Retinal Image Vessel Segmentation Based On Feature Classification

Abstract:In the field of image processing, it is one of the most important technique to research and analysis medical image. With the rapid development of image processing technology, in the aid of graphics knowledge and computer skill, the method of display and analysis for medical image has been improved greatly. Human retinal vessel is the only deeper non-traumatic micro-vascular, whose structure can directly reflect the effect of retinal diseases. However, there exist some challenges to analysis retinal vessel image, such as low contrast between vessels and background, different widths and tortures for retinal vessels, and noise pollution during the collection of retinal images. Therefore, it is important to detect retinal vessels and to establish retinal computer system, which can be significant for clinical diagnosis of retinal diseases.In this paper, the method for retinal image vessel segmentation based on feature recognition is proposed, whose main work can be shown as follows:(1) The preprocessing of retinal vessel images is studied. Analysis of the RGB components of original color retinal image denotes that the red channel is the brightest color channel, while the green channel presents the best vessel-background contrast and the blue channel provides poor dynamic range. Therefore, this paper uses green channel to process in the following steps. In order to obtain an enhanced retinal image, this paper also applies the contrast-limited adaptive histogram equalization (CLAHE) operator to improve the contrast between the retinal vessel and background.(2) A new retinal vessel segmentation method based on feature classification is proposed. Firstly, feature extraction is presented, which consists of retinal vessel enhancement, Gaussian matched filter,2D Gabor wavelet transform, Frangi filter and gradient orientation analysis (GOA). Then, use the error back propagation (BP) neural network to recognize retinal vessels and background. Besides, we evaluate the proposed method by qualitative and quantitative analysis, which includes the comparison of different methods for retinal vessels segmentation, and the computation of the average accuracy, sensitivity and specificity for the segmented methods. Experimental results show that the proposed method has good robustness and performs well on retinal vessels segmentation.