An Automated method of Optic Disc Detection from Retinal Fundus Images

Abstract

Localization of optic disc (OD) and segmentation of its boundary are important in developing systems for automated detection of retinal diseases in its early stage. A method to detect OD based on Fuzzy C Means clustering and ellipse fitting is proposed in this paper. First, morphological operations are performed on fundus image to remove the gaps within OD region which appears due to the presence of optic nerve and major blood vessels, for correct segmentation. Further, a brightest portion of the V component in HSV converted image is used to localize the OD and 250 × 250 size Region of Interest (ROI) matrix is extracted. An unsupervised learning method namely, the Fuzzy C Means (FCM) clustering is used to perform the segmentation of OD from ROI. Boundary points are clearly drawn around OD using ellipse fitting. The method is tested n three standard datasets: MESSIDOR, DRIONS-DB and DIARETDB1. The proposed method achieves an average overlap ratio of 88.08 % for DIARETDB1 images, 90.67 % for DRIONS-DB images and 90.00% for MESSIDOR images, with average accuracy of 99.64%, 99.58% and 99.93% respectively. The comparison with alternative method yielded results that demonstrate the superiority of the proposed algorithm for OD detection.