Research On Automatic Inspection Method Of Color Vision Camera

The optic disc (OD) is an important component in fundus image and its characteristics (such as shape and size) are auxiliary parameters in clinical diagnosis. Therefore, automatic detection of OD is of great significance for computer-aided diagnosis of the ophthalmology diseases.In this thesis, based on the characteristics of fundus structure, the OD detection method in color fundus images is proposed. Firstly, vein vessels in fundus image are extracted based on the bot-hat transformation, and then the extracted vein vessels are fitted by a parabola based on the least squares fitting method. Secondly, the preliminary localization of OD is obtained fusing characteristics of OD and vein vessels. Thirdly, accurate localization of OD is obtained through phase congruency and window-scanning method. Finally, iterative Otsu method and Hough transformation are performed on the located images, and the optic disc segmentation result is obtained. The proposed method is mainly based on characteristics of local vessels instead of precise characteristics of global blood vessels, which can overcome the limitations in traditional methods.The proposed method is tested on four public fundus image datasets (DRIVE, DIABETEDO, STARE and MESSIDOR), and the test results are 100%,98.6%,93.8%, 99.75% respectively, which proves the accuracy and robustness of the proposed method.