AUTOMATIC DETECTION OF RETINAL MICROANEURYSMS USING HYBRID KERNEL SVM CLASSIFIER

Retinal microaneurysms(MA) are the earliest clinical sign of diabetic retinopathy (DR) disease. DR is diagnosed by inspecting fundus images. Early detection of DR protects patients from losing their vision. In this paper an automatic method for detection of MA is proposed. This system utilizes preprocessing of retinal image to correct for non- uniform illumination and enhance contrast in the area of interest. The detection of MAs using Local Convergence index Features (LCF) and Hybrid kernel Support Vector Machine (HKSVM) classifier tells the probability of being actual MAs. The intensity based features and shape based features are extracted and combined using ensemble classifier positives will be reduced during the classification phase. The MA candidates are extracted and classified using HKSVM Classifier. The proposed method has been evaluated by public databases: Retinopathy Online Challenge (ROC) and e-optha. The efficiency and effectiveness of the method processed has been demonstrated by experimental results and thus proving its potential as a diagnostic tool for DR.