Adaptive pre-filtering for HRT retinal vessel detection in images

Accurate detection of blood vessels and the measurement of vessel diameter can be used in the analysis of disease processes of the retina. In retinal HRT images two distinct categories of blood vessels exist: those that give the appearance of being 'hollow' and those which do not. In this paper, we propose a new pre-processing step and an associated set of edge detection templates for the accurate identification of 'hollow' blood vessels. Our pre-processing step filters images using an adaptive LoG kernel, matched to the cross-section profile of the vessel, which enhances the edges of the blood vessel and strengthens the contrast between the edge boundaries, the background image and the central hollow part of the vessels. A specifically designed set of templates in 24 directions is then used to detect the position of the edges of the blood vessels and to trace their centre lines.