Localisation of optic disc in fundus images by using clustering and histogram techniques

Optic Disc (OD) is considered as one of the main features of a retinal fundus image. Segmenting the OD can be used for automatic extraction of anatomical structures. The change in the shape, color or depth of optic disc is an indicator of various ophthalmic pathologies. Before segmenting the optic disc, the position of OD has to be found. This paper proposes a new method for OD localization based on clustering and histogram approaches. In this method, first the candidate regions are determined by clustering the brightest pixels in red plane of the fundus image. Then three OD candidate pixels are determined within the candidate region of green plane by using three independent methods namely maximum difference method, maximum variance method and Gaussian low pass filtering method. Three sub images with center of these three OD candidate pixels are selected. Histogram of each of the sub images is found. Finally center of OD is located by selecting a sub image with large number of brighter pixel in blue plane. The algorithm is validated with MESSIDOR database which shows that the proposed technique can locate the optic disc even in blurred images.