Multiple Threshold Techniques for Feature Extraction of Retina using Back Propagation Neural Network Classifier

Abstract

Diabetic Retinopathy (DR) causes the retina to leak fluid into blood vessels which leads to vision distortion. Cell loss in retina will occur when there is abnormal growth of blood vessels on the surface of the retina and also lead to scarring. The main objective of this paper is to identify the abnormalities of blood vessels in retina that are changing the structure of blood vessels and flow of occlusions and neovascularization. Maximum Entropy Multiple Threshold method is adopted in the segmentation of retina images; Segmentation decomposes an image into regions or objects for example retinal blood vessels, optic nerve head or pathological. Features are extracted from the segmented retina automatically by using classification technique to find the normal and abnormal blood vessels in retina. Back propagation neural network classifier is used to divide the blocks as abnormal and normal blood vessels. This multi-layered threshold technique is used to extract the blood vessels accurately by capturing the morphological properties of retinal image.