The Key Technology Of Face Recognition Research Based On Dictionary Learning

With the advent of the ear of science and technology, the level of technology competition is more and more drastic. In order to improve the environment of technological innovation, at home and abroad have increased investment in science and technology. Under the opportunity the face recognition technology is also rapid development Due to the wide applying of face recognition technology, people have a higher degree for personal and national information security. In the actual application, the feature extraction of human face will be affected by the light, the potential and the obstructions, so the facial feature cannot express the human facial. The human face recognition rate is decreased. In the process of sparse classification, we directly make the training samples as the dictionary to have a sparse representation for the given test samples. If the number of the training samples is little, so the ability of the sparse representation for test samples is poor. In order to enhance the ability of representation and discriminated for dictionary, we can increase the number of training samples, but it can increase the complexity in the process of solving the coefficient. In view of the above problems, the dictionary learning method has been proposed. In the process of face recognition, the performance of the classifier is an important factor for face recognition. In addition to the performance of the classifier, how to make the face feature of extraction have strong ability of identifying information and characterization is also very important. In this paper we mainly do two aspects of work for the problem that are how to extract the features of face that has stronger discriminated information and the how to get a better dictionary.The low dimensional data of samples space are mapped to the high dimensional feature space by the proper kernel function. This can enhance the linear separability of data. Different kernel functions have different characteristics and have a different mapping result. The selection of kernel function parameters is made by the experience, in addition to the experience we have no others theoretical basis. Aiming at the shortcomings of the above, in this paper we think up a face recognition algorithm of multi-kernel sparse classification that based on power kernel. Power kernel function is a kind of triangle kernel function and has stronger stability. Gaussian kernel function has a certain practicality. In practical application, if the Gaussian kernels functions have appropriate parameters, the feature space that mapped through the most of samples can be linear separated. We make the Power kernel function and Gaussian kernel function linear composite together. We make some tests on the ORL database and Yale database and compare with kernel sparse classification algorithm that based on the Gaussian kernel function, Linear kernel function and Polynomial kernel function.The Fisher Discrimination Dictionary Learning gets a dictionary that has stronger discrimination ability through minimizing the divergence within the class and maximizing the discrete degree between classes. This dictionary learning algorithm does not fully combine the divergence within the class and the discrete degree between classes. Aiming at the shortcomings of the above algorithm, in this paper, we put forward to a Multiple Maximum Scatter Difference Discrimination Dictionary Learning. This algorithm has parameter constraint for the divergence within the class and the discrete degree between classes and makes the ability of minimizing the divergence within the class and maximizing the discrete degree between classes to become stronger through regulate the contribution ability between the divergence within the class and the discrete degree between classes. We can get a dictionary that has stronger discriminant ability by the Multiple Maximum Scatter Difference Discrimination Dictionary Learning. In order to verify the validity of the algorithm, we have some tests on the AR database and Yale B database and compared with the Fisher Discrimination Dictionary Learning, Sparse classification. The experimental results show that the algorithm has a good recognition performance.