Face Recognition Based On Sparse Representation In Security Protection System

Facial recognition is one of the key steps in face based security and defense system, which mainly including geometry, texture feature extraction, subspace projection and classifier design etc. This thesis primarily focuses on research of facial recognition technology based on sparse representation. The main research contents of the thesis are listed as follows:1.A serial classic facial recognition algorithms and facial recognition technology based on sparse representation are systematically tested and compared. The experiments on the public data sets of ORL and Yale facial databases using the two base classifiers KNN and SVM, along with the two subspace projection algorithms PCA and fisherface are conducted. From the experimental results, we draw the conclusion that sparse representation achieves more robust recognition results in the noisy dataset.2.A new classification algorithm based on weighting coefficients sparse representation is proposed. Unlike the traditional method which selecting the label linked to the max coefficient of sparse representation, the proposed method not only takes advantage of the structure of the image subspace, but also the plentiful values of sparse coefficients which are closed to zero. Compared with the minimal residual classification, the proposed method is less complex, therefore leads less calculation. Experimental results show that when the number of the training sample is smaller,the recognition rate achieves1.26%higher than that of the minimal residual method.3.A new application for l1/2-norm optimization is proposed. Compared with traditional l1-norm, the performance of l1/2-norm based classifiers is a little better, but sparsity of the coefficients is much stronger.4A set of prototype system of facial recognition based on sparse representation is designed and implemented, which is capable of collecting face training samples, completing pre-process the face image, doing the feature extracting and face recognition based on sparse representation. According to the tests done to the system, the current system is able to reach a relatively high recognition rate and at the same time, it is more robust and practical under conditions where noise, different types of lightings are present.