A Volumetric Shape Registration Based On Graph Matching

Three-dimensional medical image registration plays an important part in the field of medical image processing.It is widely used in medical diagnosis,surgery,virtual surgery guidance,evaluation of therapeutic effects and other aspects.The registration of a three-dimensional medical image can be described as follows: for a given medical image,one or a series of spatial transformations could be found through a certain algorithm,so that the corresponding anatomical points on another medical image are consistent in space.At present,the point registration and surface registration have been developed very maturely.However,there is little research on volumetric registration.The study of volumetric registration is of great research value.Compared with point registration and surface registration,volumetric registration can better capture the volumetric information of the 3D solid shapes and the research is more challenging.In order to improve the efficiency of volumetric registration,both local and global features of the 3D solid shapes are considered,and a method based on a locally affine-invariant constraint for volumetric registration is presented.A layered matching pipeline algorithm based on the mixed corresponding grid dense matching is studied.Considering the robustness and efficiency in mesh matching which may has noise,a layered matching pipeline based on the mixed corresponding grid dense matching algorithm is presented.The aim is to realize a dense matching between two three dimension meshes.Firstly,this algorithm finds an intrinsic map between two non-isometric,genus zero surfaces.Secondly,it uses the nature of the bottom of the measure preserving distance to release the dense corresponding of surfaces.A set of experimental results show that the proposed algorithm achieves better approximation accuracy than the other algorithms.A hypergraph matching method based on the blended tensor is studied.Because of unsuitable feature descriptor and the deformation of the feature points' positions caused by different perspectives,it is difficult to find the optimal correspondence between the two 3D grid models.Considering these problems,the matching problem between the 3D grid models can be well formulated into graph matching problem and a hypergraph matching method based on the blended tensor is proposed.Firstly,the set of feature points from the models areextracted.Secondly,the associated hyper-graph according to the feature point sets is constrcuted.Finally,we do a process of affinity-preserving random walks on the associated hypergraph to sort the nodes in the hyper-graph is processed and the match pairs are obtained.Compared with the other classical graph matching methods,the proposed method can get a more accurate one-to-one mapping between the nodes of two hyper-graphs.Based on dense matching algorithm and hyper-graph matching method,a medical volumetric shape registration with a locally affine-invariant constraint is proposed.This method introduces a novel local affine-invariant to enhance the constraint of the registration process,allowing the volumetric shape registration system to iterate in the correct registration direction.The local affine-invariant cannot only be completely linearized,but also can keep the local geometry of the volumetric shape more fully,which can guarantee the stability of the local geometry of the volumetric shape and deal with more complex deformation.Compared with other methods,constraints require fewer auxiliary variables.In order to determine the conditions of the iteration and the performance of the algorithm,the correspondence between the partial feature points is obtained by the method of dense matching and hypergraph matching.Then,in each step iteration of the volumetric registration system,the spatial compactness of the corresponding feature points that between volumetric models is calculated as the end condition of the iteration in the algorithm.A large number of experimental results show that the proposed method has high registration precision and is quite robustness,which is a stable and reliable automatic registration method.