Sparse Representation Based Face Recognition And Its Implementation

With the rapid popularization of the video monitoring, information security, and image retrieval techniques, face recognition, as a main technical mean of the Biometric Identifica-tion Technology, has received extensive attention and development. After decades of contin-uous efforts, the existing face recognition systems can yield a very good performance in well controlled conditions of the environment. However, when the environment is changed, the recognition rates of the systems fall dramatically.Along with the rise of compressive sensing theory, sparse representation classification (SRC) based algorithms were introduced to face recognition. At present, the SRC methods are still developing and perfecting. Thee main objective of this thesis is to study the process and mechanism of SRC methods for face recognition and to propose an improved face recognition system. The main works done in this thesis are as follows:1. The theory of face recognition problem, compressed sensing and sparse representation is studied. The current mainstream of methods for feature extraction and signal recon-struction algorithm is discussed in details.2. The classification framework of sparse representation is studied and the corresponding algorithms are implemented on Matlab. For the deficiencies of the existing algorithms, a new method is proposed to avoid only one atom to be selected from the dictionary in each iteration via regrouping atoms of the dictionary by category labels. Simulations show that the group sparse representation method has better recognition results and robustness.3. To overcome the shortcomings of SRC method, we propose a low-rank subspace joint sparse representation recognition algorithm. The dictionary is divided into two parts, a valid entry and error term. The classification discrimination is performed using the residues of sparse representation of two parts. By doing so, the occlusion effect on the recognition rate is effectively reduced.4. The face recognition prototype system is implemented on the platform of Matlab 2009, including face image input, preprocessing, feature extraction and classification module. By implementing a prototype face recognition system, the feasibility and effectiveness of the proposed methods are verified. Such a platform can be useful for further study of face recognition.