Fundus Image Segmentation Based On SVM And Template Matching

Fundus image segmentation is always an important problem in the application of medical images ,There are many research achievements about blood vessel segmentation nowadays, But most of the proposed approach deal well with high-contrast and wide vessels, not the low-contrast and narrow ones, However, the latter type of vessels are more challenging and useful in revealing certain systemic disease. Motivated by this goal, From theory and application, This article propose some kind of methods, Extract the blood vessel in fundus respectively step by step. The main work of this text summarized as follows.(1) Proposed an improved median filter algorithm, which retain most of the features in vessel pixel and eliminate most of the noises at the same time. Combining with the tubular feature of vessel, proposed a kind of self-adaptive local threshold algorithm and Extract the vessel's general contour and retain prototypes as many as possible.(2) Use two different kinds of methods, equilibrium background of image, elimination noises, enhance contrast one after another. Make contrast between vessel and background more evident and gray difference between backgrounds much less. Extract two different kinds of features as training sample and train support vector machine.(3) Process classified image based on mathematical morphology, Include noise immunization, hole-filling and slit connection, compare the results which segmented by different algorithm at last.All of the data come from STARE fundus image database which provided by Hoover, The test result shows that the method propose in this article can segment and extract blood vessel effectively.