Segmentation of Blood Vessels Using Improved Line Detection and Entropy Based Thresholding

Segmented blood vessel from the fundus image provides useful clinical information for the diagnosis and monitoring of eye diseases. In this paper, blood vessels from the green channel of the fundus image are enhanced using a two dimensional matched filter which enhances the contrast of the blood vessel against the background. The contrast enhanced blood vessels are then segmented using line detection algorithms which uses four directional filters. The final segmented vasculature is obtained by integrating the outputs from the directional filters. The proposed algorithm is evaluated using the DRIVE database and better performance is obtained. An accuracy of 0.9488 is achieved using the proposed method against Staal et al [7] method with an accuracy of 0.9442. The effectiveness and simplicity of this method can be used to automate screening for early detection of diabetic retinopathy.