Automated detection of exudates using histogram analysis for Digital Retinal Images

Abstract

Diabetic retinopathy is one of the serious vision-threating diabetes complications which can even lead to blindness if not diagnosed and cured at the early stage. Exudate is one of the early sign of diabetic retinopathy. Hence, proper method for exudates detection is very useful step in the screening of computer-aided diabetic retinopathy systems. This paper proposes an automatic algorithm to localize the optic disc and detect exudates from fundus retinal images. The algorithm uses histogram information. Optic disc occurs as the bright yellowish region in the retina image, which resembles exudates. It must be eliminated before exudates detection. Therefore, it is necessary to localize the center of the optic disc before it can be eliminated. Without the presence of the optic disc, exudates can be detected easily by using the histogram based thresholding technique. The DRIVE, DIARECTDB1, STARE, and 325 retinal fundus images from local dataset are examined for evaluating the proposed algorithm. The proposed method achieved the accuracy of 100%, 96%, 93%, and 97% respectively for optic disc detection. For exudates detection, the accurate rate is 99%, 90%, and 89% for DRIVE, DIARECTDB1 and local dataset respectively.