL FUNDUS IMAGES, School of Engineering and
Physical Sciences Department of Electrical Electronic and Computer
engineering, Heriot Watt University,2006.
7. V.Vijaya Kumari, N.SuriyaNarayanan, Diabetic Retinopathy-Early Detection
Using Image Processing Techniques, (IJCSE) International Journal on
Computer Science and Engineering , Vol. 02, No. 02, 2010, 357-361.
8. Neera Singh, Ramesh Chandra Tripathi, Automated Early Detection of
Diabetic Retinopathy Using Image Analysis Techniques , International Journal
of Computer Applications (0975 8887), Volume 8 No.2, October 2010
9. Huiqi Li, Wynne Hsu, Mong Li Lee*, and Tien Yin Wong, Automatic Grading
of Retinal Vessel Caliber, IEEE TRANSACTIONS ON BIOMEDICAL
ENGINEERING, VOL. 52, NO. 7, JULY 2005.

52
10.Akara Sopharak, Bunyarit Uyyanonvara, Automatic exudates detection from
diabetic retinopathy retinal images using Fuzzy C-means and morphological
methods,

53