CHAPTER 1

INTRODUCTION

Image Processing, whose goal is to investigate the history of an image

using passive approaches, has emerged as a possible way to solve the above
crisis. The basic idea underlying Image Processing is that most, if not all,
image processing tools leave some (usually imperceptible) traces into the
processed image, and hence the presence of these traces can be investigated in
order to understand whether the image has undergone some kind of processing
or not. In the last years many algorithms for detecting different kinds of traces
have been proposed which usually extract a set of features from the image and
use them to classify the content as exposing the trace or not.

Computerized analysis of medical images is gaining importance day by

day, as it often produces higher sensitivity irrespective of experience of the
analyst. For this reason various image analysis techniques are used for
automatic detection of various diseases. The retina is the innermost and most
important layer of eye, where the earliest pathological changes can be
observed. It is composed of several important anatomical structures, which can
indicate many diseases such as hypertension, diabetes and other various
diseases of eye. The most effective way to detect these diseases is to regular
screening of retinal fundus image. Diabetic retinopathy is one of the most
serious complications of diabetes mellitus and a major cause of blindness. It is
a progressive disease classified according to common cause of blindness
among people aged 30-69 years.

One-fifth of patients of diabetes type II, have retinopathy at the time of

diagnosis. In type I diabetes, diabetic retinopathy never occurs after diagnosis.
But after 15 years all most of all patients with type I and two-third of those
with type II diabetes have background of diabetic retinopathy . In case of
diabetic retinopathy blood vessels get damaged and protein and fat based
particles gets leaked out of the damaged blood vessels beside blood flow to the

1
retina decreases. These particles are referred to as exudates. Various methods
have been developed for detection of exudates. These include thresholding and
edge detection based techniques ,FCM based approach , gray level variation
based approach , multilayer perceptron based approach . Optic disk must be
detected and segmented early in the detection process,as often optic disk has
more or less same brightness and contrast as the exudates. So, if optic disk is
not segmented in early stage the process may produce wrong result. In this
paper we have proposed a method which dynamically calculates an optimal
thresholding value to detect hard exudates in fundus images.

For these reasons, taking a final decision about the authenticity of an

image relying on the output of a set of forensic tools is not a trivial task, thus
justifying the design of proper decision fusion methods explicitly thought for
this scenario.

1.1 Biomedical Image Processing

The commonly used term “biomedical image processing” means the

provision of digital image processing for biomedical sciences. By the
increasing use of the direct digital imaging systems for medical diagnostics,
digital image processing becomes more and more important in health care.
Based on digital imaging techniques, the entire spectrum of digital image
processing is now applicable in medicine. In general, digital image processing
covers four major areas - Image formation or Image acquisition, Image
visualization or Image enhancement, Image analysis and Image management.
1.1.1Image formation or Image Acquisition

Image formation or Image acquisition includes all the steps from

capturing the image to forming a digital image matrix. Numerous
electromagnetic and some ultrasonic sensing devices are frequently arranged in
the form of a 2-D array. The response of each sensor is proportional to the light
energy falling onto the surface of the sensor. Generally image acquisition stage
involves preprocessing like scaling.

2
 Nowadays, medical images have become a major component of
diagnostics, treatment planning and procedures, and follow-up studies.
Furthermore, medical images are used for education, documentation, and
research describing, morphology as well as physical and biological functions in
1D, 2D, 3D, and even 4D image data. Today, a large variety of imaging
modalities have been established, such as X-ray, Computed Tomography(CT),
Magnetic Resonance Imaging (MRI), Fluoroscopy, Ultrasound etc. which are
based on transmission, reflection or refraction of light, radiation, temperature,
sound, or spin. Obviously, an algorithm for delineation of an individual that
works with one imaging modality will not be applicable directly to another
modality.

1.1.2 Image Visualization or Image Enhancement

Image visualization or Image enhancement refers to all types of

manipulation of the image matrix, resulting in an optimized output image. The
goal is to process the image so that the result is more suitable than the original
image for a specific application. The word specific is important because the
methods for enhancing one kind of image may not be suitable for another kind,
e.g. X-ray images and space craft images. Image visualization or image
enhancement is lowlevel processing which denotes manual or automatic
techniques, which can be realized without a priori knowledge on the specific
content of images. These methods operate on the raw data as well as on pixel,
edge, or texture levels, and thus are at a minimal level of abstraction. The Low-
level methods of image Introduction processing, i.e., procedures and
algorithms, are mostly applied for pre- or postprocessing of medical images.
1.1.3 Image analysis

Image analysis includes processing used for quantitative measurements

as well as abstract interpretations of biomedical images. These steps require a
priori knowledge of the nature and content of the images, which must be
integrated into the algorithms at a higher level of abstraction. Thus, the process

3
of image analysisis very specific, and developed algorithms can be transferred
directly into other application domains. High-level image processing include
methods at the texture, region, object, and scene levels. The required
abstraction can be achieved by increased modelling of a priori knowledge.
Image analysis techniques require.

Extraction of certain features that aid in the identification of the object.

Image Analysis mainly involves segmentation, feature extraction and selection,
representation and description, classification or detection or recognition.

• Segmentation techniques are used to isolate the desired object from the scene
so that measurements can be made on it subsequently. Segmentation partitions
the image into its constituent connected regions or objects. The level to which
the subdivision is carried depends on the problem being solved. In medical
image processing, the definition accentuates the various diagnostically or
therapeutically relevant image areas, namely, the discrimination between
healthy anatomical structures and pathological tissue. By definition, the result
of segmentation is always at the regional level of abstraction. Depending on the
level of feature extraction required after segmentation, we can methodically
classify the procedures into pixel, edge, and texture or regionoriented
procedures. In addition, there are hybrid approaches, which result from
combination of single procedures.

• Representation and description almost follow the output of a segmentation

stage, which is usually raw pixel data, constituting the boundary of the region,
i.e. a set of pixels separating one region from another or all the points in it. In
either case, converting data to a suitable form for computer processing is
necessary. Therefore, the task of feature extraction is to emphasize image
information at the particular level, where subsequent algorithms operate.
Consequently, information provided on other levels must be suppressed. Thus,
a data reduction to obtain the characteristic properties is executed.

4
• Description is also called feature selection. It deals with extracting the
attributes that result in some quantitative information of interest or is basic for
differentiating one class of object from another.

• Recognition is a process that assigns a label to an object, based on its

descriptors. This is usually achieved through classification or detection of
objects or regions in an image. According to the general processing chain, the
task of the classification/detection is to assign all connected regions which are
obtained from the segmentation, to particularly specified classes of objects.
Usually, region-based features that sufficiently abstract the characteristics of
the objects are used to guide the classification process. These extracted features
must be sufficiently discriminative and suitably adopted to the application,
since they fundamentally impact the resulting quality of the classifier/detector.
The classification itself reverts mostly to known numerical (statistical) and non-
numerical (syntactic) procedures as well as the newer approaches of
Computational Intelligence (CI), such as neural networks, evolutionary
algorithms, and fuzzy logic. In general, the individual features, which are
determined by different procedures, are summarized either to numerical feature
vectors (also referred to as signature) or abstract strings of symbols. Statistical
classification regards object identification as a problem of the statistical
decision theory. A syntactic classifier can be understood as a knowledge-based
classification system (expert system), because the classification is based on a
formal heuristic, symbolic representation of expert knowledge, which is
transferred into image processing systems by means of facts and rules.
1.1.4 Major Challenges in Biomedical Image Processing

Using medical images, it is difficult to formulate a priori knowledge

such that it can be integrated directly and easily into automatic algorithms of
image processing. This is referred to as the semantic gap, which means the
discrepancy between the cognitive interpretation of a diagnostic image by the
physician (high level) and the simple structure of discrete pixels, which is used

5
in computer programs to represent an image (low level). In the medical
domain, there are three main aspects hindering bridging this gap

• Heterogeneity of images: Medical images display living tissue, organs, or

body parts. Even if captured with the same modality and following a
standardized acquisition protocol, shape, size, and internal structures of these
objects may vary remarkably not only from patient to patient (inter-subject
variation), but also among different views of the same patient and similar views
of the same patients at different times (intra-subject variation).

• Unknown delineation of objects: Frequently, biological structures cannot

be separated from the background because the diagnostically or therapeutically
relevant object is represented by the entire image. Even if definable objects are
observed in biomedical images, their segmentation is problematic because the
shape or borderline itself is represented fuzzily or only partly. Hence,
medically related items often can be abstracted most at the texture level.

• Robustness of algorithms: In addition to these inherent properties of

medical images, which complicate their high-level processing, special
requirements of reliability and robustness of medical procedures, when applied
in routine, image processing algorithms are also demanded in the medical area.
As a rule, automatic analysis of images in medicine should not provide wrong
measurements. This means that, images which cannot be processed correctly,
must be automatically, rejected and withdrawn from further processing.
Consequently, all images that have not been

rejected must be evaluated correctly.

1.2 Literature Survey:
1.2.1 A New Dynamic Thresholding Based Technique for Detection
of Hard Exudates in Digital Retinal Fundus Image
I Diptoneel Kayal, II Sreeparna Banerjee

In this paper, They present Diabetic retinopathy is one of the major

causes of blindness among diabetes mellitus patients. In case of diabetic

6
retinopathy blood vessels of retina gets ruptured and lipoprotein substances
gets leaked out of the damaged blood vessels and are deposited in the intra
retinal space. The part of optic nerve beneath these substances are barred from
being excited by light rays and failed to produce any nerve impulse to brain
and leads to partial loss of vision. These lipoprotein substances are yellowish in
color and are termed as exudates. Exudates are the primary as well as most
important sign of the presence of diabetic retinopathy. If the disease is not
detected in early stages then it may lead to complete loss of vision to the
diabetes patients. But detection of exudates is extremely difficult to detect by
visual inspection due to small inner diameter of retina. Sometimes inadequate
illumination makes the problem worse.

An efficient image analysis program can detect the presence effectively.

In this paper we have proposed one such method is proposed. The proposed
method uses various image processing techniques, such as median filtering,
image thresholding etc. with a goal to detect hard exudates.

To reduce this we need Computerized analysis of medical images is

gaining importance day by day, as it often produces higher sensitivity
irrespective of experience of the analyst. For this reason various image analysis
techniques are used for automatic detection of various diseases. The retina is
the innermost and most important layer of eye, where the earliest pathological
changes can be observed. It is composed of several important anatomical
structures, which can indicate many diseases such as hypertension, diabetes
and other various diseases of eye. The most effective way to detect these
diseases is to regular screening of retinal fundus image.Diabetic retinopathy is
one of the most serious complications of diabetes mellitus and a major cause of
blindness.

7
1.2.2 Automated Segmentation of Blood Vessels for Detection of
Proliferative Diabetic Retinopathy
M. Usman Akram, Ibaa Jamal, Anam Tariq and Junaid Imtiaz

In this paper Retinal image analysis is very effective in early detection

and diagnosis of diabetic retinopathy. Diabetic retinopathy is a progressive
disease and is broadly classify into two stages i.e. Non proliferative diabetic
retinopathy (NPDR) and Proliferative diabetic retinopathy (PDR). A sign of
PDR is the appearance of new blood vessels in fundus area and inside optic
disc known as neovascularization. The study of blood vessel is very important
for detection of neovascularization. In this paper, we present a method for
accurate blood vessel detection which can be used for detection of
neovascularization. The paper presents a new method for vessel segmentation
using a multilayered thresholding technique. The method is tested using two
publicly available retinal image databases and experimental results show the
significance of proposed work.

DR is broadly divided into two stages i.e Non proliferative diabetic

retinopathy (NPDR) and proliferative diabetic retinopathy (PDR). NPDR also
known as background DR contains the early signs of presence of DR such as
microaneurysms and dot haemorrhages caused by the breaks in tiny small
vessels called capillaries. As the disease progresses, more signs of DR appear
such as hard and soft exudates which are caused by the leakage of fats and
proteins on the surface of retina known as severe NPDR. In second stage of DR
i.e PDR, new abnormal blood vessels appear in retina called neovascularization
(NV).In this paper, we present a method for blood vessel enhancement and
segmentation to detect presence of neovascularization. Two dimensional Gabor
wavelets are used to enhance the vascular pattern. A new multilayered
thresholding technique is presented for vessel segmentation which applies a
recursive algorithm for blood vessel segmentation. A comparison for blood
vessel segmentation is done to evaluate the validity of proposed algorithm.

8
1.2.3 New Feature-Based Detection of Blood Vessels and Exudates in
Color Fundus Images

Doaa Youssef, Nahed Solouma, Amr El-dib, Mai Mabrouk, and Abo-Bakr
Youssef Exudates are one of the earliest and most prevalent symptoms
of diseases
leading to blindness such as diabetic retinopathy and wet macular degeneration.
Certain areas of the retina with such conditions are to be photocoagulated by
laser to stop the disease progress and prevent blindness. Outlining these areas is
dependent on outlining the exudates, the blood vessels, the optic disc and the
macula and the region between them. The earlier the detection of exudates in
fundus images, the stronger the kept sight level. So, early detection of exudates
in fundus images is of great importance for early diagnosis and proper
treatment. In this paper, we provide a feature-based method for early detection
of exudates. The method is based on segmenting all objects that have contrast
with the background including the exudates. The exudates could then be
extracted after eliminating the other objects from the image. We proposed a
new method for extracting the blood vessel tree based on simple morphological
operations. The circular structure of the optic disc is obtained using Hough
transform. The regions representing the blood vessel tree and the optic disc are
set to zero in the segmented image to get an initial estimate of exudates. The
final estimation of exudates are obtained by morphological reconstruction. This
method is shown to be promising as we can detect the very small areas of
exudates.

Exudates are formed by the leakage of proteins and lipids from the
bloodstream into the retina via damaged blood vessels. In retinal images, hard
exudates appear as bright yellow lesions with varying sizes, shapes, and
locations. They also have a considerable contrast with respect to the
background. The optic disk, bright circular region from where the blood vessels
emanate, is the only area in the fundus images having the same brightness and

9
colour range like the exudates. the appearance of the exudates and the optic
disc in the colored fundus images. So, detection of exudates could accurately
be done by extracting the bright yellow regions after elimination the optic disc
area from the , Image with hard exudates Several techniques have been
developed for exudates detection in fundus images. Akara et al use maximum
variance to obtain the optic disk center and a region growing segmentation
method to obtain the exudates. Blood vessel intersection property is used in to
obtain the optic disk. Based on its color characteristics, the authors in
composed a simple Bayesian classifier to detect the exudates. Extraction of
exudates and blood vessels by computing the difference map and k-means
clustering is introduced in . Color normalization and local contrast
enhancement followed by fuzzy C-means clustering and neural networks were
used by Osareh . The system works well only on Luv color space but in the
case of non-uniform illumination the detection accuracy is low. In a naïve
Bayes classifier for diagnosis of diseases from retinal image is applied and this
can provide a good decision support to ophthalmologist. Walter et al proposed
a method for automated identification of exudates in colour fundus images
using mathematical morphology techniques.
1.2.4 Automated Detection of Diabetic Retinopathy Using Fundus
Image Analysis
aspreet Kaur ,Dr. H.P.Sinha

The present study is aimed at developing an automatic system for the

extraction of normal and abnormal features in color retinal images. Prolonged
diabetes causes micro-vascular leakage and micro-vascular blockage within the
retinal blood vessels. Filter based approach with morphological filters is used
to segment the vessels. The morphological filter are tuned to match that part of
vessel to be extracted in a green channel image. To classify the pixels into
vessels and non vessels local thresholding based on gray level co-occurrence
matrix is applied. The performance of the method is evaluated on two publicly
available retinal databases with hand labeled ground truths.

10
 Hemorrhages are of two types: Flame and Dot-blot hemorrhages. Flame
hemorrhages occur at the nerve fibers and they originate from precapillary
arterioles, which located at the inner layer of the retina. Dot and blot
hemorrhages are round, smaller than micro aneurysms and occur at the various
levels of retina especially at the venous end of capillaries. Hard exudates are
shinny, irregularly shaped and found near prominent microaneurysms or at the
edges of retinal edema. In the early stage, the vision is rarely affected and the
disease can be identified only by regular dilated eye examinations.

Fundus images are used for diagnosis by trained clinicians to check for
any abnormalities or any change in the retina. They are captured by using
special devices called ophthalmoscopes. Each pixel in the fundus image
consists of three values namely red, green and blue, each value being quantised
to 256 levels. Diabetic Maculopathy(DM) is a stage where fluid leaks out of
damaged vessels and accumulates at the center of the retina called macula
(which helps in seeing the details of the vision very clearly) causing permanent
loss of vision. This water logging of the macula area is called clinically
significant macular oedema which can be treated by laser treatment.

Proliferate diabetic retinopathy, which is defined as the growth of

abnormal new vessels (neovascularization) on the inner surface of the retina are
divided into two categories: neovasculature of the optic disk and
neovascularization elsewhere in the retina. The above stages can be seen
clearly in Fig. which shows different changes that take place in the retina of a
DR patient over a period of time.

Diabetic Maculopathy is a stage where fluid leaks out of damaged

vessels and accumulates at the center of the retina called macula (which helps
in seeing the details of the vision very clearly) causing permanent loss of
vision. This water logging of the macula area is called clinically significant
macular oedema which can be treated by laser treatment The performance of
retinal vessels on drive database, sensitivity 86.39%, accompanied by

11
specificity of 91.2%. While for STARE database proposed method sensitivity
92.15 % and specificity 84.46%. The system could assist the ophthalmologists,
to detect the signs of diabetic retinopathy in the early stage, for a better
treatment plan and to improve the vision related quality of life.
1.2.5 A semi-automated technique for labeling and counting of
apoptosing retinal cells

Mukhtar Bizrah, Steve C Dakin, Li Guo, Farzana Rahman, Miles Parnell,

Eduardo Normando, Shereen Nizari,Benjamin Davis, Ahmed Younis6 and
M Francesca Cordeiro

Retinal ganglion cell (RGC) loss is one of the earliest and most
important cellular changes in glaucoma. The DARC (Detection of Apoptosing
Retinal Cells) technology enables in vivo real-time non-invasive imaging of
single apoptosing retinal cells in animal models of glaucoma and Alzheimer‟s
disease. To date, apoptosing RGCs imaged using DARC have been counted
manually. This is time-consuming, labour-intensive, vulnerable to bias, and has
considerable inter- and intra-operator variability. Automated analysis included
a pre-processing stage involving local-luminance and local-contrast “gain
control”, a “blob analysis” step to differentiate between cells, vessels and noise,
and a method to exclude non-cell structures using specific combined „size‟ and
„aspect‟ ratio criteria. Apoptosing retinal cells were counted by 3 masked
operators, generating „Gold-standard‟ mean manual cell counts, and were also
counted using the newly developed automated algorithm. Comparison between
automated cell counts and the mean manual cell counts on 66 DARC images
showed significant correlation between the two methods (Pearson‟s correlation
coefficient 0.978 (p < 0.001), R Squared = 0.956.

12
 CHAPTER 2
STRUCTURE AND FUNCTION OF THE EYE
2.1 Basic Structure of the Eye

The five human senses, the eye is one of its most important sense
organs by the quality and precision of the signals it captures and by the
reduction of quality of life within its absence. It was estimated that 70% of the
sensorial information interpreted by humans is captured by the eye (Davidovits
2001). It has the ability to capture the light and convert it in the retina into
electric signals that are sent to human brain for interpretation. Its functioning is
similar to conventional image capture systems due to its lenses that refract and
focus the incoming light in the sensorial region. The optical system (figure 2.1)
is composed by the lens, the iris and the cornea. As a result of the incoming
light and in order to refract and focus, the lens reshapes with the help of
auxiliary muscles. The control of the amount of light that enters the eye
through the pupil is done by the iris, which is a tissue that is able to contract
and expand, by decreasing or increasing the size of the pupil, respectively
(figure2.1).

The cornea is a transparent and protective layer that covers the iris and
the pupil and is the first refracting layer. With approximately 43 dioptres of
refractive power the human cornea is the major focusing element of the eye,
although it is fixed. The variable focus is obtained by the lens which in a
natural environment has approximately 18 dioptres of refractive power The
light that enters the eye is projected into the retina, which contains the
photoreceptors. There are different types of photoreceptors, some are for low-
light vision and black-white perception (rods) and others are for colour
perception and daytime vision (cones). Although these are dispersed along the
retina, which occupies 72% of a sphere with 22mm of

13
 13

Figure2.1 Light Propagation In The Retina. (A) The Light Behaviour For An
Undilated And A Dilated Pupil; (B) Light Refraction In The Eye With A
Relaxed And A Stretched Lens.

diameter, they are more concentrated on the central part of the retina, the
macula. This latter is about 6000μm of diameter and contains a high density of
photoreceptors. The photoreceptors that are outside the macula capture the
peripheral vision, which in case of damage is less noticed or even unnoticed
than in the macula. In the centre of the macula with 1000μm of diameter is the
fovea. It contains 50% of the photoreceptors of the retina being responsible for
the sharp and high resolution central vision used by humans for reading,
watching television, driving, and any other activity where visual detail is of
primary importance. The retina is divided into layers from the nerve fibre layer
to the Bruch‟s Membrane. The light that reaches the retina crosses the nerve
fibre and the ganglions‟ layers to be captured by the photoreceptors (rods and
cones).

These return the light information to the ganglions which gather and
compress the information from several neighbouring photoreceptors and send it
through the nerve to the brain. The Retinal Pigmented Epithelium (RPE) is a
layer of cells that protects and nourishes the retina, removes waste products,
prevents new blood vessel growth into the retinal layer and absorbs light not
absorbed by the photoreceptor cells; these actions prevent the scattering of the

14
light and enhance vision clarity. Finally, the Bruch‟s Membrane is a thin layer
that acts as a blood-retinal barrier and as a support to both RPE and Choroid.
The information collected by the retina is gathered into a set of nerve fibres
which are forwarded to the brain through the optic nerve. This latter includes
also the veins and arteries which supply the blood to the eye. The optic disc is
located within the retina and does not have photoreceptors which create a blind
spot in the visual field.
2.2 Introduction to Retinal Diseases

Blindness is one of the most undesirable consequences of any eye

disease. According to the World Health Organization (2007) the major causes
of blindness that occur nowadays worldwide are the opacity of the lens
provoked by Cataracts, Glaucoma which affects the optic nerve and involves
the loss of ganglions, and Retinopathies that affect the retina, such as ARMD
and Diabetic Retinopathy by distorting the image, or in an advanced stage, with
the loss of photoreceptors In this work only a particular retinopathy is
approached, the ARMD. However, in order to contextualize the reader an
overview on other retinopathies which can have confounding characteristics
such as the Diabetic Retinopathy and the Ocular Melanoma will be presented
in the following sections.

Figure2.2 Cross-Sectional View Of A Human Eye. (From (National

Eye Institute 2009)).

15
2.3 Diabetic Eye Diseases

In this region and therefore the absence of photoreceptor cells results in

a blind spot in the retina. The nutritional support to the retina is provided by the
choroid and the two main capillary networks: the nerve fibre layer network and
the connecting neuron layer network. The capillary density increases towards
the centre region of the retina and the most dense network is found in the
macula, but the fovea itself is absent of capillaries.

Therefore, the fovea is dependent on the choroidal blood supply from

the vascular layer behind the retina (choroid). The presented anatomical parts
(macula, fovea, capillaries, and optic nerve head) highlighted are the relevant
structures of the retina in terms of retinal diseases and this thesis.

There are a number of reasons that can cause reduced visual acuity,
visual impairment, and blindness. In diabetic eye diseases, the cause of visual
disturbances is in most cases related to those vascular changes diabetes is
causing to the eye. The discussion in this section concentrates on the diabetic
eye diseases that encompass a group of eye problems, such as diabetic
retinopathy, cataract, neovascular glaucoma and diabetic neuropathies . The
section discusses how the symptoms of the diabetic eye diseases emerge and
how they affect the vision.

Diabetic retinopathy is a microvascular complication of diabetes,

causing abnormalities in the retina. Typically there are no salient symptoms in
the early stages, but the number and severity predominantly increase in time. In
the following, the progress of the disease is described in detail.

The diabetic retinopathy typically begins as small changes in the retinal

capillaries. The smallest detectable abnormalities, micro aneurysms (MA),
appear as small red dots in the retina and are local distensions of the weakened
retinal capillary . Due to these damaged capillary walls, the small blood vessels
may rupture and cause intraregional hemorrhages (HA).

16
 The diabetic retinopathy also increase the permeability of the capillary
walls which results in retinal edema and hard exudates (HE). The hard exudates
are lipid formations leaking from the weakened blood vessels and appear
yellowish with well-defined borders. If the local capillary circulation and
oxygen support fail due to obstructed blood vessels, pale areas with indistinct
margins appear in the retina. These areas are small micro infarcts known as soft
exudates (Se) . Intraregional microvascu5lar abnormalities (IRMA) and
venopathy are signs of a more severe stage of diabetic retinopathy, where
intraregional microvascular abnormalities appear as dilation in the capillary
system and venopathy as shape changes in artery and veins.

(a) (b) (c)

(d) (e)

Figure 2.3: Symptoms Of Diabetic Retinopathy (Images Processed For Better

Visualisation): (A) Microaneurysm; (B) Haemorrhages; (C) Hard Exudates;(D)
Soft Exudates; (E) Neovascularisation

Stages of diabetic retinopathy and maculopathy The severity of diabetic

retinopathy is divided into two stages: nonproliferative (background
retinopathy) and proliferative retinopathy. The nonproliferative retinopathy

17
indicates the presence of diabetic retinopathy in the eye and consist of micro
aneurysms, hemorrhages, exudates, retinal oedema, IRMA and venopathy .

Figure 2.4(a) Human Vision (b) Vision Loss Due To DR

Example of nonproliferative diabetic retinopathy: eye fundus image

showing hard exudates; close up image of the hard exudates. If the
nonproliferative retinopathy is untreated or undiagnosed it will turn into
proliferative retinopathy which is also an eye-sight threatening condition. The
proliferative diabetic retinopathy may cause sudden loss in visual acuity or
even a permanent blindness due to vitreous haemorrhage or tractional
detachment of the central retina. This stage is considered if neovascularisation
or vitreous/preretinal haemorrhage is present in the retina .

Figure 2.5 - Global Causes Of Blindness Due To Eye Diseases And

Uncorrected Refractive Errors Estimated In 2006. (From (World Health
Organization 2009))

18
2.4 The History of a Changing Diabetes

There have been many improvements in diabetes care from the discovery of
insulin in the 1920s until the present day (1‐3). These developments imply
that many older patients with diabetes today have lived through some
substantial changes. Since most diabetic complications of diabetes (including
DR) develop with time, the incidence and progression of these depend on the
type of diabetes care available in the relevant decades. Consequently, patients
who were diagnosed with diabetes 60 years ago have a different prognosis
compared to those who are diagnosed with diabetes today. Insulin was first
discovered in 1921 by Banting and Best who two years later were awarded the
Nobel Prize for their discovery of its effect on patients with diabetes.

The introduction of treatment with insulin meant patients nolonger

necessarily succumbed to ketoacidosis and death at a very short duration of
diabetes. The average survival time before the introduction of insulin was less
than 5 years after diabetes onset . At the start, however, insulin was only given
once a day, and patients were initially not able to selfmonitor their glucose
levels at home. The importance of good glycaemic control for reducing the risk
of microvascular complications was not presented in the literature until 1988,
when the Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR)
presented observational data that showed this, Banting was awarded the Nobel
Prize together with John McLeod, but decided to share it with Best.

19
 CHAPTER 3
DIABETIC RETINOPATHY
3.1 Introduction

Blindness is one of the most undesirable consequences of any eye

disease. According to the World Health Organization (2007) the major causes
of blindness that occur nowadays worldwide are the opacity of the lens
provoked by Cataracts, Glaucoma which affects the optic nerve and involves
the loss of ganglions, and Retinopathies that affect the retina, such as ARMD
and Diabetic Retinopathy by distorting the image, or in an advanced stage, with
the loss of photoreceptors. In this work only a particular retinopathy is
approached, the ARMD. However, in order to contextualize the reader an
overview on other retinopathies which can have confounding characteristics
such as the Diabetic Retinopathy and the Ocular Melanoma will be presented
in the following sections.

Diabetes has become one of the rapidly increasing health threats

worldwide . Only in Finland there are 30 000 people diagnosed to the type 1
maturity onset diabetes in the young, and 200 000 people diagnosed to the type
2 latent autoimmune diabetes in adults . In addition, the current estimate
predicts that there are 200 000 undiagnosed patients Proper and early treatment
of diabetes is cost effective since implications of poor or late treatment are very
expensive. In Finland, only 10% of the total health care costs of diabetes arise
from 70% of early diagnosed patients while the remaining 90% arise from the
patients having poor treatment (30%) . This fact promotes the study of
automatic diagnosis methods for screening over larger populations. Fundus
imaging has an important role in diabetes monitoring since occurrences of
retinal abnormalities are common and consequences serious. However, since
the eye fundus seems to be sensitive to vascular diseases, fundus imaging is
considered as a candidate for non-invasive screening of diabetes. The success

20
rate of screening depends on accurate fundus image capturing and especially on
accurate and reliable image processing algorithms for detecting the
abnormalities.

Various algorithms have been proposed by many research groups for

this purpose. However, it is impossible to judge the accuracy and reliability of
the approaches because of the lack of commonly accepted and representative
fundus image database and evaluation protocol. The commonly accepted
protocol could evaluate the maturity of state-of-the-art methods, i.e., produce
the achieved sensitivity and selectivity rates. This would finally allow the
technology transfer from research laboratories to practice. The main
contribution of this work is a framework for evaluating methods for an
automatic diabetic retinopathy diagnosis. A prototype database with the ground
truth and evaluation protocol are proposed. Experimental results for a method
from the literature are reported. This study provides the means for the trustable
evaluation of automatic diabetic retinopathy. The idea for this approach
originates from the strict regulations for the valuation of biometric
authentication method, e.g., the FERET protocol for face recognition methods.

A patient with Diabetes may experience several ocular complications,

such as Cataracts, Glaucoma and Diabetic Retinopathy, which can lead to
irreversible blindness. This latter is the most common within the patients with
diabetes and is one of the leading causes of American Adults blindness
(Prevent Blindness America 2008). It is caused by changes in the blood vessels
of the retina that can lead to haemorrhages, blurring the vision. These changes
can be new vessels that grow on the surface of the retina or vessels that swell
and start bleeding. This disease often has no early signs. Only after small
haemorrhages do the patients

feel their vision blurred. Firstly this sensation can be temporary. However, after
some days or weeks, a severe bleeding is likely to occur and damage the retina
irreversibly. In a funduscopic examination of a retina affected by Diabetic
Retinopathy, the ophthalmologist may find cotton wool spots and

21
haemorrhages (figure3.1). The cotton wool spots are regions where the blood
supply has been obstructed, exhibiting a white reflection in a distorted region
as a consequence. The haemorrhages are the darker spots which exhibit an
irregular shape, as illustrated in figure 3.1
3.2 Drusen in Eye

Image enhancement methods: used normal retinal findings (vasculature,

optic disk, fovea, and abnormal findings) to estimate the illumination
component using iterative robust homographic surface fitting to compensate
the nonuniform illumination in fundus images. In detection of bright diabetic
retinopathy areas from fundus images, applied adaptive local contrast
enhancement to sub-image areas using the local mean and standard deviation of
intensities. The same approachwas used by Osareh after color normalization
between fundus images using histogram specification. adjusted the image
brightness using brightness transform function similar to gamma correction.

Detection and classification methods: determined abnormal and normal

finding areas using intensity properties for dynamic clustering. From the result
abnormal areas, hard exudates were separated from soft exudates and drusen
using intensity contrast information between abnormal areas and immediate
background. The domain knowledge of retinal blood vessels were used to
remove false artifacts. eliminated the vessels by applying morphological
closing to the luminance component of the fundus image. From the result,
within a sliding window local standard variation image was calculated and
thresholded into coarse exudate areas. More accurate countours were acquired
by thresholding difference between original image and morphologically
reconstructed image. S used yellowish color and sharp edges to distinguish
hard exudates from the fundus images.

The image pixels were classified into background and yellowish objects
using minimum distance discrimination, where the countour pixels of extracted
optic disk were used as background color reference and pixels inside the

22
contour were used as yellowish object color reference. The segmented
yellowish areas and their edge information extracted with Kirsch‟s mask were
combined to hard exudate areas using boolean operator. located the bright
abnormal regions in fundus images by applying fuzzy c-means clustering in
LUV color space. The result areas were classified to hard exudates, soft
exudates, and normal findings using support vector machine. Osareh searched
the coarse hard exudate areas using fuzzy c-means clustering with Gaussian
smoothed histograms of each color band of the fundus image.

The segmented areas were classified to exudate and non-exudate regions

using neural networks. Color, region size, mean and standard deviation of
intensity, and edge strength were used as features. segmented exudates with
combination of Canny edge detection and region growing in LUV color space.
Gradient, mean pixel value, and seed pixel value were used as criteria in region
growing. used a similar approach for bright abnormal region detection as they
used for finding abnormal red regions. In addition to the previous work, the
prior knowledge of optic disk and vascular arch were used to improve detection
accuracy. clustered similar pixels using intensity difference as criteria for
recursive region growing. The region with the most pixels were considered as
background and defined the threshold value for hard exudate areas.

Figure 3.1 – A Retina Affected By Diabetic Retinopathy, Where Cotton

Wool Spots (Bright Spots) And Haemorrhages (Darker Spots) Are Visible.

23
 CHAPTER 4
GRAYSCALE
4.1Introduction

In photography and computing, a grayscale or greyscale digital image is

an image in which the value of each pixel is a single sample, that is, it carries
only intensity information. Images of this sort, also known as black-and-white,
are composed exclusively of shades of gray, varying from black at the weakest
intensity to white at the strongest.

Grayscale images are distinct from one-bit bi-tonal black-and-white

images, which in the context of computer imaging are images with only the
two colors, black, and white (also called bilevel or binary images). Grayscale
images have many shades of gray in between. Grayscale images are often the
result of measuring the intensity of light at each pixel in a single band of the
electromagnetic spectrum (e.g. infrared, visible light, ultraviolet, etc.), and in
such cases they are monochromatic proper when only a given frequency is
captured. But also they can be synthesized from a full color image; see the
section about converting to grayscale. The intensity of a pixel is expressed
within a given range between a minimum and a maximum, inclusive.

This range is represented in an abstract way as a range from 0 (total

absence, black) and 1 (total presence, white), with any fractional values in
between. This notation is used in academic papers, but this does not define
what "black" or "white" is in terms of colorimetry. Another convention is to
employ percentages, so the scale is then from 0% to 100%. This is used for a
more intuitive approach, but if only integer values are used, the range
encompasses a total of only 101 intensities, which are insufficient to represent
a broad gradient of grays. Also, the percentile notation is used in printing to
denote how much ink is employed in halftoning, but then the scale is reversed,
being 0% the paper white (no ink) and 100% a solid black.

24
 In computing, although the grayscale can be computed through rational
numbers, image pixels are stored in binary, quantized form. Some early
grayscale monitors can only show up to sixteen (4-bit) different shades, but
today grayscale images (as photographs) intended for visual display (both on
screen and printed) are commonly stored with 8 bits per sampled pixel, which
allows 256 different intensities
(i.e., shades of gray) to be recorded, typically on a non-linear scale. The
precision provided by this format is barely sufficient to avoid visible
banding artifacts, but very convenient for programming due to the fact that
a single pixel then occupies a single byte.

Technical uses (e.g. in medical imaging or remote sensing

applications) often require more levels, to make full use of the sensor
accuracy (typically 10 or 12 bits per sample) and to guard against roundoff
errors in computations. Sixteen bits per sample (65,536 levels) is a
convenient choice for such uses, as computers manage 16-bit words
efficiently. The TIFF and the PNG (among other) image file formats
support 16-bit grayscale natively, although browsers and many imaging
programs tend to ignore the low order 8 bits of each pixel.No matter what
pixel depth is used, the binary representations assume that 0 is black and the
maximum value (255 at 8 bpp, 65,535 at 16 bpp, etc.) is white, if not
otherwise noted.
4.2 Converting Color to Grayscale

Conversion of a color image to grayscale is not unique; different

weighting of the color channels effectively represent the effect of shooting
black-and-white film with different-colored photographic filters on the
cameras.

4.2.1 Syntax
 I = rgb2gray(RGB) 

 newmap = rgb2gray(map) 

25
 4.2.2Description

I = rgb2gray(RGB) converts the truecolor image RGB to the

grayscale intensity image I. The rgb2gray function converts RGB images to
grayscale by eliminating the hue and saturation information while retaining
the luminance. If you have Parallel Computing Toolbox™ installed,
rgb2gray can perform this conversion on a GPU.

newmap = rgb2gray(map) returns a grayscale colormap equivalent to map

4.3 Colorimetric (luminance-preserving) conversion to grayscale

A common strategy is to use the principles of photometry or, more

broadly, colorimetry to match the luminance of the grayscale image to the
luminance of the original color image. This also ensures that both images will
have the same absolute luminance, as can be measured in its SI units of
candelas per square meter, in any given area of the image, given equal
whitepoints. In addition, matching luminance provides matching perceptual
lightness measures, such as L* (as in the 1976 CIE Lab color space) which is
determined by the luminance Y (as in the CIE 1931 XYZ color space) .

To convert a color from a colorspace based on an RGB color model to a

grayscale representation of its luminance, weighted sums must be calculated in
a linear RGB space, that is, after the gamma compression function has been
removed first via gamma expansion.

For the sRGB color space, gamma expansion is defined as

where Csrgb represents any of the three gamma-compressed sRGB

primaries (Rsrgb, Gsrgb, and Bsrgb, each in range [0,1]) and Clinear is the

26
corresponding linear-intensity value (R, G, and B, also in range [0,1]). Then,
luminance is calculated as a weighted sum of the three linear-intensity values.
The sRGB color space is defined in terms of the CIE 1931 linear luminance Y,
which is given by
.

The coefficients represent the measured intensity perception of typical

trichromat humans, depending on the primaries being used; in particular,
human vision is most sensitive to green and least sensitive to blue. To encode
grayscale intensity in linear RGB, each of the three primaries can be set to
equal the calculated linear luminance Y (replacing R,G,B by Y,Y,Y to get this
linear grayscale). Linear luminance typically needs to be gamma compressed to
get back to a conventional non-linear representation. For sRGB, each of its
three primaries is then set to the same gamma-compressed Ysrgb given by the
inverse of the gamma expansion above as

FIGURE4.1: Gray Scale Image

27
 CHAPTER 5
MORPHOLOGICAL TRANSFORMATIONS
5.1. what is Morphology?

Morphology is a technique of image processing based on shape and

form of objects. Morphological methods apply a structuring element to an input
image, creating an output image at the same size. The value of each pixel in the
input image is based on a comparison of the corresponding pixel in the input
image with its neighbors. By choosing the size and shape of the neighbor, you
can construct a morphological operation that is sensitive to specific shapes in
the input image. The morphological operations can first be defined on
grayscale images where the source image is planar (single-channel). The
definition can then be expanded to full-colour images.
5.2. Morphological Operations

Morphological operations such as erosion, dilation, opening, and

closing. Often combinations of these operations are used to perform
morphological image analysis. There are many useful operators defined in
mathematical morphology. They are dilation, erosion, opening and closing.
Morphological operations apply structuring elements to an input image,
creating an output image of the same size. Irrespective of the size of the
structuring element, the origin is located at its centre. Morphological opening is
( )( ) B f x m g and Morphological closing is

where m a homothetic parameter, size is m means a square of (2m +1)´(2m +1)

pixels. B is the structuring element of size 3 × 3 (here m = 1).

bwhitmiss Binary hit-miss operation

bwmorph Morphological operations on binary images

28
bwareaopen Remove small objects from
binary image

imbothat Bottom-hat filtering

imclearborder Suppress light structures

connected to image border

imclose Morphologically close image

imdilate Dilate image

imerode Erode image

imextendedmax Extended-maxima transform

imextendedmin Extended-minima transform

imfill Fill image regions and holes

imhmax H-maxima transform

imhmin H-minima transform

imimposemin Impose minima

imopen Morphologically open image

imreconstruct Morphological reconstruction

imregionalmax Regional maxima

imregionalmin Regional minima

imtophat Top-hat filtering

watershed Watershed transform

conndef Create connectivity array

iptcheckconn Check validity of connectivity

argument

29
5.2.1 Dilation

Dilation is a transformation that produces an image that is the same

shape as the original, but is a different size. Dilation stretches or shrinks the
original. Dilation increases the valleys and enlarges the width of maximum
regions, so it can remove negative impulsive noises but do little on positives
ones.
The dilation of A by the structuring element B is defined by:

If B has a center on the origin, as before, then the dilation of A by B can

be understood as the locus of the points covered by B when the center of B
moves inside
A. Dilation of image f by structuring element s is given by f Ås . The
structuring element s is positioned with its origin at (x, y) and the new pixel
value is determined using the rule:

The following figure illustrates the morphological dilation of a gray

scale image. Note how the structuring element defines the neighbourhood of
the pixel of interest, which is circled. The dilation function applies the
appropriate rule to the pixels in the neighbourhood and assigns a value to the
corresponding pixel in the output image. In the figure, the morphological
dilation function sets the value of the output pixel to 16 because it is the
maximum value of all the pixels in the input pixel's neighbourhood defined by
the structuring element is on. Easiest way to describe it is to imagine the same
fax/text is written with a thicker pen In a binary image, if any of the pixels is
set to the value 1, the output pixel is set to 1.

30
 Figure5.1: Dilation operation

5.2.2 Erosion

It is used to reduce objects in the image and known that erosion reduces
the peaks and enlarges the widths of minimum regions, so it can remove
positive noises but affect negative impulsive noises little.

Figure5.2: Erosion operation

The erosion of the dark blue square resulting in light blue square.
The erosion of the binary image A by the structuring element B is defined by:

In a binary image, if any of the pixels is set to 0, the output pixel is set to 0.

Erosion of image f by structuring element s is given by f _s. The structuring

element s is positioned with its origin at (x, y) and the new pixel value is
determined using the rule:

31
In the above equation fit means all on pixel in the structuring element covers an
on pixel in the image. The following figure illustrates the morphological
erosion of a gray scale image. Note how the structuring element defines the
neighbourhood of the pixel of interest, which is circled. The dilation function
applies the appropriate rule to the pixels in the neighbourhood and assigns a
value to the corresponding pixel in the output image. In the figure, the
morphological erosion function sets the value of the output pixel to 14 because
it is the minimum value of all the pixels in the input pixel's neighbourhood
defined by the structuring element is on.

Figure 5.3. Morphological Erosion of a Grayscale Image

5.2.3 Opening Operation

Figure5.4:Opening operation

The opening of A by B is obtained by the erosion of A by B, followed by

dilation of the resulting image by B:

In the case of the square of side 10, and a disc of radius 2 as the structuring
element, the opening is a square of side 10 with rounded corners, where the
corner radius is 2.

32
5.2.4 Closing Operation
Closing of an image is the reverse of opening operation.

Figure5.5:Closing operation

The closing of A by B is obtained by the dilation of A by B, followed by

erosion of the resulting structure by B:

The first method proposed is the block analysis where the entire image
is split into a number of blocks and each block is enhanced individually. The
next proposed method is the erosiondilation method which is similar to block
analysis but uses morphological operations (erosion and dilation) for the entire
image rather than splitting into blocks. All these methods were initially applied
for the gray level images and later were extended to colour images by splitting
the colour image into its respective R, G and B components, individually
enhancing them and concatenating them to yield the enhanced image. All the
above mentioned techniques operate on the image in the spatial domain. The
final method is the DCT where the frequency doma in is used. Here we s c a l e
t h e D C coefficients of the image after DCT has been taken. The DC
coefficient is adjusted as it contains the maximum information. Here, we move
from RGB domain to YCbCr domain for processing and in YCbCr, to adjust
(scale) the DC coefficient, i.e. Y (0, 0). The image is converted from RGB to
YCbCr domain because if the image is enhanced without converting, there is a
good chance that it may yield an undesired output image. The enhancement of
images is done using the log operator.

33
5.3 Image Background Analysis By Blocks

Morphological transformations (Opening by reconstruction, Erosion-

Dilation method) and Block Analysis is used to detect the background of gray
level and colour images . These techniques are first implemented in gray scale
and are then extended to colour images by individually enhancing the colour
components. For aiding better results, the compressed domain (DCT) technique
is used exclusively for colour image enhancement. The major advantage of the
DCT method is that it can be used for any type of illumination. In all the above
methods, the enhancement of the background detected image is done using
Weber‟s law (modified Weber‟s law for compressed domain) a critical analysis
of the various advantages and drawbacks in each method are performed and
ways for overcoming the drawbacks are also suggested. Here, the results of
each technique are illustrated for various backgrounds, majority of the
minimum poor lighting condition. In image acquisition, background detection
is necessary in many applications to get clear and useful information from an
image which may have been picturized in different conditions like poor lighting
or bright lighting, moving or still etc. This Section deals with background
analysis of the image by blocks. In this project, D represent the digital space
under study.

For each analyzed block, maximum ( i M ) and minimum ( i m ) values

are used to determine the background measures. is used to select the
background parameters. Background parameters lie between clear ( )
and dark ( ) intensity levels. If is the dark region then background
parameters takes the maximum intensity levels ( i M ) then is the clear
region, background parameters takes the minimum intensity levels ( i m ).
Enhance images are we get after applying the below equation , and

34
 - Dilation operation, - Erosion operation

Dilation and erosion are the two most common morphological operations
used for back ground analysis by blocks.
5.4 Block Analysis For Gray Level Images

Let f be the original image which is subdivided into number of blocks

with each block is the sub-image of the original image. For each and every
block n, the minimum intensity i m and maximum intensity i M values are
calculated i m and i M values are used to find the background criteria i in the
following way :

is used as a threshold between clear and dark intensity levels.

Based on the value of i , the background parameter is decided for each
analyzed block. Correspondingly the contrast enhancement is expressed as
follows:

It is clear that the background parameter entirely is dependent up on the

background criteria value. For i f , the background parameter takes the
maximum intensity value i M within the analyzed block, and the minimum
intensity value i m otherwise. In order to avoid in determination condition, unit
was added to the logarithmic function.

35
(left half of the image) and output image (right half) for block analysis is shown
below:

Figure 5.6: Block Analysis for Gray level images

5.5 Image Background Analysis Using Opening By Reconstruction

For Gray Scale Images

This method is similar to block analysis in many ways; apart from the
fact that the manipulation is done on the image as a whole rather than
partitioning it into blocks. Firstly minimum Imin (x) and maximum intensity
max I (x) contained in a structuring element (B) of elemental size 3 × 3 is
calculated.

The above obtained values are used to find the background criteria i as
described below

Where min I (x) and max I (x) corresponds to morphological erosion and
dilation respectively,Therefore

36
In this way the contrast operator can be described as in equations By
employing Erosion-Dilation method we obtain a better local analysis of the
image for detecting the background criteria than the previously used method of
Blocks. This is because the structuring element μB permits the analysis of eight
boring pixels at each point in the image. By increasing the size of the
structuring element more pixels will be taken into account for finding the
background criteria . It can be easily visualized that several characteristics that
are not visible at first sight appear in the enhanced images. The trouble with
this method is that morphological erosion or dilation when used with large size
of μ to reveal the background, undesired values maybe generated.

In general it is desirable to filter an image without generating any new

components. The transformation function which enables to eliminate
unnecessary parts without affecting other regions of the image is defined in
mathematical morphology which is termed as transformation by reconstruction.
We go for opening by reconstruction because it restores the original shape of
the objects in the image that remain after erosion as it touches the regional
minima and merges the regional maxima. This particular characteristic allows
the modification of the altitude of regional maxima when the size of the
structuring element increases thereby aiding in detection of the background
criteria as follows:

Where opening by reconstruction is expressed as

It can be observed from the above equation that opening by

reconstruction first erodes the input image and uses it as a marker. Here marker
image is defined because this is the image which contains the starting or seed
locations. For example, here the eroded image can be used as the marker. Then
dilation of the eroded image i.e. marker is performed iteratively until stability

37
is achieved. Image background obtained from the erosion of the opening by
reconstruction Background parameter b(x) is calculated by eroding the above
obtained background criterion τ(x) which is described below:

As it is already mentioned that morphological erosion will generate

unnecessary information when the size of the structuring element is increased,
in this study, the image background was calculated by choosing the size of the
structuring element as unity.
5.6 What Is Median Filtering

Median filtering is a nonlinear process useful in reducing impulsive, or

salt-and-pepper noise. It is also useful in preserving edges in an image while
reducing random noise. Impulsive or salt-and pepper noise can occur due to a
random bit error in a communication channel. In a median filter, a window
slides along the image, and the median intensity value of the pixels within the
window becomes the output intensity of the pixel being processed.For
example, suppose the pixel values within a window are 5,6, 55, 10 and 15, and
the pixel being processed has a value of 55. The output of the median filter an
the current pixel location is 10, which is the median of
the five values.

Like lowpass filtering, median filtering smoothes the image and is thus
useful in reducing noise. Unlike lowpass filtering, median filtering can
preserve discontinuities in a step function and can smooth a few pixels whose
values differ
significantly from their surroundings without affecting the other pixels.

One method that tends to preserve 2-D step discontinuities will is to

filter a 2-D signal along the horizontal direction with a 1-D median filter and
then filter the result along the vertical direction with another 1-D median filter.
This method is called separable median filtering, and is often used in 2-D
median filtering u(n1 ,n2 ) with a 2-D 5 x 5-point median filter.

38
 Figure5.7:Graph For Median Filter

A median filter is a nonlinear system, and therefore many theoretical results on

linear systems are not applicable. For example, the result of separable median
filtering depends on the order in which the 1-D horizontal and vertical median
filters are applied. Despite this difficulty, some theoretical results have been
developed on median filtering. One result states that repeated application of a
1-D median filter to a 1-D sequence eventually leads to a signal called a "root
signal" which is invariant under further applications of the 1-D median filter.
5.7 Out-Range Pixel Smoothing

Like median filtering, out-range pixel smoothing is a nonlinear

operation and is useful in reducing salt-and-pepper noise. In this method, a
window slides along the image, and the average of the pixel values, excluding
the pixel being processed, is obtained. If the difference between the average
and the value of the pixel processed is above some threshold, then the current
pixel value is replaced by the average. Otherwise, the value is not affected.
Because it is difficult to determine the best parameter values in advance, it may
be useful to process an image using several different threshold values and
window sizes and select the best result.

39
5.8 Edge Detection

An edge in an image is a boundary or contour at which a significant

change occurs in some physical aspect of an image, such as the surface
reflectance, illumination or the distances of the visible surfaces from the
viewer. Changes in physical aspects manifest themselves in a variety of ways,
including changes in intensity, color, and texture.

Detecting edges is very useful in a no of contexts. For example in a

typical image understanding task such as object identification, an essential step
is to segment an image into different regions corresponded to different objects
in the scene. Edge detection is the 1 st step in image segmentation.

Another example, is in the development of a low bit rate image coding

system in which we can code only edges. It is well known that an image that
consists of only edges is highly intelligible. The significance of a physical
change in an image depends on the application. An intensity change that would
be classified as an edge in some application might not be considered an edge in
other application.

In object identification system, an object's boundaries may be sufficient

for identify and contours that represent additional details within the object may
not be considered edges. An edge cannot be defined outside of the context of
an application.

40
 CHAPTER 6

THE METHODS OF DETECTION REGION

6.1 Introduction

The methodology that is proposed for the drusen automatic

detection defines all image processing to determine the area affected by drusen,
in order to establish a uniform analysis criterion, although it does not perform
any automatic diagnosis. The retinal images are the only underlined data and
therefore, no definitive diagnosis can be made, since complementary exams to
other health indicators are usually required. Any result produced by an
automatic lesion detection system is an exam that requires interpretation by a
clinician, other exams and the patients‟ biological profile to be validated. As
automatic diagnostic systems do not have this overall view of the patient, they
can always produce false negative and false positive results. The retinal images
used in this work are complex.

These contain anatomical structures that are in its surface, such as the
optic disk, the vessels and lesions when they exist (drusen, exudates,
haemorrhages and others), alongside with the high variability among different
patients‟ images due to the differences in each individual‟s physiology.
However, this methodology concentrates only on the drusen and in particular
on those that affect the patients‟ central vision. The other structures, vessels
and optic disk, are also used in this methodology but as auxiliary indicators for
contrast adjustment and to obtain the image resolution. The automatic detection
and quantification of drusen is described as a methodology composed by five
steps, starting with the image pre-processing and finishing with the
quantification of drusen.

41
6.2 The Methodology

Although there are no strict rules on how a digital image processing

system should be structured, it is usually divided into a sequence of four
operations (Gonzalez and Woods 2007). It starts with a digital format image
capturing, followed by three bottom-up sequences of operations for image pre-
processing, image interpretation and cognitive processing. The image pre-
processing is characterized by low-level image operations where the inputs and
outputs are images, such as image enhancement, noise reduction, geometrical
operations and others, in order to prepare and normalize the image for further
processing. The image interpretation is a mid-level operation which describes
the image contents.

This involves image segmentation, to separate the image in regions or

objects, object characterization and object recognition to identify the structures
in the image. At this level the inputs are images and outputs are attributes
extracted from the structures in the image. The cognitive processing on its turn
is to use the information retrieved and process it accordingly. It can include
analysis of sequences of images, alarms generation and other applications. The
methodology proposed for drusen automatic detection defines all image
processing to determine the area affected by drusen in fundus images. The
choice to analyze the fundus images was that they are the most widely available
and used method to diagnose drusen. The image capture and the cognitive
processing are out of the scope of this methodology. The images are already
supplied in digital format ready to be processed, while the cognitive processing
is the medical diagnosis itself. Following the ophthalmologists
recommendations involved in this work and the Wisconsin grading system
(Klein, Davis et al. 1991), the analysis is focused only on the central vision,
namely the inner macula, i.e., a circular region of interest (ROI) of 3000μm of
diameter centred on the fovea. Within this region, drusen are identified and
quantified, as presented. The proposed methodology covers only the image pre-
processing and interpretation. Since the image acquisition conditions differ

42
between Clinical Centres and equipments, the image pre-processing normalizes
the image for the drusen detection and quantification process. This
normalization will include both image intensity and contrast.

Due to eye lens problems and to deficient patient collaboration during

the image acquisition the illumination over the retina is in many cases non
uniform. The image pre-processing is also responsible for correcting this non
uniform illumination. The image interpretation is composed by the detection of
drusen locations followed by an analytical characterization and finally by the
affected areas calculation. These suboperations include: Drusen detection,
Drusen modelling and Area calculation. For drusen detection a new algorithm
for local maximums detection (Gradient Path Labelling) with improved
accuracy and performance was developed. It retrieves the drusen location and
amplitude, which are required as initial parameters to the modelling step. The
modelling of drusen is obtained with a fitting of modified Gaussian functions
for the detected drusen, allowing its analytical characterization. With the
drusen model, the affected area, as well as other indicators such as the number
of drusen or the overall reflectivity, can be finally quantified. The
parameterization in the automatic image processing systems is one of their
weak points.

If on one side a parameterizable system is capable of being adjusted to

more types of images, on the other side it is less adaptive by requiring pre-
saved configurations for each image type. Also, for the definition of a Gold-
Standard drusen assessment a flexible method is needed, otherwise its status
can be questionable. Having a Gold-Standard definition as a possible objective
to achieve, in this work there was a concern to design a methodology with as
less parameters as possible.

The procedure for setting the base parameterization is in a first phase to

use images graded by the clinicians to provide feedback for the initial
parameters. Then the algorithms are adjusted to produce similar results as do

43
the specialists. As the consensus among the specialists is sometimes weak, as
well as a moderate intra-operator variability, the comparison between the
automatic detection algorithm and all the specialists should have a statistical
approach to allow some degree of uncertainty in the model. A second phase
pursuing the development of the Gold-Standard is to present the automatic
detection of drusen to the specialists in order to assess its results and adjust the
parameters accordingly to achieve a consensus among them. With the final
parameterization defined, the last phase is to validate the results with a blind
study. Images are randomly presented both to the ophthalmologists and to the
software in order to assess the agreement.
6.3 Region of interest

The first operation over the image is to locate the region of interest
(ROI) where the image operations will produce effects and where drusen are to
be detected and quantified. This is done by locating the macula and specifying
a circular ROI around it. The fundus images are usually already oriented and
aligned. However, to verify the alignment, most of the retinographs add an
orientation mark on their top-right side (figure 6.1). With this orientation it is
possible for the clinicians to orientate the image and identify which eye it is. A
right eye has its Optic disk on the same side as the orientation mark, while the
left eye has the opposite.

In the images presented it is noticed a darker area near the centre in 3 to

4 mm to the temporal side of the optic disk, which is the macula. It can also be
identified by the convergence of the retinal vessels in the centre .This latter is
an important reference in images containing advanced macular degeneration,
since the macula may gain some clarity and make difficult its location.

The ROI will be a circular area centred on the macula with a diameter of
two optic disk diameters (ODD). The ODD is used in several studies as a
standard reference to calculate the image resolution. Despite that it is a stable
measure among the humans (Jonas, Gusek et al. 1988), there is no consensus

44
within the scientific community in the reference value, ranging from 1500μm
to 1850μm. In this work a reference value of 1500μm is used according to the
International grading recommendations (Bird, Bressler et al. 1995). Although
there are some works relating to the automatic macula and optic disk location
(Hoover and Goldbaum 2003; Tobin, Chaum et al. 2007), they are not in the
scope of this work. These works use the vessels and optic disc geometric
configuration in order to estimate the exact location of the macula.The work
from Tobin et al. (2007) achieved an accuracy higher than 90%, in a set of
images containing several retinal pathologies. Despite these good results, the
work being presented is focused in drusen detection, leaving for the specialist
the manual detection of the macula, which executes a similar cognitive
procedure as the one showed in the Figure. First, they measure the ODD in
pixels and then mark a circle with two ODD centred on the macula. After
defining the ROI, the image is cropped to a square centred on the ROI with a
width of 1.2 x 2 x ODD, in which the factor 1.2 is the margin used to avoid a
special image.
6.3.1 Colour Channel Selection

From the analysis of the image colour channels, as presented and despite
the fact that the images are predominantly red, it was noticed that for colour
images the green channel is the one providing the best contrast and which is
less affected by the illumination non-uniformity.

which shows the individual colour and grey channels, it is again noticed
that the blue channel has almost no information, the red channel is the one with
higher intensity and with eventual saturation, the grey channel has a low and
non-uniform contrast, and the green channel has a medium intensity and good
contrast. The image histograms confirms these observations.

45
 The grey channel exhibits a good histogram shape, although the image
contrast is not uniform. From the four histograms presented, the green channel
is the more balanced. It has a peak value near the centre, representing the
background, and has clearly defined secondary peaks corresponding to other
structures as the vessels, the optic disc, the drusen and others. The use of the
hue channel from the IHS colour space is also an alternative which has also
been used gray images. However, the inconsistency of the image colour
between the imaging techniques makes it a less reliable approach.

Figure 6.1 Excudate area calculation. (top) contour defined at 20% of

amplitude, (middle) the marked image; (bottom) the areas
values.

46
6.3.2 Non-Uniform Illumination Compensation

In ophthalmologic imaging it is usually difficult to obtain good quality

images. Illumination is not always uniform over the retina due to numerous
factors, such as, the retina convex surface, the lower macula reflectance in
comparison to others parts of the retina, the existence of ocular pathologies
which affect the lens, poor patient collaboration, among other factors. The
compensation of non-uniformity of illumination is perhaps one of the key
operations in this image pre-processing. This illumination effect is responsible
for the appearance of brighter and darker areas and different contrasts among
them. The purpose of this pre-processing operation is to normalize the
illumination, in order to have uniform intensity and contrast patterns along the
retina and especially in the macula. The illumination pattern that focuses on the
retina is not known a priori since it is dependent on various unknown
independent interferences. Therefore, it should be estimated based on the
acquired image. Two approaches for correcting illumination were analysed, one
based on frequency filtering and other based on illumination estimation and
compensation, which will be presented next.

Figure 6.2 Methodology for automatic drusen detection and quantification

Since the image acquisition conditions differ between Clinical Centres

and equipments, the image pre-processing normalizes the image for the drusen

47
detection and quantification process. This normalization will include both
image intensity and contrast. Due to eye lens problems and to deficient patient
collaboration during the image acquisition the illumination over the retina is in
many cases non uniform. The image pre-processing is also responsible for
correcting this non uniform illumination. The image interpretation is composed
by the detection of drusen locations followedby an analytical characterization
and finally by the affected areas calculation.

These suboperations include: Drusen detection, Drusen modelling and

Area calculation. For drusen detection a new algorithm for local maximums
detection (Gradient Path Labelling) with improved accuracy and performance
was developed. It retrieves the drusen location and amplitude, which are
required as initial parameters to the modelling step. The modelling of drusen is
obtained with a fitting of modified Gaussian functions for the detected drusen,
allowing its analytical characterization. With the drusen model, the affected
area, as well as other indicators such as the number of drusen or the overall
reflectivity, can be finally quantified.. Also, for the definition of a Gold-
Standard drusen assessment a flexible method is needed, otherwise its status
can be questionable. Having a Gold-Standard definition as a possible objective
to achieve, in this work there was a concern to design a methodology with as
less parameters as possible.

48
 CHAPTER 7
AREA CALCULATION AND CONUTING
7.1 Introduction

The purpose of thresholding is to extract those pixels from some image

which represent an object (either text or other line image data such as graphs,
maps). Though the information is binary the pixels represent a range of
intensities. Thus the objective of binarization is to mark pixels that belong to
true foreground regions with a single intensity and background regions with
different intensities. Thresholding is the transformation of an input image f to
an output(segmented).Binary image g as follows

g(i,j)=1 for f(i,j)

>=T. g(i,j)=0 for
f(i,j) <T.
Where T is the thresholding, g(i,j)=1 for image elements of objects, and

g(i,j)=0 for image elements of background If objects do not touch each other,
and if their gray-levels are clearly distinct from background gray-levels,
thresholding is suitable segmentation method. Correct threshold selection is
crucial for successful threshold segmentation,this selection can be determined
interactively or it can be the result of some threshold detection method. Only
under very unusual circumstances can threshold be successful using a single
threshold for whole image(global thresholding) since even in very

simple images there are likely to be gray-level variations in objects and

background; this variation may be due to non-uniform lighting, nonuniform
input device parameters or a number of other factors.

49
7.2Drusen Quantification

With the retinal drusen model it is now possible to proceed to the drusen
quantification. In the epidemiological studies which have been done in the past,
the calculated indicators were mainly the number of drusen and the size of the
biggest druse present, but only within a range of four to five possible values.
However, a high variability among the analysis done by different specialists
was observed. In the Wisconsin Grading Center, considered as one of the
reference Retina Image Grading Centres, a variability of 32.9% for drusen size
measurement was reported (Klein, Davis et al. 1991). The quantification
method that is proposed here quantifies the affected area, estimates the drusen
integral and is able to evaluate the number of drusen. The simplest method of
image segmentation is called the thresholding method. This method is based on
a clip-level (or a threshold value) to turn a gray-scale image into a binary
image.

Segmentation using variable thresholds (adaptive thresholding), in

which the threshold value varies over the image as a function of local image
characteristics.
• A global threshold is determined from the whole image f :
T=T (f) .

Local Thresholds are position

dependent T=T(f, ),

This thresholding can be useful, for instance, in microscopic blood cell

segmentations, where a particular gray-level interval represents Cytoplasma,
the background is lighter, and the cell kernel darker. This thresholding
definition can serve as a border detector as well; assuming dark objects on a
light background, some gray-levels between those of objects and background
can be found only in the object borders. Thresholding algorithms For a
thresholding algorithm to be really effective, it should preserve logical and
semantic content. There are two types of thresholding algorithms

50
In global thresholding, a single threshold for all the image pixels is used.
When the pixel values of the components and that of background are fairly
consistent in their respective values over the entire image, global thresholding
could be used. In adaptive thresholding, different threshold values for different
local areas are used. Basic Global Thresholding Simplest of all thresholding
techniques is to partition the image histogram by using a single global
threshold, T Segmentation is then accomplished by scanning the image pixel
by pixel

and labeling each pixel as object or background, depending on whether the

gray level of that pixel is greater or less then the value of T.

Figure 7.1: Detection And Counting Cells Using ROI

This Threshold achieved a clean segmentation by eliminating the

shadows and leaving only the objects themselves. The objects of interest in this
case are darker than the background, so any pixel with a gray level<=T was
labeled black (0) and any pixel with gray level>T was labeled white (255). The
key objective is merely to generate a binary image, so the black-white
relationship could be reversed. Algorithm: Basic Global Thresholding The
following algorithm has been used to obtain the initial threshold T, in the
following section automatically:

51
1. Choose initial estimate for T.
2. Divide the histogram using T. This will produce 2 groups of pixels:
• G1 - Set of all pixels with grey level values >T.
• G2 - Set of pixels with values <=T.

3. Compute the average grey level values m1 and m2 for the pixels in
regions G1 and G2.
4. Compute a new threshold value:

7.3Integral Estimation

The genesis of Drusen deposits is an extra-cellular material

accumulation between the basal lamina of the retinal pigment epithelium and
the inner collagenous layer of Bruch‟s membrane, what causes the retinal
pigment epithelium to be raised with a smooth slope. The materials that cause
these elevations reflect more light than a normal retina appearing in fundus
images with a brighter and yellowish colour. Using the two-dimensional
coordinates -(x,y) - and the image intensity - I(x,y) - from the retinal images it
is possible to estimate a tridimensional view of its surface where elevation of
the spots can be perceived.

Figure7.2: Flow chart of the cells counting

process.

52
Figure7.2 shows the flow chart of the cellscounting process. In data
acquisition, as an input image and it has shown in Figure . These images need
to be enhanced for further analysis. This is a pre-process of an image sequence
before feeding into the segmentation process. Cells segmentation and
extraction is the process to distinguish between red blood cell and other cell in
the blood smear image. The last process is counting the number of cellsusing
Hough transform technique.

Figure 7.3 – Integral quantification procedure.

7.4. Number of Drusen

The number of drusen is used for some epidemiologic studies together

with their main class (small, medium, large size). In this methodology this
value is obtained directly by the GPL algorithm output, i.e., each RIM detected
is considered as a druse. However, this latter is an approximation, given that
the number of GGD functions used in the model is not a reliable measure since
it can be detected more than one GGD functions for each real drusen. Here it is
visible, especially on larger drusen, that there is more than one RIM for each
druse The image normalization is very important as it prepares the image by
standardizing the illumination and contrast for further processing, which can
be, in this work, the drusen detection or can be other kind of analysis such as,
tumours evaluation, diabetic retinopathy detection or others.

53
Any point (x, y) for which f (x, y) > T is called an object point, otherwise the
point is called background point. Thresholding normally results in binary
image and the mathematically; the
operation can be expressed as;

where the pixels labeled 1 is corresponded to object whereas the pixels labeled
0 are corresponding to the background. From the lower pixel image we perform
the morphological area closing to fill the hole and eliminate the unwanted
small pixel. The other image which is higher pixel value has been used as an
input for next processes which are dilation and area closing.

Figure 7.4 – Examples of drusen detection with more than one GGD per
drusen.

on the lower pixel image and higher pixel image. It can be seen that the hole in
the cell has been filled up. From the result after morphological from both lower
and higher pixel, the range value of pixel is determined. The histogram is
suitable to describe where the position of the range value we want in this work.
Histogram of saturation, S image is shown in Figure 6.

54
 CHAPTER 8
RESULT AND DISCUSSION

INPUT IMAGE: This image is consider as input in this project

GRAYSCALE IMAGE: Input image converted as grayscale image by

using i = rgb2gray(rgb) .It will remove the colors level of image

55
MORPHOLOGICAL IMAGES: By using morphological operation it can
perform dilation .Dilation increases and enlarges the width of maximum
regions, so it can remove negative impulsive noises but do little on positives
ones. It is used to reduce objects in the image and known that erosion reduces
the peaks and enlarges the widths of minimum regions, so it can remove
positive noises but affect negative impulsive noises little

REGION OF INTEREST: The image is to locate the region of interest (roi)

where the image operations will produce effects and where drusen are to be
detected. This is done by locating the macula and specifying a circular ROI
around it.

FIGURE 8: OUTPUT IMAGES

56
 CHAPTER 9

Conclusion:

In this project we have proposed a method to detect hard exudates in digital

retinal fundus image which exhibits a sensitivity value of 97.25% and
specificity value of 96.85% after validating the algorithm using 10 images from
database. It provided a review of common segmentation algorithms for retinal
image features. From both number and diversity of algorithms used for
retinopathy detection it was clear that there is no gold standard which solves
entire problem. The segmentation of blood vessels in color retinal images using
Morphological filters. It was found that the appearance of vessels is highly
sensitive in the gray scale image containing only the wavelength of green.
Therefore, for segmentation of vessels was performed using only green channel
of RGB color image. Morphological filter, whose application in the field of
machine vision and automatic object detection, were explored to detect and
enhance vessel features in retinal image. morphological filter provided a better
result.. The morphological operations include dilation, erosion, opening, and
closing. Dilation is a process that thickens objects in a binary image, The result
demonstrates that the proposed algorithm can be used for detection of diabetic
retinopathy and Vessel diameter measurement in their day to day use.

57
REFERENCES:

[1] Diptoneel Kayal, Sreeparna Banerjee, “New Dynamic Thresholding

Based Technique for Detection of Hard Exudates in Digital Retinal Fundus
Image,” 2014 International Conference on Signal Processing and
Integrated Networks (SPIN), pp. 133–162,
[2].M. Usman Akram, Ibaa Jamal, Anam Tariq and Junaid Imtiaz
“Automated Segmentation of Blood Vessels for Detection of Proliferative

Diabetic Retinopathy” Proceedings of the IEEE-EMBS International

Conference China, 2-7 Jan 2013
[3]. Doaa Youssef, Nahed Solouma, Amr El-dib, Mai Mabrouk, and Abo-
Bakr Youssef” New Feature-Based Detection of Blood Vessels and

Exudates in Color Fundus Images”on Image Processing Theory in IEEE 2012

[4]. A. Kumar, “A Segment Based Technique for Detecting Exudate From

Retinal Fundus Image”, International Journal of Computer Science
& Engineering Technology feb , 2-7 Jan 2013

[5] Mukhtar Bizrah, Steve C Dakin, Li Guo1, Farzana Rahman, “A semi-

automated technique for labeling and counting of apoptosing retinal cells,”
Bizrahetal.BMCBioinformatics2014,15:169
http://www.biomedcentral.com/1471-2105/15/169

[6] Jaspreet Kaur ,Dr. H.P.Sinha , “ Automated Detection of Diabetic

Retinopathy Using Fundus Image Analysis,” Jaspreet Kaur et al, / (IJCSIT)
International Journal of Computer Science and Information Technologies, Vol.
3 (4) , 2012,4794 - 4799

[7] Sharath T. ChandraShekar Gomata L. Varanasi, “Region Of Interest

Coding In Medical Images Using Diagnosticall significant Bitplanes,” in
Proc. 2011 18th IEEE Int. Conf. Image Processing (ICIP), Sep. 2011, pp.
1929–1932.

58
 [8] Nasrul Humaimi Mahmood and Muhammad Asraf Mansor, “Red Blood
Cells Estimation Using Hough Transform Technique,” Signal & Image
Processing : An International Journal (SIPIJ) Vol.3, No.2, April 2012

[9] Shafer, G.: A Mathematical Theory of Evidence. Princeton University

Press, Princeton, New Jersey (1976)

[10] Sumeet Chourasiya G Usha Rani, “Automatic Red Blood Cell Counting
using Watershed Segmentation,” Sumeet Chourasiya et al, / (IJCSIT)
International Journal of Computer Science and Information Technologies, Vol.
5 (4) , 2014, 4834-4838

[11] Z. Li, C. Liu, C. Zhao, Y. Cheng, “An Image thresholding Method Based
on Human Visual Perception”, 2009, IEEE.
[12]. D. Kayal, S. Banerjee, “An Approach to Detect Hard Exudates
Using Normalized Cut Image Segmentation Technique in Digital Retinal

Fundus Image”, First International Conference on Signal Processing Image

Processing and Pattern Recognition, 2012, New Delhi, India
[13]. N. Zhu, G. Wang, G. Yang, W. Dai, “A Fast 2D Otsu Algorithm

Based on Improved Histogram”, IEEE Transaction on Image Processing,

2009..

59