Registration And Feature Detection For Fundus Images

Fundus abnormalities caused by ocular and systemic diseases can be shown in retina and choroid. Therefore, the examination of the fundus is an applicable tool in diagnosing and monitoring both eye diseases and whole body conditions. With the development of digital image processing techniques, Computer-Aided Fundus Image Processing (CAFIP) system can be utilized to help ophthalmologist on focus location and diagnosis. My thesis is a significant part of CAFIP system, and composed of three kernel parts.1. Image Rectification: Based on the physiological and anatomical characteristics as well as some prior knowledge of the eyeball, a novel systematic image rectification method was proposed to correct the geometric distortion. In addition, gray adjustment method and gray adaptation method were proposed to rectify the gray level distortion caused by nonlinear illumination and edge gray level difference respectively.2. Image Registration: According to the characteristics of retinal fundus images, one novel method of registration based on Phase-Correlation was proposed. This method takes advantage of optic disc to match two images coarsely. Meanwhile Edge Detection was applied to extract the edge of abundant retinal veins and arteries in order to raise the precision of registration. This method also overcomes the gray-sensitive deficiency in Phase-Correlation method. And the experiment results show that the method not only enhances the precision of registration but also reduces time complexity remarkably.3. Feature Detection: Based on characteristics of fundus images, the Active Contour method was improved and used to detect the Optic Disc in the retinal fundus. And because of its low sensitivity to the noise and low contrast of fundus images, the result was satisfactory.