Detection and classification of exudates using k-means clustering in color retinal images

Abstract—Detection and classification of exudates using k-means clustering in color retinal images. Diabetic retinopathy (DR) is one of the < Final Year Projects > leading causes of blindness in the world among patients suffering from diabetes. It is an ocular disease and progressive by nature. It is characterized by many pathologies, namely microaneurysms, hard exudates, soft exudates, hemorrhages, etc, among them presence of exudates is the prominent sign of non-proliferative DR. Both hard and soft exudates play a vital role in grading DR into different stages. In this paper, we present an efficient method to identify and classify the exudates as hard and soft exudates. The retinal image in CIELAB color space is pre-processed to eliminate noise. Next, blood vessels network is eliminated to facilitate detection and elimination of optic disc. Optic disc is eliminated using Hough transform technique. The candidate exudates are then detected using k-means clustering technique. Finally, the exudates are classified as hard and soft exudates based on their edge energy and threshold. The proposed method has yielded encouraging results.