Study On Face Recognition Based On Fusion Of Local Feature And Global Feature

Face recognition is a biometric technology possessing great developable potential, the key issue of successful face recognition approach is how to extract discriminant feature from a face image. In recent years, however, face recongtion technology has achieved unperecdented progress, facial personalized feature extraction quality personalized subject to posture, facial expressions, illumination and other factors, which lead to its recognition precision in practical applications still cannot satisfy the expectant demands of people.Isolated grey values of pixels cannot reflect the characteristics contained in human face directly, adopting appropriate face description method, such as Gabor Wavelet, LBP operator, mapping them into the feature space to recognition is an effective approach to improve the recongtion peformance. A method for face description is proposed to solve the problem of the curse of dimensionality, which extracts the histogram sequence of local Gabor binary patterns from the magnitudes of Gabor coefficients. The proposed method is robust to illumination, expression and misalignment by combing the Gabor transform, LBP and spatial histogram.Meanwhile, the global profile characteristic of face is also important to face recognition, that the low-frequency information is usually regard as retained the overall contour of face, low-pass filter is adopted to retain the low-frequency information of the image, reducing the impact of local changes on the profile information. Finally, Fisher linear discriminant analysis based on Gabor transform which using local feature and global feature is used to cluster analysis of facial feature respectively. The main work is as follows:(1) The dissertation studied face recognition method based on histogram sequence of local Gabor binary pattern, and analysis the effect of different parameters on face recognition rate;(2) The dissertation improved the traditional model based on histogram sequence of local Gabor binary patterns, presented statistic Fisher weight on different areas of face, which have different contribution to face recognition;(3) According to the CMU latest research, the sharpness of face images is not proportional to face recognition rate, the dissertation proposed retained low-frequency information of images by Gaussian low-pass filter, that is, the contours information of the global face, which is needed in face recognition;(4) Finally, the dissertation fuse the local feature and the global feature of the human face, then, clustered of facial feature by Fisher discriminant analysis, and the weight of the global characteristics and the local characteristics is adjusted.A good recognition results on both the first and the fourth test of FRGC2.0face database show that the method has good recognition ability, and it is robust. The subject of this article is from the actual development project, and the proposed algorithm has been used as a part of the design for permission to enter by a tablet PC computer manufacturer, which to further validate its usefulness.