Research Of Face Recognition Based On Two-dimensional Image

Face recognition technology is one of the most important biometric identification technologies. Everyone can take advantage of the unique facial features for identification. In this society with the highly developed information technology, the identification and protection of personal information has become a top priority. With its high efficiency, accuracy, non-contact and other unique advantages, face recognition technology rapidly becomes an important mean of identification of personal information. This paper firstly introduces the research background and development status of face recognition technology, and specifically describes four classical face recognition methods, which are the principal component analysis, linear discriminant analysis, locality preserving projection and face recognition method based on kernel. After studying these classic methods, this paper improves four two-dimensional discriminant analysis methods used in face recognition to solve the problems of some existing methods. This part is the main research content of this paper.(1) Face recognition method based on bidirectional two-dimensional iterative uncorrelated discriminant analysis(2D2UDA). Two-dimensional uncorrelated discriminant transformation(2DUDT) method can only extract the feature in the vertical direction, which can only reduce the dimension of one direction. The method can calculate both horizontal and vertical directions of the optimal set of discriminant vectors, which meet the uncorrelated conditions. Experimental results show that the proposed method is more thorough to reduce the dimension of face sample matrixes.It also has a higher recognition rate.(2) Face recognition method based on bidirectional two-dimensional uncorrelated set of discriminant vectors(2D2UDV). Two-dimensional uncorrelated discriminant vectors(2DUDV) method can only extract the feature in the vertical direction, which can only reduce the dimension of one direction. This method uses the non-iterative algorithm to calculate the uncorrelated set of discriminant vectors, whichhave the characteristics of 2DPCA and 2DLDA model. Experimental results show that the proposed method can reduce the dimension of face sample matrixes, which can make the feature samples with fewer dimensions in the feature subspace. It also has a higher recognition rate and less calculation time.(3) Face recognition method based on two-dimensional local linear discriminant analysis(2DLFDA). One-dimensional local linear discriminant analysis(LFDA)method has destroyed the matrix structure information in the process of stretching the image matrix into one dimension vector. The method can extract the features and project directly on two-dimensional matrixes, which avoid the loss of the information of the matrix structure when two-dimensional matrixes stretched into one-dimensional vectors. The data structure of the matrixes can be well protected. Experimental results show that the proposed method can effectively reduce the computational complexity,saves the storage space, and has high recognition rate.(4) Face recognition method based on kernel two-dimensional uncorrelated discriminant analysis(K2DUDA). This method combines the advantages of the Kernel method and uncorrelated discriminant transformation(2DUDT) method.The method can project the two-dimensional matrixes to the high dimensional subspace, which can be divided linearly. Then we use uncorrelated discriminant analysis method for feature extraction in the high dimensional subspace to find out a set of uncorrelated discriminant vectors. The vectors can project the original samples in the high dimensional subspace. Experimental results show that the proposed method has the high recognition rate.