Classification of visualization exudates fundus images results using support vector machine

Abstract

This paper classifies the characteristics of normal and exudates fundus images by determine its accuracy for diagnostic purposes. Image normalization was performed on 149 fundus images (81 normal and 68 exudates) from MESSIDOR databases to standardize the colours in the fundus images. The OD removed fundus image and fundus image with the exudates areas removed. The SVM1 classifier was applied to 30 test fundus images to determine the best optimal parameter. The kernel function settings; linear, polynomial, quadratic and RBF have an effect on the classification results. For SVM1, the best parameter in classifying pixels is linear kernel function. The visualization results using CAC and radar chart are classified using ts accuracy. It has proven to discriminated exudates and non exudates pixels in fundus image using linear kernel function of SVM1 to diagnose DR.

Keywords: Diabetic retinopathy (DR); Optic disc (OD); Support Vector Machine (SVM); AC); Radial Basis Function (RBF).