Retinal Vascular Image Enhancement And Detection

Vascular diseases are common diseases threat to human health, vascular diseaseprevention and diagnosis has important implications, but the distribution of blood vesselsin human tissues is complex and intensive, with various shape and form, and low contrastof small vessels. It is not convenient to the doctors to use image to diagnose the disease,therefore, through a variety of image processing methods to suppress the noise anddistinguish different organizations near the blood vessels, we can achieve the vascularenhancement and precise positioning, and provide more reliable diagnostic information forphysicians to clinical diagnosis and treatment. This study mainly include the followingthree aspects:(1) The structure tensor is a detection operator based on the response of the first gradientof structure detection operator, generally applied only to the step edge of the target, andHessian matrix is belong to the second-order differential operator, which can be used forthe detection of ridge-like shape. In this paper, we combine the structure tensor and theHessian matrix to detect the vascluar, and take into account the detection of axis andboundary location, So as to achieve the goal of enhancing vascular. Moreover, throughanalysising diffusion distribution for the gradient vector flow of the axis of vascular, whichcan be used of the detection of vascular centerline, we can reach the detection of vascularridge, combining with the statistics of the structure tensor gradient vector,(2) Blood vessel detection based on the second derivative of Hessian matrix to achievethe detecion of the vasucular axis, generates the pseudo-response of step edge, which ledsto the detection of inaccuracy. In this paper, a nonlinear gauss differential combination isapplied, which convolute the blood vessels on both sides with the first order Gaussianderivate.to get the minimum response of the two edges of the vessel blood, achieving theridge detection of vessel blood. This method combine the Hough transform can be use todetect tubular straight line in the side view of industrial paper in order to achieve thedetection of the amount of paper.(3) Although the traditional regional growth of the vascular segmentation method issimple, it needs artificial implantation of the seed points, and it is also noise-sensitive. Inthis paper, the shortest path vascular tracing algorithm reliance on high-level featureinformation of the vascular, include the estimation of the scale and direction of bloodvessels is raised, it can track vasucular pixels of similar characteristics, and achieve accurate segmentation, It can overcome traditional various features of the vessel trackingmethod, which must rely on the limitations of the artificial set of seed points.This paper through the analysis of the vascular high order differential, Combines thesrtucture tensor and the other enhanced operators to improve vascular enhancementalgorithms, And using Dijkstra algorithms to solve the shortest path method, andautomatically set the seed point, tracking the similar vascular properties, to achieve aextract the segmentation of the blood vessels automatically.Comparing the results of thispaper expermental and the standard databaes, the improvement of performance of theenhancement and segmentation algorithm is verified.