Extracting Vessel Centerlines From Retinal Images Using Topographical Properties and Directional Filters

In this paper we consider the problem of blood vessel segmentation in retinal images. After enhancing the retinal image we use green channel of images for segmentation as it provides better discrimination between vessels and background. We consider the negative of retinal green channel image as a topographical surface and extract ridge points on this surface. The points with this property are located on the centerline of vessels. In presence of noise and non-uniform illumination the extracted ridge points appear as separated points which consist parts of vessel centerline. In order to connect separated ridge points and extending them for thin vessel extraction, we introduce a bank of directional filters to determine proper direction for extending the ridge end points. The ridge end points grow to provide link between separated parts of centerline using the introduced procedure. The result of experiment on images in the DRIVE database shows the proposed method outperforms the existing methods. Performance of the proposed method was evaluated based on accuracy, false positive and false negative criteria.