Probability map based localization of optic disk

Fundus imaging is one of themost commonly used modalities for examining retinal structures. The images are mostly analyzed by physicians. These diagnoses are subjective with a high rate of inter-observer variability, therefore an automated or a computer aided diagnosis is needed to speed up the monotonous screening process and to provide objective measurements. In this article we present an automatic algorithm for localization of the optic disk center in fundus images. The application field of the algorithm is preprocessing for further algorithms such as segmentation of the optic disk boundary, vessel tracking etc. The algorithm is evaluated using the DRIVE database and our own public high resolution image database. The tests show a localization error less than 0.2 optic disk diameter (ODD) in case of the high resolution database, and less than 0.35 ODD in case of the lower resolution DRIVE database.