Studies On The Application Of Wavelet Theory In Face Recognition

The biologic character identification is regarded as one of the most reliable and safest person identification techniques since the peculiar individual features of mankind are applied to verify personal identity. Face recognition, as an important branch of biometric features recognition utilizes multi-disciplinary knowledge, including image processing, signal processing and pattern recognition. The researches on the face recognition need to answer various challenges, such as the huge volume of face images, the high dimension vector property, head pose, illumination etc. In addition, the real face recognition system requires not only identifying the target from the images with distortion and noises, but also improving the recognition rate and reducing the time complexity. All these requirements heighten the difficulty for face recognition.In the dissertation, the face recognition theory and methods, the influence of illumination on face recognition, illumination invariant extraction and the application of wallet-based theory to face recognition are investigated systematically. A novel and interesting combination of uncorrelated linear discriminant analysis and wavelets is presented to extract features for face recognition; illumination invariant extraction method is proposed to deal with the illumination problem based on the wavelet transformation and denoising model. More importantly, we propose a novel wavelet based approach that considers the correlation of neighboring wavelet coefficients to extract an illumination invariant for face recognition and bimodal processing approaches for training images and testing images were introduced for the first time.Fisher discriminant analysis is an important method for features extraction. Uncorrelated linear discriminant analysis (ULDA) is a very famous one among Fisher discriminant analysis. However, small sample issue is still a problem when dealing with Fisher discriminant analysis, specifically for high resolution, high dimension images. To cope with this problem, uncorrelated linear discriminant analysis and wavelets were combined to extract features for face recognition. In the proposed technique, the face images are divided into smaller sub-images by 2-D DWT and the uncorrelated linear discriminant analysis is applied to approximations sub-images. The time-cost of the proposed method is greatly reduced and recognition rates ranging between 95% and 97.5% are obtained on the ORL database. An average error rate of 1.4% is obtained on the NUST603 database. The proposed algorithm has an improved recognition rate and a decrease in computational load for face images with high resolutions when compared with conventional FSD+PCA method and ULDA+PCA method. The effect of number of discriminant vectors on the recognition system was systematically discussed.The features of a face can change drastically as the illumination changes. In contrast to pose position and expression, illumination changes present a much greater challenge to face recognition. Thus, the recognition of frontal facial appearance with illumination is a difficult task. In this study, based on the investigations on the theories of the wavelet transformation and denoising model, we present an illumination invariant extraction method to deal with the illumination problem. The illumination invariant is extracted in wavelet domain by using wavelet-based denoising techniques. This invariant represents the key facial structure needed for face recognition. The 2D wavelet transformation is applied to the images with illumination in the logarithmic domain to separate low and high frequency wavelet coefficients. Through manipulating the high frequency wavelet coefficient combining with denoising model, the edge features of the illumination invariants are enhanced and more useful information is restored in illumination invariants, leading to an excellent face recognition performance. Experimental results on Yale face database B and CMU PIE face database show that satisfactory recognition rate can be achieved by the proposed method.By studying the relationship between wavelet coefficients and the illumination invariant, we found that the correlation of neighboring wavelet coefficients (NWC) plays a very important role in illumination invariant extraction. Thus, we propose a wavelet based approach that considers the correlation of neighboring wavelet coefficients to extract an illumination invariant. This method has better edge preserving ability in low frequency illumination fields and better useful information saving ability in high frequency fields by using wavelet based NeighShrink denoise techniques. This method applies different process approaches to training images and testing images since these images always have different illuminations. More importantly, by having different processes, a simple processing algorithm with low time complexity can be applied to the testing image. This leads to an easy application to real face recognition systems. Experimental results on Yale Face Database B and CMU PIE Face Database show that through optimizing the soft thresholding value, parameterλ(0.9-1.2), as high as 100% of recognition rate can be obtained for various training sets and testing sets. Finally, the validities of NWC-based illumination invariant extraction method and bimodal method were discussed.