Study On Vascular Segmentation Method Of Colored Fundus Image

The retinal vascular network is the only deep blood vessel that can be observed directly with non-invasive means in the human body.The changes of characteristics or morphology in the microvasculature can be leaded by any pathological changes of systemic and hematological characteristics.The segmentation of retinal blood vessels is a key step in retinal image processing and analysis,it has a good research value for the early prevention and diagnosis of systemic and hematological diseases.The characteristics of retinal images are complex,the automatic segmentation of retinal blood vessels is easy to be affected by external conditions and pathological changes,and in the retinal images,the difficulty of segmentation is increased by the reason that contrast ratio of tiny vessels and its background is low,therefore,to improve the segmentation accuracy is an important topic to research.The background and significance of retinal vascular and the structural characteristics of eyeball and retina have been introduced in this thesis,and then the present situation of the research on retinal vascular image processing at domestic and abroad have been described,the characteristics and difficulties of blood vessel image have been analyzed in the end,the retinal vascular image denoising and segmentation mainly be focused.In this paper,DRIVE standard image library and STARE standard image library of color fundus image are used for simulation experiment.The main contents are as follows:(1)A bilateral filtering method for image denoising is improved based on non-local mean filter.Works mainly on retinal vascular images denoising has been done,the bilateral filtering and non-local mean filtering has been deep analyzed,and the advantages and disadvantages of the two filtering methods has been summarized,then a new method based on bilateral filtering is improved.Meanwhile,the Integral graph method and raised cosine function which approximating gray similarity function are used for implementing non-local mean filtering and bilateral filtering respectively.The simulation results show that the improved method has a better performance on denoising and consumes less time on operation.(2)A color fundus segmentation method based on level set function is studied.By analyzing the fundus segmentation method,in this paper,a color fundus segmentation method based on horizontal set function is studied.Firstly,adaptive histogram and two-dimensional Gabor transform were used to preprocess the fundus image.Then,the horizontal set theory was defined by using the variable region fitting energy definition.(3)An algorithm based on generalized linear model for color fundus image segmentation is improved.According to the characteristics of fundus blood vessels,an algorithm based on generalized linear model is improved.Firstly,this improved algorithm enhanced retinal image application for adaptive histogram equalization method.Secondly,using two-dimensional Gabor wavelet multi-scale to transfer fundus image.Finally,the generalized linear model(GLM)classifier is used to vessel segment the fundus images.