Research Of Face Recognition Algorithm Based On Compressed Sensing

Face recognition is a popular subject in the field of computer technology. It exploits biological characteristics to identify the different persons. Due to its advantages, face recognition has been widely used in multimedia processing, pattern recognition, image processing, computer vision and other fields. In recent years, face recognition technology has made great progress, a large number of face recognition system is used in the authentication.Compressed sensing theory is a new theoretical framework proposed by Donoho and Candes et al in2006, Soon to be introduced into the field of face recognition and sparked a wave of research upsurge.The most successful application of the theory is the SRC algorithm(Sparse Representation-based Classification). Compared with most existing algorithms, the SRC method takes advantage of the sparse distribution of high dimensional data, could deal with the problem of high dimension of face image effectively. In addition, the process use the original image pixels, greatly reduces the loss of information caused by some preprocessing operation. However, it is easy to cause the alignment error by changes in facial pose and expression. This problem also affects the performance of the SRC method while the existing SRC method requires the training images and test images strict alignment, It has become the main obstacle that block the SRC method to application.This thesis firstly studies the face recognition algorithm based on sparse representation, introduces the traditional PCAn LDA method and carries out some simulation tests, then realizes the SRC algorithm by introducing the compressed sensing and get good recognition effect. This thesis presents the basic principle of the SRC algorithm, and Simulates the experiment with this method and the traditional method on the ORL face database, calculates the recognition rate, compares and reveals the difference and connection between these methods.Following the above research, this thesis realize the recognition method based on the two-stage sparse representation because of the shortcomings in SRC method. The experiment shows that the advantage of two-stage sparse representation compared to the traditional SRC. Then aiming at the problem of lacking training samples existing in the application, this thesis realize the two-stage sparse representation method based on improved face database. This thesis demonstrates the effectiveness of this method through the simulation experiment in the YALE、ORL、FERET and AR database. At the same time the method weakened the dependence on the number of adjacent sample, enhance the robustness.