Research On Face Recognition Algorithm Based On Sparse Representation And Feature Extraction

Because of the great theory and application value, the fast and effective automated authentication and identification technolog ies attract increasing attention with the development of pattern recognition, image processing and machine learning. As an important mean o f identification in biometrics, face recognition received significant progress in information security, criminal investigation, video surveillance and other areas. And it has been widely used in smart home, safe city, intelligent monitoring and intelligent building. Face recognition with sparse representation based classification has become the research focus because of the robustness to noise and partial occlusion, the main idea of SRC is the probe sample can be linearly represented by gallery samples of the same class, and then the classification is performed by minimizing the reconstructed error. Firstly, solve the optimization problem and obtain the sparse coefficients of the probe sample associated with different gallery samples, and then the reconstruction probe sample can be presented as the linear combination of gallery samples, the class of the least reconstruction error is the classification result.In this paper, the research is carried out from the challenges and problems which wildly exist in face recognition, especially the variations of illumination, pose and expression from actual scene. Based on the further study on the related theories of sparse representation, the thesis discusses the problem of the face recognition based on sparse representation, and make contribution to the improvement and innovation of algorithm. The main contents of this paper are presented as follows:1. The paper briefly overviews the framework of face recognition and the challenges it faced. The basic theories of spars e representation are analyzed and discussed, and the research emphases of sparse representation including dictionary learning and coefficient solving are studied. Besides, the traditional SRC procedure and the classical extended algorithm are summarized.2. In view of the fact that the excellent representation ability of classical feature operators, the paper proposed a new algorithm named iteration sp arse representation based classification which utilizes the combination of sparse representation and feature extraction. To enhance the efficiency of face recognition, the algorithm first removes the classes which have the larger reconstruction error and update the training set, then the discrimination information is described by LBP feature or Gabor feature. Finally, the training set with effective feature is used for representation and classification.3. The locality structure has been proved to be critical in pattern recognition and draw extensive attention. A simple face recognition system is introduced in this paper. With the locality and sparsity of image signal, a new classification algorithm based on weighted sparse representation is proposed. To obtain accurate Similarity Measurement of sample, the images presented by LBP feature are used to obtain the locality structure information.4. Considering the nonlinear distribution of images and the loca lity structure information, face recognition algorithm with weighted kernel sparse representation is proposed. The original data is mapped into a high kernel feature space firstly. Meanwhile, to deal with the illumination variation, the samples processed by MSR algorithm are utilized to compute the weighted matrix. With the integration of sparsity and locality, the algorithm is an effective solution of the problem of illumination and occlusion, which are widely existing and challenging.