Shape matching and retrieval based on multiple feature descriptors  

作　　者：WANG Weiming LIU Xiuping LIU Ligang 

机构地区：[1]School of Mathematical Sciences,Dalian University of Technology [2]School of Mathematical Sciences,University of Science and Technology of China

出　　处：《Computer Aided Drafting,Design and Manufacturing》2013年第1期71-78,共8页计算机辅助绘图设计与制造（英文版）

基　　金：Supported by National Natural Science Foundation of China(61222206,61173102,U0935004);the One Hundred Talent Project of the Chinese Academy of Sciences

摘　　要：A lot of 3D shape descriptors for 3D shape retrieval have been presented so far. This paper proposes a new mechanism, which employs several existing global and local 3D shape descriptors as input. With the sparse theory, some descriptors which play the most important role in measuring similarity between query model and the model in the dataset are selected automatically and an affinity matrix is constructed. Spectral clustering method can be implemented to this affinity matrix. Spectral embedding of this affinity matrix can be applied to retrieval, which integrating almost all the advantages of selected descriptors. In order to verify the performance of our approach, we perform experimental comparisons on Princeton Shape Benchmark database. Test results show that our method is a pose-oblivious, efficient and robustness method for either complete or incomplete models.A lot of 3D shape descriptors for 3D shape retrieval have been presented so far. This paper proposes a new mechanism, which employs several existing global and local 3D shape descriptors as input. With the sparse theory, some descriptors which play the most important role in measuring similarity between query model and the model in the dataset are selected automatically and an affinity matrix is constructed. Spectral clustering method can be implemented to this affinity matrix. Spectral embedding of this affinity matrix can be applied to retrieval, which integrating almost all the advantages of selected descriptors. In order to verify the performance of our approach, we perform experimental comparisons on Princeton Shape Benchmark database. Test results show that our method is a pose-oblivious, efficient and robustness method for either complete or incomplete models.

关 键 词：SIMILARITY INVARIANT shape descriptor subspace clustering 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]