The Study On Detection Method Of Pathological Changes In Fundus Image

Retinal vessel behavior analysis of trend, bending and bifurcation has became an important means in medical diagnosis of systemic vascular diseases. Because of most collected fundus images has light uneven phenomenon, traditional vessel segmentation methods is difficult to detect the tiny blood vessels. With the development of medical image processing, using computer to observe and check ocular fundus disease can help doctors to improve the efficiency and accuracy of disease diagnosis. As the basis of ocular fundus disease detection, segmenting retinal blood vessels accurately is the premise of whether the system can check ocular fundus diseases correctly or not. However, due to the defects of the fundus camera optical design, a lot of background noise and uneven illumination phenomenon occurs in the collected images. The traditional method of vessel segmentation usually lose the tiny blood vessels, and can hardly ensure whole blood vessel’s correct segmentation, bringing bad affects in segmentation of ocular fundus diseases. Therefore, how to ensure the correct segmentation of large blood vessel, and at the same time correctly segment tiny blood vessels as far as possible, is the key to improve the ability of disease detection in ocular fundus image.This topic researches and improves segmentation method of retinal blood vessels in fundus images, using improved method to segment retinal vessels in pathological changed fundus image, so as to detect background region lesions and vascular region lesions. According to the high recognition ability to segment structure of the Hessian matrix, improved method firstly constructs a multi-scale Hessian filter by using the Gauss kernel function to enhance vascular pixel contrast. Then build the scale space, using the improved Top-hat transform to segment enhanced retinal blood vessels. Finally use morphological reconstruction to remove isolated noise points in the segmentation result image. The main research contents are embodied in the following aspects:(1) This paper studies the traditional segmentation of retinal blood vessels, proving feasibility of the application of Hessian matrix, and analyses the influence of spatial scale factor for Hessian segmentation and enhancement. Discuss the segmentation process theory and multiscale morphological segmentation; analyze advantages and disadvantages of using scale space morphological segmentation.(2) The key of Hessian enhancement is to construct vascular similarity function. Considering fundus images have characteristics like uneven illumination background and the fluctuation of pixels gray value, this topic improves the original vascular similarity function of Hessian matrix, leading to better adaptive ability and vascular enhancement effect, at the same time smooth fundus image to reduce the influence of noise.(3) Improving the traditional Top-hat transform, introducing the scale space, can realize preliminary segmentation of retinal blood vessels based on Top-hat scale space transform. Scale space according to the scope of the vessel diameter, select a series of size of structure element, using each structure element of a top hat segmentation operation, generate a plurality of result image, and finally take maximum gray of each image at all pixels.(4) Improving segmentation results image quality by morphological reconstruction operation. Combining the scale space and morphological reconstruction operation, using multi scale reconstruction to revise the preliminary segmented retinal blood vessel image, can stand out vascular pixel, eliminate false edges and isolated noise points.