Research On Stitching Technology Of Multiple Retinal Images

Retinal fundus images are not only important bases for the diagnosis of glaucoma,macular degeneration and other eye diseases,but also significant guidance issues for the diagnosis and treatment of other system diseases,such as diabetes,?hypertension?arteriosclerosis and so on.However,the limited field of view of most fundus camera or the confocal laser scanning optical microscope restricts the single fundus image to only local area of fundus.Therefore,image mosaic is needed in the fields of ophthalmology to provide panoramic retina images for better diagnosis of diseases.This thesis devotes itself in developing algorithms for mosaicking together multiple fundus images from different views seamlessly.The main work and contribution of this thesis are as follows:First,an improved SURF feature point matching method is proposed to solve the problem of the existing SURF feature point matching method in adaptively obtaining sufficiently larger number of correct matching point pairs.In this algorithm,initial matching point pairs are first obtained by Brute-Force matching of SURF feature points extracted from retinal fundus image,after which the mismatching point pairs are eliminated according to the characteristics of the lines linking matching point pairs and the number correct matching point pairs estimated with an improved strategy based on the spatial order.With this purified subset of matching point pairs,parameters for transformation models of all floating images to the central target one are obtained,with which images taken in different views are co-registered to the one in the central field of view.Experimental results show that this algorithm can not only obtain sufficiently larger number of matching point pairs but also achieve higher correct ratio adaptively and hence can improve the accuracy and success rate of retinal fundus image registration.Second,in order to solve the problem in determining complex overlapping areas in the process of multiple retinal images stitching,a fast stitching method for multiple fundus images is proposed based on maximum fusion and gray mean correction ratio.In this method,all floating images taken at different views are first aligned to the central reference one simultaneously upon performing image registration among them,after which the maximum fusion algorithm is used to obtain a preliminary mosaic image from which the stitching seam is detected with a novel seam detection method.Finally,an innovative gray mean correction ratio algorithm is used to eliminate the stitching seams from the preliminary mosaic image Experimental results show that our method not only can perform better in dealing with the abrupt changes near the splicing seams and in preserving the texture information but also can be implemented faster than existing algorithm.