基于超图的三维模型检索方法研究

项目名称： 基于超图的三维模型检索方法研究

项目编号： No.61502337

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 聂为之

作者单位： 天津大学

项目金额： 22万元

中文摘要： 随着多媒体采集设备和三维建模技术的发展，三维模型数量出现爆炸性的增长。如何对其进行有效管理，从而实现便捷的三维模型检索成为当前亟待解决的问题。本课题研究中拟挖掘三维模型的多视角二维视图表征中潜在的局部和全局结构特性，通过合理的超图构建和匹配实现不同三维模型的相似性度量，从而实现有效的基于视图的三维模型检索。主要研究内容包括三方面：首先，拟通过三维模型的空间结构特征和视觉特征构建图模型结构，利用有效的图聚类算法提取三维模型的特征视图；其次，拟挖掘三维模型特征视图集合潜在的多层级局部结构化特性，从而实现超图模型节点和边的构建；最后，拟基于高阶张量理论实现超图全局结构的特征表征，从而通过超图匹配目标函数构建和求解实现三维模型相似性度量。本课题研究成果对计算机视觉和模式识别领域具有重要意义，并将为虚拟现实、数字娱乐等相关领域的发展提供技术基础。

中文关键词： 三维模型；多媒体检索；图匹配

英文摘要： With the rapid development of multimedia acquisition devices and 3D modeling technology, a large number of 3D models emerge rapidly. It is mandatory to develop advanced 3D model retrieval methods to realize effective management of the big data of 3D models. In this proposal, we plan to discover and leverage the latent local and global structural characteristics in the multi-view images of one 3D model to realize the similarity measure between pairwise 3D models by constructing and matching the hyper-graphs of 3D models. It mainly contains three steps: 1) Constructing graph model by fusing geometry and visual features and then extracting representative views of individual 3D model with graph clustering algorithm; 2) Designing the nodes and edges of the hyper-graph by exploring the latent hierarchical local structural characteristics in the representative views of 3D models; 3) Formulating and optimizing the objective function of hyper-graph matching for the similarity measure of pairwise 3D models with the global structural characteristics in the hyper-graph represented by high-order tensor computation. This research will not only benefit the research of computer vision and pattern recognition but also provide novel techniques for multiple industry applications, including virtual reality, digital entertainment and so on.

英文关键词： 3D Model;Multimedia Retrieval;Graph Matching