Face Recognition Based On Dictionary Learning And Sparse Representation

Face recognition is a hot research topic in the field of computer vision and pattern recognition, it has a very wide range of application. However, there are still many problems to be solved in face recognition, for example, the problems such as occlusion, the change of illumination and posture to a certain extent limit the development of face recognition technology. Unlike traditional face recognition technology, face recognition technology based on sparse representation for its many advantages has been becoming a current hot spot. Although sparse representation method has achieved good results and robustness of recognition, but there are still some problems need to be sovled.For example, the current most algorithms sparse representation are based on of global features, such features can not be good to overcome the Illumination variation and a series of changes in posture and expression; And the performance of dictionary in the sparse representation algorithm is very critical for identifying effect, So how to choose and construct an effective dictionary to represent the face is the focus of research.Therefore, based of sparse representation in the face recognition, this thesis carried out further research work; Thesis’s main work and innovations are summarized as follows:1. We study the main feature extraction methods in face recognition, and base on the sparse representation theory, the sparse model of face recognition is built. In addition, the dictionary learning methods is studied in order to learn the adaptive dictionary for face representation.2. The method of Gaussian Mixture Sparse Representation for Image Recognition Based on Gabor Features and Dictionary Learning is studied. In order to increase the dictionary representation capability and use the discriminant information,we introduced the Fisher discriminant constraints to the dictionary learning process so that the dictionary has a certain ability to identify categories, we use Gaussian mixture sparse representation for classification, the discriminating items can be expressed to the maximum likelihood function of residuals,so the problem of identification is converted to the optimal weighted norm approximation problem.3. In order to solve the problem such as occlusion and noise in images, the method of extended sparse representation for face recognition based on Gabor features and Metaface learning is studied. The method extracts Gabor features of images firstly, and then a new dictionary with stronger sparse representation power can be obtained from the Gabor feature sets by Metaface scheme;meanwhile in order to overcome the effect of occlusion of the image more effectively, the Gabor occlusion dictionary is employed to encode the occluded part of the image; finally, the test image can be reconstructed by the over-complete dictionary bases, the residual between the sample and the reconstructed sample is used for classification by minimizing residual.