Research On 3D Model Recognition And Retrieval Based On Tensor Subspace

As the development of 3D model, the retrieval technology of 3D model has been widely concerned by scholars. Among them, the main retrieval method is the content-based 3D model retrieval, and the core of content based retrieval technology is feature extraction. Based on the feature recognition method of other multimedia information, this paper considers the application of tensor subspace technology to the feature extraction of 3D model, and describes the 3D shape of the model by three visual tensor. At the same time,we propose weighted multi linear principal component analysis(WMPCA) feature extraction method and weighted multi linear principal component analysis(PCA) and weighted linear discriminant analysis combining the feature recognition method WMPCA+WLDA on the basis of multi linear principal component analysis(MPCA).In this paper, the main work following:1. 3D model standardization and pretreatment, due to the impact of model is not unified, after preprocessing of 3 D model into two-dimensional projection images from multiple perspectives.2. 3D model, in order to maintain the integrity of the internal information, the 2D projection image as a multidimensional tensor proposed a weighted linear principal component analysis(WMPCA) method, and apply it to the 3D model in feature extraction.3. This paper proposed a method of features fusion based on weighted linear weighted principal component analysis and linear discriminant analysis(WMPCA + WLDA), the method can not only save the projection image spatial correlation, but also be used to model the class label information and the weights of characteristic value to improve the characteristics of the 3D model recognition.4. Introduce the implementation method of 3D model retrieval system based on tensor subspace.