热核距离下的3D对称图形匹配算法

3D Symmetric Shape Matching Algorithm Using Heat Kernel

对称混淆问题一直是图形匹配的难点之一,其中,特征点选取、对称点检测、初始匹配对最终匹配结果影响很大.针对此问题提出了一种基于热核的3D对称图形匹配算法.在前期工作基础上,考虑测地线在对称图形检测中的不足,提出一种以热核信号(HKS)为工具的算法.首先将HKS采样和最远点采样结合并融合采样,获得分布稳定且具有代表性的特征点;其次用HKS进行特征点分类实现对称点的准确检测;然后再利用基于热核的一点匹配算法进行初始匹配,以提高正确率;最后利用LS_MDS算法将图形嵌入到欧氏空间并投影到2D平面,根据投影中顶点的法向以及网格上的曲率信息,标定图形方向,最终完成对称匹配.在TOSCA数据库上进行数值实验,并将文中算法与已有的经典算法进行比较,结果表明,该算法不仅提高了匹配准确率,而且适用于较特殊的对称混淆问题,如大猩猩对称图形的匹配问题.

In terms of the problem of symmetry flip,it is always one of the difficulties in shape matching,in which feature point selection,symmetry point detection and initial matching have a great influence on the final matching result.In view of this problem,this paper proposes a 3D symmetry shape matching algorithm based on HKS.Besides,a new method using HKS as a tool is proposed considering the shortage of geodesic in symmetry shape detection on the basis of previous work.Firstly,it combines the HKS sampling and the farthest point sampling and fusing the sampling,so as to obtain the characteristic points with stable distribution and representativeness.Then,it uses the HKS to classify the characteristic points to realize the accurate detection of symmetrical points.After that,it uses the one-point matching algorithm based on HKS to carry out the initial matching and improve the accuracy.Finally,it uses the LS_MDS algorithm to embed the shape into the Euclidean space and project it into 2D.According to the normal direction of the vertex in the projection and the curvature information on the grid,the direction of the shape is calibrated,and the symmetric matching is finally completed.In the TOSCA database,it carries out the numerical experiments,in which the results of the comparison between the methods in this paper and the existing classical methods show that the method not only improves the matching accuracy,but also is suitable for the relatively special symmetrical confusion problems,such as symmetry shape matching in gorillas.