Study On Image Segmentation Of Angiography

Diabetes, hypertension, arteriosclerosis and other serious impact on human health of the disease, often accompanied by the generation of new blood vessels. The correct segmentation of vascular images and the analysis of the changes of blood vessels play a significant role in the prevention and diagnosis of the disease. Although existing blood vessels segmentation method has achieved good results, but the vessel segmentation problem is far from resolved. A segmentation algorithm can be used in most vascular images, and also can get good result have yet not to be brought out.This paper do a rough classification for the existing vascular segmentation algorithm, that mainly divided into two categories: vessel segmentation method based on unsupervised learning and vessel segmentation method based on supervised learning. While based on unsupervised learning method for blood vessel segmentation can be divided into the one based on the matching filter, segmentation method based on morphological analysis of blood vessels, segmentation method based on vessel tracking vessel, segmentation method based on a model of vascular, segmentation method based on scale analysis. Through a brief discussion of the above segmentation algorithms, this paper discusses and compares the advantages and disadvantages of various algorithms.In view of the different blood vessel image, the paper carries on the research of the different segmentation algorithm:(1) In this paper, a new blood vessel extraction algorithm based on frequency domain is proposed to solve the fuzzy vascular image, the method avoids the tedious operation process that removing blur first and then segmenting, it directly segments the fuzzy blood vessel image, Firstly, the improved gradient domain is used to enhance the dynamic compression of the blood vessel edge, which highlights the small vessel; Then the non blind deconvolution based on sparse prior proposed by Horacio et al. is used to restore the divergence image; Then the median filter is used to reduce noise and reduce the impact of noise in process of the segmentation; Finally, using the method of maximum between-cluster variance to segment the blood vessel image and get the final blood vessel. This method has a better effect on the relative background image of the blood vessel.(2) In order to find a new and effective retinal segmentation algorithm, this paper presents a new algorithm for the segmentation of blood vessels based on principal component analysis. This method takes advantage of the new segmentation method to achieve better segmentation results, First of all, we use four methods,such as morphological transform, analysis of blood vessel direction based on line orientation, analysis of blood vessel curvature based on gradient calculation, vascular ridge, to extract eight feature space. Then, PCA was used to red uce the dimension of the eight feature maps, and the main blood vessel features were extracted. Finally, the OTSU method is used to segment the reduced dimension image to get the final segmentation result of blood vessel. In this paper, the experiment was carried out on the international DRIVE fundus image database, and the performance evaluation was carried out from three aspects of accuracy, sensitivity and specificity, the experimental results show that the method has high sensitivity.