Research On Face Recognition Algorithm Based On Kernel Method And Null Space

The technology of face recognition is a technology that uses the computer to analyze the image and discriminate identity or recognize status from the worked image. It is a research area spanning several disciplines such as image processing and analysis, computer vision, human intelligent, pattern recognition, Biology, and so on. The research achievement has a broad application prospect. For the particularity of human face images, face recognition with a computer is a very difficult problem and there are still many works to do before such technology can be used wildly. With the development of the society, the application of face recognition systems will be wilder and brings much challenge to the researchers.The process of face recognition mainly consists of three parts: image preprocessing, feature extraction and recognition. In the period of preprocessing we firstly make a precise location of human eye. Then we crop the face image according to the eye location, and scale the cropped face image to the same size. Lastly, the face image is normalized by image gray. The image preprocessing eliminate the impact of illumination in some extend. In the process of human eye location, in order to locate the human eye in precisely, hybrid projection function is proposed in this paper. Hybrid projection function combined the advantages of the integral projection function and the gradient projection function. Hybrid projection function not only reflects the value of the image gray, but also reflects the variance of the gray along a certain direction. Therefore, the utility of hybrid projection function in the eye location can obtain strong adaptability and high accuracy.For the part of feature extraction, in this paper we have introduced classical face recognition algorithm in present, including Principle Component Analysis and Fisher discriminant analysis, and the face recognition algorithm based on kernel method, such as kernel-based principle component analysis and kernel-based fisher discriminant analysis. In the meanwhile, we also made focus on the problem of choosing kernel and setting of the corresponding parameters. What's more, we made a further research on the feature extraction algorithm combined with null space, which can extract the useful discriminant information effectively.In the period of classification, both the Nearest Neighbor method and Support Vector Machine are used as classifiers. The experiment results show that the Nearest Neighbor method can make a good performance in the classification. And it has a wide application. What's more, Support Vector Machine also has been discussed in this paper, which has been applied in the face recognition successfully. However, the parameters setting problem need a further research.Finally, the feature extraction methods, such as Principle Component Aanalysis, Fisher discriminant analysis, kernel Principle Component Analysis, kernel Fisher discriminant analysis and so on, are combined with the Nearest Neighbor method or Support Vector Machine to realize the face recognition. Many experimental results show that the human eye location by the method of Hybrid Projection Function can effectively eliminate the impact of hair and background. As a result, we can combine the null space and kernel Fisher discriminant analysis method to extract the useful discriminant information, which can increase the recognition rate.