Automatic extraction of features from retinal fundus image

Vessel, fovea and optic disk are the three most important features of human retina that are frequently used for retinal image registration, illumination correction as well as for pathology detection inside retina. In this paper, we present a fully automated approach that can detect and localize these organs from retinal fundus image effectively. For vessel detection, we have adopted an exploratory tracing algorithm that has employed directional templates to trace the vessels. After that, we have employed a novel method that utilizes circular matched filter to compute cross-correlation to detect and localize the optic disk and fovea accurately. Since the circular matched filter cross-correlates with a pre-computed ROI, it reduces the computational cost for matching significantly. The proposed method dynamically approximates the diameter of optic disk and fovea regions, and eventually approximates the shapes of these organs as well. Extensive results of our experiment show that the proposed method is effective and encouraging.