Research On Image Registration Technology Based On Graph Models

Image registration technology is the basis of many disciplines,and its research results are widely used in computer vision,object detection,stereo vision,image classification and retrieval,and 3D reconstruction.With the development of technology application,the efficiency,adaptability and accuracy of image registration have become the goal of many scholars.As an effective representation of structural feature information of the set of points to be matched,graph model has been paid more and more attention in the research of image registration technology because of its visualization and visualization.In this paper,the problem of image registration in different cases is studied through the construction of different graph models.The main contents and results are as follows:1.A graph matching algorithm based on attribute relational graph model is proposed.The algorithm uses the distribution of feature points to construct the attribute relationship between feature points and edges.Firstly,the number of feature points on the one side of the line between the feature points and the central point of the point set is used as the attribute of the feature points,and then the direction relation and attribute information of the edges are determined according to the number of points on both sides of the line between the feature points.Then the affinity matrix between the two images is constructed and the matching results are solved by the iterative solution method with integer constraints.The experimental results show that the algorithm has good performance and good matching effect on images with large deformation.2.An image registration algorithm combining hypergraph constraint and improved normalized cross correlation is proposed.In order to improve the accuracy and adaptability of image registration algorithm,hypergraph constraint and the improved NCC are applied to image registration.The proposed algorithm uses the Hessian-Affine detection affine invariant region instead of the fixed window to improve the NCC method and obtains the initial matching point pairs;The similarity degrees between the hyperedges of hypergraph are calculated by Martensitic distance,and the matching scores of the matching pairs calculated by hypergraph constraint are used to sort the matching pairs;The transformation matrix is initialized with the some matching points of higher matching scores and circularly updated by filtering matching pairs to get the optimal transformation matrix,which is used to implement registration.The experimental results show that the proposed method has better performance in matching and rejecting mismatch,and also has better registration performance in different types of image registration.