基于空间-法向量特征的三维模型检索

3D Model Retrieval Based on Spatial and Normal Feature

针对全局法向直方图缺少空间分布信息问题,提出了一种集成空间法向信息的直方图构造方法。在对三维模型进行姿态调整和线性细分预处理操作以后,把模型按照空间分布分割成若干个子块。对每个子块,构造局部法向量直方图来分析其表面特性。为了能够描述直方图不同分量的重要性程度,对每个分量定义特征权值,其大小由该分量对应的三维模型表面积大小所决定。最后,根据局部法向量直方图和特征权值,采用一种改进的直方图相交函数计算不同模型的相似度。对一个标准三维模型测试数据库的检索实验表明,由于考虑了模型的空间分布特征,本文方法可以得到更好的检索准确率。

Global normal histogram has not considered the spatial information of the given 3D shape. To resolve this problem, this paper proposes 3D retrieval algorithm combining spatial and normal information. After pose estimation and linear subdivision, the given 3D model is decomposed into a set of blocks by spatial distribution. For each block, normal histogram is constructed to describe its local shape attribute. Furthermore, feature weight is defined for each component of the constructed normal histogram. Finally, an improved histogram intersection function is presented to calculate similarity between 3D models. Differing with normal histogram applied in 3D object recognition, our method extracts spatial distribution features as well as normal features, and defines feature weight for normal histogram. Experiments show that 3D retrieval based on the proposed algorithm can achieve better retrieving precision than traditional normal histogram.