Study And Implement Of Computer-Aided Diagnosis System Using Fundus Images

The key technical problems of retinal image processing have been studied in this paper, which involves the extraction of the retinal image, the measurement of the width of retinal blood vessel, the registration of retinal images. Firstly, comparing the performance of retinal vessel segmentation methods, we chose the KNN method .In this way, the precision of the retinal image segmentation can be greatly improved. Secondly, based on the extraction of the retinal image, the measurement of the width of retinal blood vessel is carried out through four steps. The first step is to decide the direction of the blood vessel by taking into account the centerline of the blood vessel. The second step is to construct the matching template. The third step is to present the Gaussian model that can approximately simulate the gay distribution of the blood vessel. The last step is to solve the width of the blood vessel by applying Least Squares method optimize the model parameters. Thirdly, a method based on the blood vessel resemblance is proposed for the retinal image registration after considering the feature of retinal images, the precision, speed and practical application of the registration algorithm. In this method, the registration objective function is developed by using the maximal overlap degree of the retinal blood vessel information and its parameters is resolved by introducing a global optimization algorithm --Genetic Algorithm (GA). Finally, Based on VC++ platform, a preliminary computer-aided diagnosis system based on the fundus images is carried.