Segmentation of vessels in retinal images based on directional height statistics.

In this paper we present a fast and simple, yet accurate method for the segmentation of retinal blood vessels. Many diseases of the eye result in the distortions of the vessels. The precise location of the major optic veins may be used for the localization of other anatomical parts, such as the macula and the optic disc. Also, many microaneurysm detection methods consider an additional vessel segmentation step. The proposed method realizes the recognition of vessels through considering cross-sections of the image at different orientations. Peaks on the profiles are localized and their heights are measured. This way, a set of height values are assigned to every pixel of the image. Simple statistics are calculated for every pixel, and combined to construct a vessel score map. We apply a simple thresholding procedure and postprocessing step to obtain a binary vessel mask. The method has been tested on the publicly available DRIVE database, and it proved to be competitive with the state-of-the-art.