Retinal Registration Based On Bifurcation Structure And Thin-plate Spline Function Transformation

Fundus retinal image has become an important diagnostic basis for a variety of clinical diseases,and retinal blood vessel is the unique method to observe the physical form of blood vessels for non-invasive,However,the fundus images obtained by the fundus camera can only be local rather than global,therefore,it is necessary to obtained the global retinal image by registration.At present,there are many fundus image registration methods,and each has advantages and disadvantages,so these methods can not be widely applied to various fundus images.Therefore,the research of automatic registration for retinal images is still a challenging subject.In this paper,we propose a registration strategy based on improved bifurcation and thin-plate spline function transformation.And improving the process of selecting some bifurcation structures points by the range information,The slope information and the information of the bifurcation structure's own concentration information.In addition,the methods of fundus blood vessel segmentation of are explored in this paper.After comparison and analysis,we know the topological vascular tree segmentation method is more effective.The main process of registration of retinal image includes retinal image preprocessing,retinal vascular segmentation,skeleton extraction,feature extraction and registration.The improvement of registration strategy is the most important part of this paper.The experiments of the proposed method are implemented on the public available STARE(the Structured Analysis of Retina)and DRIVE(Digital Retinal Images for Vessel Extraction)database.From the experimental results,we can see that the method of blood vessel segmentation in this paper can get more accurate vessel,and the Thin plate spline function registration strategy also reducing partial deviation problem of the previous bifurcation method,and it is better than the other existing methods such as SIFT feature registration and manual splicing method.It is more applicable and more robust than previous registration methods.