On Face Recognition Based On Gabor Feature Sparse Representation

Face recognition is a challenging topic in computational vision. Although there are lots of representative methods:such as geometric feature, subspace analysis, elastic graph matching and characteristics of face of face recognition, etc. These classic algorithms, which only consider the global distribution of training samples, have a high computational complexity, and a poor robustness to variations upon pose, illumination, and so on. These problems make the classic identification methods have a lot of limit in the real application.Different from the classic face recognition methods, sparse representation become recently a hot topic because of its advantages of high recognition rate and robust. Although sparse representation of the face recognition is very innovative and practical, it still exist some issues which need further research and discussion. For example the SRC method been used at present mostly is based on the characteristics of the face, random face, and fisher face features that are overall characteristics. The overall characteristics cannot be very effective against the variation upon facial expression, illumination, and local deformation, and other issues. The SRC method is assumed that the each class has enough training samples, when the training samples are of small the performance of the SRC method also needs a further discussion.In this paper, a survey about the sparse representation for face recognition and a number of relevant improvement methods are given, and Based on it a further research has been done. In particular, the research of Gabor Feature based Sparse Representation Face Recognition is highlight.In brief, main work and innovation are given as follows:1) The thesis presented a real-time video detection of human face recognition system by way of classic face recognition method. Practical application showed that this system is limited by change of pose, illumination, etc. the actual recognition rate is low. 2) In order to further improve the robustness of the SRC, this paper discussed the minimum l1-norm Sparse precision problem in Gabor Feature based Sparse Representation, GSRC) method. In this paper the Vector Total Variation model (VTV) was introduced in GSRC framework to replace the original popular minimum l1-norm algorithm, and improves the precision on classification, then improve further the face recognition rate;3) Recent studies have shown that complexity of face recognition method based on Collaborative Representation (CRC) is decreased much significantly than the SRC method, and CRC has a strong competitive classification results. But in CR, validation is still of the overall features, and the recognition rate remains to be further improved. To this end, this thesis applied Gabor feature to collaborative sparse representation method and the recognition rate of face recognition has great increased;4) Finally, to solve the problems of the computing complexity and computing time caused by the cooperation Gabor feature CRC, the paper further tested several measurement matrix, discussed and compared the impact on reducing the computational complexity or feature dimension, and the Possible research target was given for the future.