Automated detection of anatomical structures in retinal images

An efficient defection of anatomical structures in retinal images is the fundamental step in an automated retinal image analysis system. This paper presents an algorithm for the localization and segmentation of optic disc and fovea. Optic disc localization is achieved by iterative threshold method to identify initial set of candidate regions followed by connected component analysis to locate the actual optic disc. The fovea is identified by finding the darkest region in the image following the priori geometric criteria based on anatomy of human eye. The algorithm is evaluated against a carefully selected database of 50 ocular fundus images. The system achieves a sensitivity of 92% for the optic disc and 94% for the fovea.