一个相关反馈三维模型深度图像检索算法

3D model relevance feedback retrieval algorithm based on range image

基于图像的三维模型检索比形状特征和拓扑结构特征具有更易于实现且检索效果好等优点,但现有基于图像的三维模型检索方法存在一些问题,如没有考虑三维深度信息、所提特征不能全部表达三维模型信息且不能实现用户交互操作等。对此提出一种基于深度图像的三维模型相关反馈检索算法:首先提取三维模型深度图像边界方向直方图和Zernike矩特征;然后利用特征距离度量进行三维模型检索,并采用相关反馈技术实现分类器的构造和模型的标注;最后利用调整好的权值分类器进行检索和标注。实验表明,该算法实现了三维模型的相关反馈检索和标注,提高了检索的精确性,并增强了系统的实用性,为进一步进行语义检索打下基础。

In 3D model retrieval area,compared to shape features and topological structure features,the method based on image has the advantages that it is easy to realize and the retrieval results are good.But the existing image retrieval methods in 3D model do not consider the depth information and the features cannot express all information of 3D model.This paper proposed the 3D-model relevance feedback retrieval algorithm based on range images.Firstly,this algorithm computed the features of edge direction histogram and Zernike moment of 3D model＇s range images,then the feature distance was used to retrieval 3D model,and relevance feedback was used to build the classification and label the 3D model in database,at last the trained classification was used to retrieval and label the 3D model.The experiments show that the algorithm improves the accuracy and practicability of system,and realizes the relevance feedback retrieval and the 3D model＇s label which lay the foundation for semantic retrieval.