Abstract

Presence of long term diabetes affects different body organs, one of the dominant effect it causes on, is retina of human eye ,called Diabetic Retinopathy(DR).DR progresses from mild Non-Proliferative DR too Severe Proliferative DR- leading to loss of vision. The syndromes it causes on retina are treatable if diagnosed in time. Earlier indications occur, due to leakage in retinal capillary, forming red deposits on retina, termed Microaneurysms. The occurrence of microaneurysms counts in diabetic retinopathy and its close correlation to the gravity of the disease is well noted. As a result, identification of Microaneurysms is must; to avoid the further impairments. A novel three stage approach for MA detection is proposed in this paper. Pre-processing is done using advanced - Adaptively Clipped – CLAHE and Directional feature enhancement using Gabor Filter, Candidate region segmentation is performed using Single Optimal Thresholding. Blood vessels are extracted and removed using cascade of morphological operations and line detectors. Further, with feature vector extraction and One Class-SVM, candidate regions are classified as MA’s and outliers. The proposed method is tested on images from publicly available DiaretDB1 dataset and accomplished the results compatible to the state-of-the-art methods.