Study And Implementation Of Fundus Mosaic Based On SIFT Algorithm

Since fundus abnormalities caused by ocular and systemic diseases can be shown in fundus blood vessel, fundus examination can early find out eye diseases, which possesses guiding significance for diagnosing both eye diseases and whole body conditions. Considering the limitation of shooting angle, the imaging regions obtained each time always smaller compared to the whole fundus images, it is important in clinic medicine to mosaic lots of fundus images together, with which the doctor can achieve relatively full information of fundus blood vessel. The fundus examination will be more favorable through this projection. Knowing the fact that the software which can achieve nice mosaic effect for fundus images do not exist until now, this study accomplishes a simple system for fundus images mosaic using feature point mosaic method based on SIFT algorithmic.First of all, this paper corrects the radiation distortion. Because of low SNR and contrast of fundus images, this paper analyzes and researches algorithms of image denoising and enhancement, It proposes the use of In selective masking smoothing algorithm to smooth and remove noise,and gray-scale transformation method for retinal image contrast enhancement. Aiming at the still existing noise effect after pretreatment, this paper do not apply traditional method extracting vascular intersection or branch point as feature point, but suggest to use SIFT algorithm which also has good robustness to noise extract image characteristics, then match the SIFT characteristics description of two images. The study puts manually choose similar feature regional forward as auxiliary of RANSAC algorithm for purification matching, and gets transformation relationship of the images, then complete the retinal image registration. Based on the transformation matrix fundus, images will be brought to the same coordinate system. The study also combines two fundus images by weighted smoothing method, and put forward the corresponding solutions for the brightness abnormal problems of the overlapping area on the touchline. It comes to many fundus images mosaics, overlap mosaic strategy is suggested to solve the problem due to patient eye strong movements when imaging causing obtained real image changing and impact splicing results. Using this method we may receive a relatively complete image detail while complete smooth picture overlapping area.After study and analysis in theoretic, the study finishes pretreatment of fundus images, taking and matching of the feature points, and finally achieves mosaic and amalgamation for fundus images based on transform matrix got by images matching. The study accomplishes a fundus images mosaic software including the function modules as said before. It has achieved good experiment results.