Abstract
Retinal images contain forests of mutually intersectingand overlapping venous and arterial vascular trees. The ge-ometry of these trees shows adaptation to vascular diseasesincluding diabetes, stroke and hypertension. Segmentationof the retinal vascular network is complicated by incon-sistent vessel contrast, fuzzy edges, variable image quality,media opacities, complex intersections and overlaps. Thispaper presents a Bayesian approach to resolving the con-figuration of vascular junctions to correctly construct thevascular trees. A probabilistic model of vascular joints (ter-minals, bridges and bifurcations) and their configuration injunctions is built, and Maximum A Posteriori (MAP) estimation used to select most likely configurations. The model is built using a reference set of 3010 joints extracted from theDRIVE public domain vascular segmentation dataset, andevaluated on 3435 joints from the DRIVE test set, demonstrating an accuracy of 95.2%. Keywords: Retinal vessels configuration, vessels connectivity, junction resolution, vessels trees reconstruction.