3D Model Classification and Recognition Method Based on Tensor Principal Component Analysis

This paper propose a 3D model classification and recognition method based on tensor principal component analysis. Firstly, the 3D model is transformed into 2D view and expressed in tensor, and then feature vector from multiple angles of tensor is extracted. On the basis of multi-linear principal component analysis and linear discriminate analysis, we present a method for merged weighted multi-linear principal component analysis with weighted linear discriminate analysis (WMPCA + WLDA). This method is not only can save spatial correlation of the projected view, but also can improve feature recognition of the 3D model through using class label information and features weight. Experiments on Princeton Shape Benchmark has shown that the method is superior to the conventional MPCA, MPCA + LDA, WMPCA + LDA and other methods on classification results of 3D model.