一种多特征动态集成的3D模型检索方法

3D Model Retrieval Based on Multi-Feature Dynamic Itergration

基于内容的3D模型检索是多媒体信息检索的热点研究问题之一,其主要的问题是提取3D模型的形状特征,但是采用单一特征很难保证检索系统对于任意输入查询模型都有很好的检索准确率。为了提高准确率,引入多种形状特征,通过加权求和在输出层融合,得到总体上模型间的相似度。每种特征采用动态权值,针对输入查询模型的不同,根据用户反馈对知识库中的权值自动更新。采用了形状分布特征和球面调和特征进行系统的建模,实验表明,文中方法比采用单特征检索的准确率大大提高。

The content- based 3D model retrieval is one of the hot research problems in the multi - media information retrieval, and the main problem is how to distill 3D models＇ shape features. But single feature is not enough to describe the shape of 3D models. Present a method by itergrating multi - feature to improve retrieval accuracy. Two features are used and each feature has a weight. These features are syncretized on output hyer. By plusing the two results of the features with their weights, can get the whole result. According to users＇ feedback, the system change the weights of each feature dynamically. Experinents show retrieval accuracy has a greater improvement than on single feature.