Optic disc detection using geometric properties and GVF snake

Abstract

The optic disc (OD) segmentation in an retinal image is prerequisite for an computerized detection of diabetic retinopathy and also for monitoring changes due to diseases such as glaucoma. The OD segmentation is also used for the detection of other anatomical structures like fovea and vascular tree. Many algorithms based on thresholding, active contour model, GVF snake and clustering have been proposed for the segmentation of OD. In this study, a novel method is proposed for optic disc segmentation. The method makes use of P-Tile thresholding for detecting patch of OD. Connected component analysis is performed for eliminating false positives. This step yields initial patch of optic disc for which centroid correction is performed. GVF snake model is used for finding the contour of OD. The method is robust and effective even in the low contrast images as well as in the presence of other pathological structures like exudates. The experimentation has been done using benchmark retinal image databases, namely, diaretdb0, diaretdb1, DRIVE. The results show accuracy of 98% with diaretdb0, 97% with diaretdb1 and 100% with DRIVE.