Research Of The Techniques Of Retinal Vessel Segmentation Based On Feature Extraction And Supervised Learning

Retinal vascular les ions can reflect a lot of systemic disease. It is of great significance that automatic detection and analysis of the vasculature can assist in the implementation of screening programs for retinal disease. Retinal vessel segmentation algorithms are a fundamental component of automatic retinal disease screening systems, which directly affects the performance of the system. Study on the retinal vessel segmentation method, which can satisfy the clinical examination requirement of accuracy, objectivity and repeatability, has great theoretical value and practical significance. Under this background, combined with the structural characteristics of retinal vessels, the retinal vessel segmentation method based on feature extraction and supervised learning is studied in this dissertation.Firstly, the green channel of retinal image is selected for subsequent image processing through the analysis of its RGB channel. We compared two image enhancement methods, the results show that the CLAHE method is better. In order to reduce the effects of image noise and invalid interference information, the area of ROI is expanded, and a fundus FOV extraction method is designed based on YIQ color space model.Secondly, on the basis of biological visual receptive field, a feature extraction method is studied based on the COSFIRE filter model. A feature vector library is composed of two extracted features by the method and one feature from green channel.Finally, on the basis of feature extraction, we discussed and analyzed two supervised learning algorithm about simple parameter tuning in order to segment retinal vessel. One is the K-Nearest Neighbor algorithm, the other is the Bayes ian Gaussian mixture model. Meanwhile, the proposed method and other existing methods are compared, the results show that effectiveness of the proposed method.