Non - Iterative Three - Dimensional Maximum Edge Criterion And Its Application In Face Recognition

Face recognition is an active research field in biometric identification. Especially in the new era of anti-terrorism and security issues have an important role. It has become a very important research direction in the field of computer technology and pattern recognition to grasp the characteristics of the human face and the local details.Face recognition research is needed to overcome a lot of difficulties, differences in face shape and the environment diversity by the influence of age and skin damage, which makes the stability of face recognition is very difficult to do, including pose, illumination, facial expressions, and accessories(eyes, whiskers) the influence of external factors, it is difficult to improve the stability and accuracy of recognition. Before many peacekeeping classification method for face recognition based on reduction, mostly the two-dimensional image into a one-dimensional, this will virtually ignored the face originally hidden data structures, and often lead to the small sample size problem, the later two dimensional discriminant analysis technology has become a people are interested in the technology, the method can overcome the shortcomings mentioned in the front of. However, two-dimensional feature extraction method is still based on the whole image of the rows or columns, so the two-dimensional feature extraction method to extract the features may still have some redundant information. In order to fully grasp the details of the human face and contour features, this paper proposes a three-dimensional data for the maximum margin criterion(3DMMC). This method is based on the maximum margin criterion(MMC), and MMC is a redundant, stable and efficient criterion. This method is based on the premise of keeping the data in a three dimensional structure, the data is reduced from three directions simultaneously, and the classifier with high recognition rate is obtained.In this paper in order to capture the contour of the face feature, using the Gabor wavelet transform to extract object contour region feature, research shows, 2D Gabor wavelet kernel function and the mammalian visual neurons cell structure is very similar, with strong spatial position, scale and orientation selection characteristics. So this method will extract more important information, and the recognition ability will be improved significantly.Finally, in order to verify the validity and stability of the proposed method. In this paper, four main face databases and a handwritten data database, the six algorithms are compared. Compare the performance of the recognition algorithm under different training samples. Experimental results show that the proposed 3DMMC+LDA, 3DMMC is a stable and effective method in the field of face recognition.