A 3D model Retrieval Approach Based on Bayesian Networks Lightfield Descriptor

摘要：

A new 3D model retrieval methodology is proposed by exploiting a novel Bayesian networks lightfield descriptor (BNLD). There are two key novelties in our approach: (1) a BN-based method for building lightfield descriptor; and (2) a 3D model retrieval scheme based on the proposed BNLD. To overcome the disadvantages of the existing 3D model retrieval methods, we explore BN for building a new lightfield descriptor. Firstly, 3D model is put into lightfield, about 300 binary-views can be obtained along a sphere, then Fourier descriptors and Zemike moments descriptors can be calculated out from binaryviews. Then shape feature sequence would be learned into a BN model based on BN learning algorithm; Secondly, we propose a new 3D model retrieval method by calculating Kullback-Leibler Divergence (KLD) between BNLDs. Beneficial from the statistical learning, our BNLD is noise robustness as compared to the existing methods. The comparison between our method and the lightfield descriptor-based approach is conducted to demonstrate the effectiveness of our proposed methodology.

关键词： 3D model retrieval Bayesian network learning lightfield descriptor Kullback-Leibler Divergence

作者: Qinhan Xiao YanJun Li

作者单位: College of Astronautics, Northwestern Polytechnical University,XianChina,710072 College of Astronautics,Northwestern Polytechnical University,Xian China,710072

会议类型: 国际会议

会议名称: International Conference on Space Information Technology 2009(2009年第三届空间信息技术国际会议)

会议地点: 北京

会议语种:英文

页码: 1-9

在线出版日期: 2009-11-26（万方平台首次上网日期，不代表论文的发表时间）