A Study On Face Recognition Algorithm Based On Multi-resolution And Multi-direction Feature

Face recognition,as a biological feature recognition,has been one of the most active research areas in computer vision,pattern recognition and biometrics.Face recognition has many advantages over other biometrics modalities.Besides being natural and nonintrusive,the most important advantage of face recognition is that it can be captured at a distance and in a friendly manner,no need cooperation of people.In recent years,with the rapid development of computer technology,face recognition technology has been widely researched and developed,various face recognition algorithms have been devised in the literature.The existing face recognition systems can obtain satisfactory results under the conditions of user cooperation and ideal environment.However,face recognition is still faced with a number of challenges such as varying illumination,facial expression,pose,occlusion and so on.Feature extraction is the most key step of a face recognition system,which can provide an effective representation of face images to decrease the computational complexity and make for classification.A good feature extraction method can greatly improve the recognition rate of a face recognition system.This dissertation discusses and establishes effective feature extraction methods based on two multi-resolution and multi-direction features,wavelet transform and Gabor wavelet transform.The main contributions are summarised as follows:(1)We propose a novel technique,called TWSBF+PCA,which is a joint of pixellevel and feature-level fusion at the top-level's wavelet sub-bands for face recognition with the help of principal component analysis(PCA).The traditional wavelet-based approaches directly use the low frequency sub-band of wavelet transform to extract facial features.However,the high frequency sub-bands also contain some important information corresponding to the edge and contour of face,reflecting the details of face,especially the top-level's high frequency sub-bands.We convert the problem of finding the best pixellevel fusion coefficients of high frequency wavelet sub-bands to an optimization problem with the help of PCA and propose an alternating algorithm to solve the corresponding optimization problem for finding transformation matrices of dimension reduction and optimal fusion coefficients of the high frequency wavelet sub-bands simultaneously.The proposed method makes full use of four top-level's wavelet sub-bands rather than the low frequency sub-band only.The experimental results indicate that the proposed method is effective and robust.(2)We propose an improved technique,called TWSBF+LDA,which is a joint of pixel-level and feature-level fusion at the top-level's wavelet sub-bands for face recognition with the help of linear discriminant analysis(LDA).Firstly,we use the idea of LDA to find the best pixel-level fusion coefficients of high frequency wavelet sub-bands which can make full use of the labels of face images and retain more discriminative information.Secondly,we combine pixel-level fusion and feature-level fusion to avoid the loss of high frequency information.Thirdly,we convert finding the best fusion coefficient vector to an optimization problem and propose an alternating algorithm to solve the corresponding optimization problem for finding transformation matrices of dimension reduction and optimal fusion coefficients simultaneously.The experimental results demonstrate TWSBF+LDA has great improvement compared with TWSBF+PCA.(3)We propose a novel face recognition method based on local binary pattern(LBP)preprocessing and wavelet transform.Firstly,a given face image is processed by the LBP operator,and an LBP image is obtained.Then wavelet transform is used to extract discriminate feature from the LBP image.The experiment results on LFW,Extended YaleB and CMU-PIE face databases show that the proposed method outperforms several popular face recognition methods,which solves the problem of face recognition under variational illumination to some extent.(4)We propose a new feature from the different direction resulted sub-images of Gabor wavelet transform corresponding the same scale inspired the idea of LBP.The new feature,Direction-Gabor LBP(D-GLBP),together with its extensions make full use of the relationship among the resulted sub-images of Gabor wavelet transform,overcome the drawback of losing information after down-sampling and robust to noise and illumination changes.The numerical results show that D-GLBP and its extensions are powerful tools for feature extraction in face recognition.(5)We propose a novel face descriptor,the Scale-Gabor LBP(S-GLBP),to explore the holistic and neighboring relationships of the different scale resulted sub-images corresponding the same direction after Gabor wavelet transform.Compared with the existed methods,S-GLBP utilizes the deep relations between neighboring Gabor sub-images instead of directly combining Gabor wavelet transform and LBP.In addition,using the Gabor filter with fewer scale achieves better performance which not only reduces the computational complexity but also improves the recognition rates.The experimental results show that the proposed method outperforms other related face recognition methods.