Abstract:
Human eye is known as the most complex part of the body. The di erent parts of the eye take
the light in the body and perceive the things around us in a proper color, detail and depth. A
number of ocular diseases exist which can a ect the vision. Retina is the most important part of
eye as it contains the sensors and other components like macula and fovea which are responsible
for proper vision. Digital fundus images are used by ophthalmologists to diagnose di erent retinal
related diseases i.e. diabetic retinopathy, hypertensive retinopathy, glaucoma and AMD etc. The
most common lesions associated with di erent retinal diseases are hemorrhages and exudates.
Computer aided diagnostic (CAD) systems to diagnose ocular diseases can be really helpful for
ophthalmologists in form of saving their time by giving initial diagnostic reports. One issue related
to digital fundus images is occurrence of non-uniform illumination which e ects the results of any
CAD system. This research presents a CAD system to detect retinal abnormalities accurately.
The implemented CAD system consists of novel algorithms for macula, hemorrhage and exudate
detection. It uses illumination equalization as a preprocessing step to cater for non-uniform illu-
mination. The main novelty for exudate detection lies in use of multi-resolution images to detect
exudates of di erent sizes. We have also proposed a new algorithm for macula detection in this
research even in the presence of other lesions. It generates a mathematical model for macula
detection using regression. The proposed system is tested on some publically available databases
and also on one database which is gather locally with the help of Armed Forces Institute of Oph-
thalmology (AFIO). The results are veri ed with the help of ophthalmologists from AFIO and the
results show the signi cant of proposed system.
