The Research Of Image Registration Based On Manifold Learning

Image registration is a important aspect of image processing and it is one of the technologies which has been developed rapidly.In recent years,with the development of computer visualization,image registration applications are widely utilized in many fields.Image registration can be divided into two categories,one category is intensity-based image registration technique and the other one is feature-based image registration technique.They have been applied in medical diagnosis,patient longitudinal tracking,satellite image processing,panoramic image stitching and intelligent robots.The latest technologies of large deformation are analyzed in depth,which has made progress in this field.We find that the large deformation in the image information and topology structure has a greater change and the current algorithms cannot maintain the consistency between local and global manifolds,which results in asymmetric energy convergence and so on.Considering the characteristics of large deformation image registration.we study on the manifold learning technique and diffeomorphic registration model and propose a model based on manifold method(Manifold registration of large deformation,MRL).The new model aims to reflect essential information of local and global in the registration process at the same time,the algorithm can greatly reduce the time consumption in large deformation image registration.The experiment results show that the proposed algorithm is effective and feasible.