Segmentation Of Retinal Blood Vessels Based On Data-Driven Markov Chain Monte Carlo

Quantitative and qualitative analysis of retinal blood vessels is an important method for non-invasively diagnosing hypertension, arteriosclerosis and other cardiovascular diseases, and the segmentation and extraction for the retinal blood vessels is a primary task. Most of existing retinal blood vessel segmentation methods are effective only for normal retinal images, but insufficiently effective for pathological retinal images due to their sensitiveness to some factors such as non-uniform illumination and focus of infection. In this paper, a relatively-robust method for segmenting blood vessels is proposed. It segments retinal blood vessels by firstly attempting to use a unified mechanism of top-down and bottom-up in computer vision. Specifically, the proposed method uses reversible jump Markov chain Monte Carlo to search parameter space in the Bayesian statistical framework and obtain the approximate global optimal result, Meanwhile a data-drive technology is used for driving the Markov chain. Experimental results show that our method is robust in segmenting both non-pathological images and pathological images.