Segmentation of Retinal Blood Vessel by Merging Shape, Region and Edge Information

Automatic extraction of retinal blood vessels is significant for clinical application. In this paper, we first us Gabor filters to enhance blood vessels by exploiting their shape feature. It is then embedded in an improved local binary fitting model as a region force. Simultaneously we broaden the definition of the distance function by adding a gray-level similarity constraint. Finally, an edge based energy term using the distribution features of gradient vector flow is proposed, which tends to smoothen the vessel boundaries with the potential regularity of gradient diffusion. Experimental results show that the proposed method can extract the main retinal blood vessels and small blood vessels with high accuracy.