针对3维统计特征对模型几何信息描述过于弱化的问题,提出了一种基于3维模型的多视点深度图的几何统计特征提取方法.因为模型库中的每个3维模型的多视点深度图经过相位傅里叶变换后,就可以得到对应3维模型表面面片的法向方向和面片面积大小的2维统计直方图,而且查询模型通过主元分析方法还可得到一个最佳视点深度图和相应的2维统计直方图,所以可通过与模型库的多视点统计直方图进行匹配计算来实现3维模型的相似性匹配和检索.实验表明,该方法对模型的简化是鲁棒的,并具有平移、尺度和旋转不变性,这就很好地解决了3维统计特征对模型几何信息过于弱化的问题,并适合模型的粗分类.

This paper proposes a Geometric Feature Map method bused on multi-viewpoint range images. After every range image of 3D model is phase encoded, a histogram about the planar surface normal and size of 3D model can be worked out. By using principal component analysis, we can obtain a series of range images and the corresponding histogram of 3D model based on the best viewpoint range image. Similarity measurement between two models can be obtained by calculating the distance of the corresponding histograms of two models. The experimental result shows that our method is invariant to the translation, the rotation and the scaling of 3D model and is robust to the simplification of 3D model, and suitable for the classification of 3D model.