基于形状特征的三维模型检索技术综述

A survey of 3D model retrieval based on shape features

在CAD产品设计、3D产品展示、影视动漫制作等应用中经常使用成千上万共享的3D模型,但现有的搜索引擎还不提供有效的3D模型检索服务。基于内容的检索技术能有效地根据3D模型的实际内容,如几何形状,拓扑结构等,查找到用户期望的模型,因而可以用于自动智能检索。本文综述了目前基于形状特征的三维模型检索的主要关键技术,对常用的特征类型进行总结,并重点对特征提取方法进行分类总结。其中基于统计特性的特征提取方法比较简捷,计算简单,但对三维模型的内容描述不够充分;基于扩展特征的方法适用于结构变化较小的网格模型,但计算量较大;基于体素化的特征利用了模型的体积特征,加速了检索过程,效率较高,但不支持多分辨率检索;基于射线采样的特征适合较为简单的模型形状,但容易出现二义性;基于视图的特征主要是在三维模型的二维投影视图上提取的,计算复杂度较低,但容易丢失一些表示三维结构的重要信息。本文同时列举分析了当前实现并发布的典型检索系统。此外,对未来的研究方向进行了展望性描述。

A large number of 3D models are shared in 3D applications,such as CAD product design,3D product exhibition,movie and animation production.But the existing search engines cannot provide an effective 3D model retrieval service for those applications yet.The content-based 3D model retrieval technique is mainly based on the intrinsic shape of a 3D model,such as geometry and topology,so it can be used to search 3D models automatically.This paper summarizes the key techniques of 3D model retrieval based shape features,and collects the commonly used features for retrieval,especially their extraction methods.Methods based on statistical features are easy to implement,but these features cannot describe the mesh shape adequately.Methods based on extended features are good for a model with gentle structure change,but computational cost is high.Voxelization based methods use volume characteristics of the model to speed up the retrieval process,but fail to support multi-resolution search.Methods based on ray sampling are suitable for a simple model,but it is prone to ambiguity.View based methods mainly extract features from two-dimensional projection view of a model and has low computational complexity,but may lose some important 3D structure information.This paper also enumerates and analyzes typical 3D model retrieval systems having been achieved and published up to now.Finally,we conclude this paper with an outlook for the future development of 3D model retrieval.