一种基于单一视图的三维模型检索方法

3D shape retrieval method based on single view

针对已有研究存在的视图冗余性问题,提出采用单一视图进行三维模型形状特征度量,并通过动态规划进行相似性计算,实现三维模型的单一视图特征描述和匹配.算法主要由三步组成:首先,将三维模型进行姿态调整,并通过渲染得到最能表达三维模型外形特征的主视图.其次,对渲染得到的视图进行轮廓采样,通过内在距离和内角提取模型形状特征描述.最后,采用动态规划算法计算不同模型之间的相似程度.实验结果表明:尽管只采用单一视图进行特征描述,但所提出的算法检索精度要高于一些典型三维形状特征描述符,包括采用深度视图定义的形状描述符.

In order to solve the redundant problem of views in the existing research works, an approach based on a single view for 3D model shape feature measuring is proposed. The similarity between two models will be calculated through dynamic programming algorithm. So the single view feather description and matching of 3D model can be implemented. The algorithm consists of three steps as follows： Firstly, the pose of 3D model will be adjusted and the main view of the 3D model, which can best express the characteristics of the 3D model shape, can be gotten through rendering. Then, outline of the main view will be sampled and the model shape feathers can be described through inner distance and inner angle. Finally, the degree of similarity among different 3D models can be calculated through dynamic programming algorithm. The experiment results indicate that although only a single view is used for feature description, the retrieval precision of this algorithm is more accurate than the typical 3D shape feature descriptor, including the 3D shape descriptor based on depth buffer.