Vessel Intensity Profile Uniformity Improvement for Retinal Vessel Segmentation

Abstract

This paper presents a novel approach for image enhancement. In the proposed method, retinal image is divided into three sub-parts, optic disc region, vessels region and non-vessels region (black region outside vessels). Color image enhancement is then performed on these three regions in HSV domain. Then the color image is converted to gray using a multi-scale line detector based technique. For vessel detection, a modified Frangi vessels detector is used. This filter produces a nearly uniform response in every vascular structure and precisely boosts the border between the background and the vascular structures. This filter accomplished the best result in segmentation quantitative metrics performance, vessel enhancement signal-to-noise ratio and uniformity of the response in vascular structures as compared to other filters. Excremental results show that the performance of modified Frangi method is much better than the old Frangi and also comparable with most of the state-of-the-art methods.