Human Facial Recognition Based On The Canonical Correlation Analysis

Face recognition plays an important role in the process of people's social communication. The goal of face recognition is to make the computer accurately recognize faces though facial information. The initial research of face recognition mainly focuse on the single feature. With further studies on face recognition, people recognize the face features from different channels between are certain complementary. Though making full use of the correlation between different channel characteristics, the result of face recognition can be much better.The face recognition based on multimodal has became the main research area.In the past 10 years, canonical correlation analysis(CCA) was wildly used in the field of pattern recognition, computer vision and biomedicine. CCA has made a breakthrough in many fields such as face recognition, disease diagnosis, movement classification. This paper mainly study the application of canonical correlation analysis in the field of face recognition.The main research works of this paper were organized as follows.1. A new feature fusion algorithm is put forward--------Generalized supervised locality preserving canonical correlation analysis algorithm and apply it to face recognition system. The nonlinear characteristic of the traditional fusion algorithm existing problem which is can't use or can't make full use of sample classification information during the feature fusion. When compared with the traditional fusion algorithm of nonlinear characteristics, this algorithm fully takes advantage of the sample classification information and supervise effectively about the related characteristics.So that the extracted characteristics will be more suitable for the classification.2. To solve the problem that the fitting error increased while adding supervised information to the sparsity preserving canonical correlation analysis(SPCCA) and the effective information losses while using Principal Component Analysis(PCA) to extract principal features of the scattering matrix, an improved supervised sparsitypreserving canonical correlation analysis algorithm is proposed based on exponential dimensionality reduction. The problem that the fitting error increased while adding supervised information to the SPCCA is solved though the fusion of the class label information and Sample feature. The local manifold structure of the data is preserved at the same time. For the problem of High-dimensional small sample size, index scattering matrix is used to keep effective information while building the non-singular scattering matrix. The problem of effective information losses while using PCA to extract principal features of the scattering matrix is solved though this method.