Retinal Image Vessel Segmentation Based On AdaBoost And Naive Bayes

Retinal blood vessels are an important part of the systemic microcirculation system.The morphological changes of retinal blood vessels are closely related to diseases such as diabetes,hypertension,glaucoma,obesity,arteriosclerosis and retinal artery occlusion.However,the distribution of retinal blood vessels is complex and small,and it is hard for doctors to segment the blood vessels and the number of doctors is limited.Therefore,this thesis studies the algorithm of retinal vascular segmentation,the main work can be shown as follows:(1)Retinal image preprocessing process are studied.By analyzing the retinal images and the monochromatic images of green,red and blue channels,the contrast between the blood vessels and the background of the green channel image is the highest,and thus the green channel of the image is extracted.Using the contrast-limited adaptive histogram equalization operator to further enhance retinal image.In order to reduce the influence of the camera aperture boundary,the region of interest is expanded and the field of view of the fundus image is extracted.(2)Realize the retinal image vascular segmentation method of feature extraction.In order to preserve the original information of the image extracted retinal image green channel image brightness features.Using the prior knowledge of the vascular cross-section approximation to the Gaussian distribution to extract matched filtering features.Identifying vascular structures by studying the eigenvalues of the hessian matrix after convolving the image with a Gaussian kernel,the similarity function of blood vessels is constructed and Frangi filtering is performed.Combined with Morphological Operation and Gaussian Second Order Derivative Filter Filtering Features.Fusion Grayscale Voting and 2D-Gabor Filtering Results.In order to solve the problem of small blood vessels in the mergence interval,the extension at the cross point is generated,the false blood vessel response is generated near the background point of the blood vessel,and the multi-scale linear detection feature is extracted.Symmetric and asymmetric B-COSFIRE filters are configured to extract symmetrical and asymmetric B-COSFIRE filter responses.(3)Proposed a retinal vessel segmentation method based on AdaBoost and Naive Bayes.Retinal target blood vessels and background were classified by AdaBoost and Naive Bayes method respectively.We found that the segmentation result obtained by AdaBoost mainly contains coarse blood vessels.Use Naive Bayes to process retinal images for more small blood vessels.In this thesis,make experiments on DRIVE and STARE database retinal images,in the DRIVE database segmentation sensitivity and specificity were 0.8004,0.9684,respectively,in the STARE database segmentation of the sensitivity and specificity of blood vessels were 0.8130,0.9574,and compared with some classical algorithms experiments show that the proposed algorithm has a high segmentation accuracy,especially for small blood vessels.And the algorithm suitability experiment is carried out.The experiment results show that the proposed algorithm has good robustness.And the use of graphical interface to build a complete automatic segmentation retinal image interface,so that medical staff or researchers can easily operate,play a computer-aided diagnosis.