基于张量子空间的三维模型特征提取及检索方法

为提高三维模型的检索效率, 针对三维模型特征提取方法进行了研究, 在多线性主成分分析(MPCA:Multi-Linear Principal Component Analysis)的基础上, 提出了一种加权多线性主成分分析(WMPCA: Weighted Multi-Linear Principal Component Analysis)方法, 并将其应用于三维模型特征提取中。 该方法首先将三维模型转化为多角度的二维投影图像, 然后从多方向上通过张量进行特征提取, 最后将提取到的特征应用到三维模型检索中。 对 Princeton Shape Benchmark 的实验表明, 该特征提取方法比经典的形状分布方法平均检索效率提高7%, 比传统的 MPCA 特征提取方法的平均检索效率提高 3%。

Method of 3D Model Feature Extraction and Retrieval Based on Tensor Subspace Learning

Feature extraction method for 3D model was studied. A method of WMPCA(Weighted Multi-linear Principal Component Analysis) for feature extraction based on MPCA(Multi-Linear Principal Component Analysis) was proposed, and it was applied to 3D model feature extraction. Firstly, the 3D model was transformed into multi-angle 2D projection images, and then were extracted features using tensor from multi-direction. Finally, the extracted features were applied to 3D model retrieval. Experimental results on Princeton Shape Benchmark show that the feature extraction method is better than classical shape distribution method and traditional MPCA method.