Towards Understanding Color Ocular Fundus Images

Abstract

We have studied some fundamental problems on the way of understanding color ocular fundus images by computer. They are the extraction of blood vessels from the retinal background and labeling them arteries and veins. We analyze the chromatic characteristics of color ocular fundus images to get the signal level knowledge on them which is used in the stages of extracting blood vessels and most reliable labeling of initial line segments. We propose a dynamic threshold selection scheme for binarization of gray images that have an amount of shadings or uneveness of gray values. In the labeling stage we use a deterministic procedure which takes into account the physical level knowledge on blood vessels. Some other methods are also discussed.