Segmentation of retinal blood vessels based on divergence and bot-hat transform

A vessel segmentation algorithm for pathological retina images is proposed. Firstly, the vessel centerlines are extracted by using the divergence of the normalized gradient vector field. Secondly, the main vessels are segmented by a sequence of bot-hat operators with different scales and directions. Thirdly, the skeleton lines of main vessels are generated after a skeletonization procedure. The distances from each extracted vessel pixel to vessel centerline and to skeleton line are compared. The noisy pixels with larger distance to the centerline than to the skeleton are removed. Finally, a repair procedure is performed to regain the pixels at the positions of vascular intersections and bifurcations. Experimental evaluation on the publicly available DRIVE database and STARE database shows that the proposed algorithm shows a global performance improvement not only for pathological retinal images but also for healthy retinal images.