Automatic Detection and Elimination of an Optic Disc for Improving Drusen Detection Accuracy

Abstract

Precise localization of optic disc (OD) in color retinal images is an important sub-problem of automated retinal image analysis system. Drusen detection algorithms generally find lot of false positives at the OD region since the intensity of the drusen resembles with that of OD region. The exact identification an OD seems to be difficult as the disc boundaries are not clearly visible. Several parts of the disk is obscured by blood vessels. Further the size and position of the OD varies from one image to another. These factors make the OD detection a challenging task. In this paper we present a novel method to automatically detect the position of OD and eliminate it. The method starts with a color channel selection that provides a better contrast and reduces computational complexity. The contrast and the illumination of the image is normalized using adaptive histogram equalization (AHE) and homomorphic filtering respectively. The OD localization is then achieved by active contour segmentation using morphological operators. The algorithm is evaluated using 30 images. The proposed method was evaluated by comparing with conventional OD detection using a 2D circular Hough transform. The results prove an efficiency of the proposed method with an accuracy of 93%.