Reseach On The Segmentation Algorithm For Retinal Image

Retinal image processing and analysis is an important application of computer technology in medical field.The morphological structure of fundus retinal tissue is highly related with certain diseases of mankind,such as glaucoma,diabetes,hypertension,arteriosclerosis,the other fundus diseases and cardiovascular diseases.The variation can be used as an important basis for disease diagnosis,as well as an important basis for judging the severity of a disease.Hence,the use of image processing technology to automaticly segment the main structure of a retina image is very important in researching and analyzing medical pathology and large-scale disease screening.In this dissertation,the vascular,optic disc and other tissue structures in retinal image are selected as the research object.The study is conducted from the enhancement,localization and segmentation of retinal vascular and optic disc.The idea of morphological operation,multi-scale technology,Gabor wavelet transform and active contour model snakes are introduced in.this dissertation,these techniques are used to segment the retinal vessels and optic disc.Specifically,the main work and contributions of this dissertation include.(1)A method of retinal vessel segmentation based on multi-scale morphological operation and seed point tracking is proposed.Aiming at the problem of the change of retinal vascular caliber,a multi-scale method to enhance the vascular diameter with different calibers by using morphological operation with different scale structure elements is proposed,the linear structure elements,which can eliminate the influence of large area of hard hemorrhage and hard infiltration of macula and optic disc,is used for this method;to solve the problem of the effects of hard hemorrhage and hard osmosis on vascular segmentation,according to the vascular connectivity,the method of point tracking is proposed to eliminate the isolated noise,retain the blood vessels and improve the accuracy of segmentation.(2)A method of retinal vascular segmentation based on local enhancement and regional growth is proposed In the process of retinal image acquisition,the contrast of different parts caused by uneven illumination causes the contrast between the microvessels and the surrounding tissues to be low.Aiming at this problem,a local contrast normalization method is proposed,which adjusts the contrast between the target and the background only in the local area around the pixel point,the enhancement of retinal blood vessels adopts the directional Gabor wavelet transform,and then the regional growth method is used to track the blood vessels to remove the isolated noise.The algorithm improves the segmentation accuracy of the retinal vessels and reduces the error of the fine blood vessels in the lower contrast region.(3)The location and segmentation algorithm of optic disc based on confidence and Snakes model is proposed,which include three steps:coarse location,optic disc segmentation and center locating.Am at the problem of the large calculation capacity and slow running speed,a fast locating method based on confidenceis proposed,which includes two steps:vertical coarse locating and horizontal coarse locating.According to the symmetry of the structure of retinal vascular tissue,the coarse location of vertical direction is proposed.The basic features of the optic disc are:high in density of blood vessel pixel,high in value of pixels and large in the contrast between the background and the corresponding blood vessels.According to these features,the horizontal coarse positioning method based on confidence coefficient is proposed.In different images,the boundaries of the optic disc are smooth and the shapes differ,am at this problem,a method of optic disc segmentation based on the Snakes model is proposed.With the combination of internal force and external force,the Snakes model can ensure the smoothness and stability of the optic disc and improve the accuracy of optic disc segmentation.Finally,the center point of the optic disc is redefined according to the optic disc boundary.