Vessel segmentation of retinal image based on pixel strength and multi-agent

Accurate vessel segmentation in retinal images prove to be a challenging process due to variations of intensity distribution and small scale of vessels, and development of clinically applicable analysis tools presents even more difficult tasks. Major obstacles include detection of very small vessels. This paper reports on a unique approach to these problems, which defines a simple but effective feature concept called pixel strength, and applies a combination of global and local methods as well as post-processing based on multi-agent. The proposed algorithm was tested using DRIVE database, whose results showed consistent performance improvement over existing solutions we have seen in literature. The effectiveness and robustness of the method were also proved with clinical images. The vessel segmentation results demonstrated its near readiness to be deployed and used for clinical studies of retinopathy.