Retinal vessel extraction using a piecewise Gaussian scaled model

Automated analysis of retinal images usually requires estimating the positions and appearance of blood vessels, which contain important features for abnormality detection. Although there is a wide literature on detecting vessel positions from retinal images by modelling cross-sectional profiles, little attention has been given to extracting vessel appearance in intensity. In this paper, we introduce a piecewise Gaussian scaled model to characterise the intensity distributions of vessel cross-sections. Based on a newly developed vessel detection scheme, we describe the use of the proposed model for extracting vessel appearance. The preliminary experimental results obtained from angiographic pairs and images of an SLO sequence are reported.