Face Recognition Method Based On Gabor And Partial Least Squares Regression

In the 21 st century, we have stepped into the information age and more and more products have introduced face recognition technology. Face recognition can help people quickly and accurately identify identity, with great research value and significance. In this paper, we study the key technologies of face recognition: facial feature extraction and classification methods, and propose an improved method of constructing a new dictionary for face recognition based on sparse representation. The main work is as follows:First, in the stage of feature extraction, this paper analyzes the shortcomings of principal component analysis(PCA), which is the classic feature extraction method. On the one hand, PCA puts the training samples as a whole and find the optimal linear mapping when the mean square error is the maximum, while maybe ignoring the important information in the other direction mapping. On the other hand, although the extracted features outline the information of training samples, they ignore the category attribute and have no explanatory power for the category. Because of these inadequacies, this paper proposes a new feature extraction method, which is the method based on Gabor filters and partial least squares regression(PLSR). After the convolution between the Gabor filters and the training samples, the training samples projects onto a plurality of directions at various angles, which can explore the optimal mapping in multiple directions and reduce the loss of information. And PLSR method puts training samples as independent variables and puts the category matrix as the dependent variable. The feature extracted by this way can summarize the training sample and can explain the category attributes.Second, this paper introduces a classification method based on PLSR, which is faster and more efficient than the classical method. Combined the feature extraction method proposed by this paper with the classification method, this paper describes a face recognition method based on Gabor and PLSR. This method is proved to be effective by experiments.Third, this paper describes the traditional method of face recognition based on sparse representation classification(SRC) and analysis the shortcomings of the traditional method, which are that the method puts training samples as the dictionary that is the basis for SRC. So the dictionary should be representative and immunity. This paper proposes a method for constructing a dictionary: firstly, executing a convolution between the training samples and Gabor filters, and then using PLSR to reduce dimensions. Gabor filter is good at edge extraction and texture expression and separation, and it can highlight the eyes, mouth, and nose from the face image. Even more, Gabor filter is not sensitive to light, which makes the dictionary more robust. And PLSR makes the dictionary have explanatory power for the category. At last, the experimental results show improved method of face recognition based on sparse representation in the processing of light, posture and facial expressions are superior to the traditional, and it has strong anti-jamming capability.