Tensor Subspace Analysis Based Multi-view Face Recognition

In recent years, with the rapid development of biometric identification, artificial intelligence and the new generation of human-computer interaction technique, automatic face recognition has become the research hotspot because of its convenience, friendliness and reliability. Studies have shown that nonlinear view variations lead to a considerable decline of recognition performance. Therefore, the research on multi-view face recognition is of significant importance in practice. In this paper, tensor subspace analysis (TSA) based multi-view face recognition algorithms are studied. The main contributions of this paper are summarized as follows:A multi-view face recognition algorithm based on Gabor features and TSA-view manifold modeling is studied. Firstly, multi-scale and multi-direction Gabor filters are introduced to extract facial features. Subsequently, principal component analysis (PCA) is used for dimensionality reduction. Finally, view information of face images is extracted and modeled by TSA for the purpose of view estimation, which is followed by the multi-view face recognition. Experimental results show that the proposed method is effective to extract the texture structure of multi-view face images and more accurate to describe nonlinear view manifold, which leads to the improved recognition accuracy.In order to preserve as much as possible the non-linear structure of multi-view faces, a multi-view face recognition algorithm based on TSA and locality preserving projections (LPP) is studied. Firstly, TSA is utilized to remove the identity information of face images and reconstruct the face image sets. Next, LPP is introduced for projecting the reconstructed images to a low dimensional subspace. Finally, the view estimation of the test image can be achieved by the K nearest neighbor (KNN) classification method, which is followed by the multi-view face recognition. Experimental results demonstrate that the presented algorithm not only maintains the nonlinear structure of the face images, but also possesses good robustness for the evaluation of the face view.