Study Of The Facial Recognition Methods Based On Matrix PCA

The traditional biometrics is the technology using physiological or behavioral characteristics owned by animals and people to realize authentication. To some extent, the extracted features belong to everyone and can constant for a long term. But they are different to each other. As the biometric owned by everyone, the face’s uniqueness and the advantage that it’s hard to be replaced make it become the most attractive technology in biometric identification and pattern recognition areas. The application areas include guard system, e-commerce, ultra-linguistic, cultural background, the robot developing technology, polygraph, smart environment, recruitment, test, the man-machine interface for a variety of ways, intelligence toys and so on.At the beginning, I introduce the main contents, status and significance of the study. Then, I analyze a lot of common methods and choose the method based on PCA,2DPCA and the wavelet transform to compare them with each other. Secondly, I propose an improved method called2DPCA reconstruction within a class. Using this method, we can compute the two-dimensional image matrix directly, which can resolve the too high dimension phenomenon. The first step is to do2DPCA in each training set; we can get the projective space in every class. Then we reconstruct the test image in several projective spaces respectively. As the reconstruction error in the same class is least, we can get the classification results according to the differences between the reconstructed image and the original image. We can get the comparing result in MATLAB between PCA reconstruction within a class and improved method. It shows that the method used by this paper can improve the recognition rate and the meantime it can ensure real-time. Finally, I optimize the facial recognition method based on wavelet transform. According to the traditional theory, the image’s main contents are concentrated in the low frequency components; we can ignore the contours and edge information carried by the high frequency components. For this, I propose an improved method based on wavelet packed fusion and2DPCA. Firstly, I do second floor wavelet packet decomposition of the image, and then fuse the four of the most conducive to the classification of high-frequency sub-graphs. Do2DPCA in high-frequency and low-frequency sub-graphs. Finally, we do decision level fusion to get the result. In MATLAB experimental platform, I compare the improved method with traditional facial recognition method based on wavelet transform. It’s clear that the approved method is effective.