三维模型在异构空间中的语义迁移方法研究

项目名称： 三维模型在异构空间中的语义迁移方法研究

项目编号： No.61303132

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 王新颖

作者单位： 长春工业大学

项目金额： 23万元

中文摘要： 三维模型检索是多媒体信息检索的重要组成部分。由于模型的低层特征不能很好表达高层语义，使得现有的检索技术很难取得令人满意的结果。我们认为，克服语义与特征之间"语义鸿沟"的关键在于三维模型内容特征与表达语义的有效结合。因此，课题以语义与特征异构空间为切入点，探索三维模型高层语义和底层特征的有效结合方式，进而揭示两者的相互作用和内在联系。首先，分别对特征空间和语义空间展开研究。一方面，研究特征空间中异构关键特征的选择性融合方法；另一方面，以三维模型有效且可量化的语义表达方式为目标，构建描述模型语义关系的异构语义网络。其次，结合语义空间和特征空间，研究语义与特征异构子空间的构建方法及异构空间中的语义迁移方法。最后，在此基础上提出三维模型语义与特征异构距离度量方法。课题的研究能对大幅度提高三维模型检索的有效性提供有力的支持。课题建构在通用的模型描述之上，对其它多媒体信息检索领域也有一定借鉴作用。

中文关键词： 异构语义网络；联合聚类；多线性主成分分析；拉普拉斯特征映射；拓扑稀疏编码

英文摘要： 3D model retrieval is an important part of multimedia information retrieval. Because the low-level features of the model can not express the high-level semantics effectively, the retrieval and classification techniques currently are difficult to achieve satisfactory results. Taking into account the key to overcome the "semantic gap" is effective combination of shape features and semantics of 3D model. Then our project takes the heterogeneous relationship between semantic and content as the starting point, taking advantage of the guiding role of the high-level semantics to the low-level shape features of 3D models to reveal the intrinsic link between the two. Firstly, research on the feature space and semantic space respectively. On one hand, study the selective fusion method of heterogeneous key features in feature space. On the other, taking effective semantic expression of 3D model as the goal, to build the heterogeneous semantic network which can describe the semantic relationship of 3D models. Secondly, learning semantic transfer method in heterogeneous space by subspace learning. Finally, on this basis we put forward a heterogeneous distance metric method based on semantics and features of 3D model. Researches can improve the effectiveness of 3D model retrieval greatly, as well as to provide strong support

英文关键词： Heterogeneous semantic networks；Co-clustering；Multi-linear principal component analysis；Laplacian Eigenmap；Topographic Sparse Coding