Research Of Face Recognition Based On Improved Sparse Representation

Face recognition has been a hot topic in the field of pattern recognition and artificial intelligence for decades. And it has been widely used in various industries. In recent years sparse representation has been widely used for face recognition and achieved good results. The thesis explores the optimization structure of redundant dictionary and the relation between sparsity and collaborative representation, to improve present face recognition methods based on sparse representation.Firstly, extract fusion feature as dictionary atom and introduce LDA to rebuild the dictionary, to form linear discriminative redundant dictionary with better structure. For face recognition based on sparse representation, the design of redundant dictionary is very important for sparse representation of the testing sample. Through extracting fusion feature to combine their advantages and get better face representation with a low feature space dimension, then introduce LDA to transform the dictionary, in order to reduce the dimensionality of the dictionary and improve the performance. The extensive experiments demonstrate that the proposed dictionary has better recognition rate and operating efficiency, while it can easily reject distractor faces.Then, the thesis proposed a fast face recognition method with regularized least square via sparse representation based on linear discriminative redundant dictionary, namely Fast Sparse Representation Classification with Regularized Least Square(FSRC_RLS). Most sparse representation methods require a redundant dictionary that the number of atoms in dictionary is much larger than the dimension of it. and they ensure sparsity by solving l1-norm minimization. Both of the procedures will increase the complexity of the algorithm. The proposed method reduces the number of atoms through extracting principal component of each class in linear discriminative redundant dictionary and enhances the sparsity of l2-norm to improve computation speed. The extensive experiments demonstrate that FSRC_RLS can improve the computation speed significantly compared with other sparse representation methods at low dimension, while ensuring the recognition performance and the ability of rejecting distractor faces.Finally, based on linear discriminative redundant dictionary and FSRC_RLS, the thesis designs and implements video face recognition prototype system. The system can perform video input, face detection, feature extraction, dictionary building, face recognition and face label with good instantaneity, accuracy and robustness. And it provides convenient interface in order to make administrator collect face images and observe test face identity.