Blood Vessel Segmentation Based On SVM

Blood vessels,as a conduit for blood circulation,are spread throughout all parts of the body,especially there is a large number of arterial blood vessel in retinal.Now,it is found that a lot of diseases can cause changes of blood vessels in retinal,such as leukemia,high blood pressure,coronary heart disease and so on.It is the diameter of vessels in retinal that may be changes in different degrees by those diseases.In order to help doctors to diagnose and cure diseases quickly and accurately through the retinal vascular network,one of the most important task is to segment blood vessels accurately.In this paper,a vascular segmentation based on SVM is studied.First,images are preprocessed by the way of transforming the original color image into gray image of the green channel to enhance the contrast ratio of blood vessel and background preliminarily;Then,gray images are processed by the improved two dimensional Gauss matched filter to enhance the contrast ratio of blood vessel and background further.We use the mean shift algorithm to pre-classify the filtered image,change the classification of pixels into regions.Furthermore,The feature vector consist of the features of the pre-classified image which are extracted by the line detector is used to train SVM classification.Finally,the SVM classifier is used to classify the images,and then do the blood vessel segmentation.In order to solve the problem of low contrast ratio between the medium and small blood vessels and background,we uses a method of Gauss matched filter to enhance the changed gray image.In order to improve the efficiency of the filter,put forward a improved method of Gaussian matched filter.We do convolution of Pixels and the best matching template,optimizing the filtering process,simplifying the calculation amount and improving the contrast ratio between small blood vessels and background.In order to improve the performance of the algorithm’s classification,in this paper,we use the mean shift algorithm to pre-classify the filtered image,then classify the vessel image based on region by SVM classifier.The advantage of the method is that the performance of region classification is more efficient than pixel classification,and it can classify images accurately and quickly.To test the retinal image on drive database,The segmentation effect of the retinal vessel based on SVM was compared with the previous method.It is shown that the proposed method in this paper is more comprehensive and accurate to segment the retinal vascular network,and it can accurate segment some small blood vessels which can not segment by the previous method.