Research Of Graph Matching Based On Geometric Constraints

Graph matching has attracted much attention in the field of pattern recognition and computer vision in recent decades.As one of the basic problem in computer vision,graph matching is applied not only to scientific research,but also to object recognition,object tracking,behavior analysis and so on.Graph matching involves establishing correspondences between the vertices of two graphs.Although the graph matching problem has been investigated for decades,it is still a challenging problem.It is a well-known general NP-hard problem of which global optimum is hardly guaranteed.Consequently,approximate algorithms seeking acceptable suboptimal are popular,and the research interest of this problem mainly focuses on investigating more accurate and faster algorithms.In this paper,we mainly divide the current graph matching algorithm into two categories by thoroughly reviewing,one is the optimization method based on the discrete domain,and the other is based on continuous domain,and the algorithm of graph matching based on geometric constraint algorithm is also an optimization algorithm based on continuous domain.Actually,the graph matching problem is easily affected by outliers,deformation noise and so on,therefore,this paper has finished the following tasks,and the algorithms are experimented on five well-known database including Synthetic dataset,CMU House dataset and so on.Firstly,we analyse the problem of the existence of singular points in the optimization process of the path following algorithms;secondly,in order to address the problem and improve the matching accuracy,we design an effective strategy to estimate the singular point,and explore multiple branches around the detected singular point for potential better matching results;In addition,due to the strategy we used to explore the singular point may result in the computational consumption,therefore,in order to reduce the consumption of computing,and accelerate the convergence of the algorithm,the adaptive step strategy in the iterative process is proposed;finally,because the GNCCP algorithm is based on the original path following algorithm,and it achieves the partial graph matching in an implicit way.Consequently,in this paper,we apply the strategy we proposed to the GNCCP to solve the problem of singular point for better matching results.The experimental results in all cases indicate that the performances of our proposed algorithm in both objective ratio and matching accuracy are superior to other algorithms.