Detection and Classification of Hard Exudates in Human Retinal Fundus Images Using Clustering and Random Forest Methods

Diabetic Retinopathy (DR) is a vascular disorder where the retina is damaged because fluid leaks from blood vessels into the retina. One of the primary lesions of diabetic retinopathy is exudates, which appear on retinal images as bright patches with various borders. In this work an image processing framework is presented to automatically detect and classify the presence of hard exudates in the human retinal fundus images. A total of 50 images have been used to detect the hard exudates from the Messidor database. Digital image processing methods help to extract the location and level of abnormalities in retinal fundus images. The contrast adaptive histogram equalization is used for preprocessing stage and Fuzzy CMeans (FCM) and K-means clustering algorithms are applied to segment the exudates in abnormal images. A set of features such as the standard deviation, mean, energy, entropy and homogeneity of the segmented regions are extracted and fed as inputs into random forest (RF) classification to discriminate between the normal and pathological image. The proposed method achieved 92.94% accuracy for early detection of DR.