Research On Hypergraph Matching Algorithm Based On Tensor Refining

Hypergraph matching as the important way to solve image matching problem,is widely used in target detection,target tracking and multimedia analysis and it plays an important role in the development of computer vision.With the development of research and expansion of application,hypergraph matching has attracted more and more attention.Although many hypergraph matching algorithms have been produced,the solving of the mathematical model in hypergraph matching is often NP-hard,and most of the algorithms solve this problem by replacing the global optimal solution of the objective function with the approximate local optimal solution.In practical applications,the hypergraph matching is also easily affected by noise,outlier points,deformation factors and other factors,resulting in a decrease in matching accuracy.There are no algorithms that can completely overcome these difficulties,the main purpose of hypergraph matching is to find a way to make matching results more accurate and faster.In view of this,this paper focuses on the hypergraph matching algorithn,the main work includes:(1)We summarize and analyze the existing hypergraph matching algorithm,point out that the current tensor hypergraph matching algorithm has the problem that the tensor itself does not update in the solving process of objective function,causing the matching accuracy is reduced.And the objective function is difficult to relax and influence the matching effect.(2)In order to solve the problem that the tensor is not updated,this paper proves the method of optimizing tensor on probabilistic model,and proposes the refining tensor hypergraph matching RTM to obtain the matching result with higher matching accuracy.Based on the RTM,a refining tensor hypergraph matching with adaptive alternating growth RATM is proposed to deal with the additional computational time consumption caused by tensor refining.The adaptive growth strategy is used to adaptively adjust the matching vectors in three directions during iterative solution to accelerate convergence and reduce computation time.(3)For the problem that objective function relaxation of tensor hypergraph matching is difficult to solve,this paper proposes a hypergraph matching algorithm CCTM based on CCRP that using the convex and concave relaxation procedure.This algorithm only needs gradient of the objective function to perform the relaxation and reduce the difficulty of solving the hypergraph matching and improves the matching accuracy.The proposed algorithm is experimentally verified on three publicly available datasets in the field of image matching including Synthetic,CMU House and Willow.All the experimental results show that the hypergraph matching algorithm proposed in this paper achieves a good matching effect both in terms of matching accuracy and objective ratio.At the end,a brief summary of the work and a prospect for future work are presented.