Blood Vessel Extraction Of Retinal Fundus Images Based On Scale Space Theory

Retinal blood vessel segmentation plays a very important role in the analysis of retinal images, blood vessel is one of the main physiological structure in retinal fundus images. Various ophthalmologic and cardiovascular diseases such as hipertension, diabetes, can cause serious damage to retinal blood vessel, thus extracting accurate and clear vascular structures can greatly help specialist to perform better diagnosis and treatment. Therefore, the research of retinal blood vessel detection has great practical significance.The retinal vessel appears as a tree structure, present the characteristics of linear structure, and shows long and narrow connected bar model, spread along a certain direction, and the vessel cross-sectional profiles approximate a Gaussian shape. The appearance of retinal vessel is multivariate, includes the change of direction and size,its direction can change from 0 to360 degree and its width can drastically change in 1~ 20 pixels. These all make robust detection of retinal vessel still be a difficult problem. As a typical linear object, its detection method is also great significant for other linear structures in other images.In order to adapt to the change of the blood vessels’ width, this paper studied retinal blood vessels’ enhancement and segmentation based on multi-scale space theory,the main research content includes the following two aspects:(1) This paper studied two image enhancement methods based on multi-scale approaches, i.e. multi-scale Gabor filter and multi-scale top-hat transform. Also we did the experiment comparison based on single scale and multi-scale approaches, we can find the necessity of using multi-scale approaches to enhance retinal images.(2) This paper proposed an effective method for vessel segmentation based on multi-scale Hessian matrix filtering and line detector. First we use the eigenvalues of the Hessian matrix to build blood vessels’ similarity function to enhance blood vessels,then we adopt the improved line operator to extract the vascular feature.At last,vascular detection is realized by using support vector machine.The performance of algorithms is compared and analyzed on two publicly available databases(DRIVE and STARE) of retinal images using a number of measures which include accuracy, sensitivity, specificity and local accuracy. The results show that the method can achieve comparative accuracy and betterperformance on sensitivity compared with other methods, only with a small number of training samples.