结合描述性文本的三维模型语义检索方法

摘要

To improve the retrieval performance of 3D model, concerning the problem that the semantic-based 3D model retrieval system is hard to support customers' subjective words, a 3D model semantic retrieval method based on content and descriptive text was proposed. This method constructed a semantic tree for 3D models firstly. Then, it calculated the similarity among the input and node of tree by the word statistics method, and got some 3D models from those nodes with high similarity,and a smaller 3D models set by semantic constraint. Finally, user input' s 3D model examples may match the shape similarity in the smaller set of 3D model through semantic constraint, and returned search results to users. The WordNet definitions of some words were as input in experiments. The experimental results on PSB show that this method performs better than the content-based 3D model retrieval method on recall-precision.%为了提高三维模型的检索性能,针对当前三维模型检索系统的语义检索功能无法支持用户的主观性描述文字的问题,提出一种基于内容和描述性文本结合的三维模型语义检索方法.该方法首先为三维模型构造语义树;然后,利用语料统计的方法,计算输入的描述性文本和语义树节点扩充信息的相关程度,将相关度较高的一部分节点的三维模型实例提取出来,得到一个经过语义约束的较小的三维模型集合;最后,使用用户输入的三维模型实例在这个经过语义约束的较小的三维模型集合里进行形状相似性匹配,依据匹配度的大小返回给用户三维模型检索结果.实验中,使用WordNet对一些名词的释义作为描述性文本输入.在普林斯顿大学的PSB三维模型数据集上的实验结果表明,该方法在大多数类别中的查准率-查全率性能好于传统的基于内容的三维模型检索方法.