Bayesian tracking for blood vessel detection in retinal images

A new statistical-based tracking method is proposed for the detection of blood vessels in retinal images. Our algorithm adopts a statistic sample scheme to estimate the candidate edge points of local blood vessel. This sampling scheme combines local grey level profile and the vessel geometric properties, which improves the accuracy and robustness of the tracking process. Edge points of blood vessels are detected iteratively based on a Bayesian approach with the Maximum a posteriori (MAP) Probability criterion. Evaluation of our algorithm is presented on both synthetic and real retinal images.