Study On Face Recognition Under Varying Illumination

Face recognition is one of the important biometric identification technologies. The main research topic on face recognition is how to make computer identify a specific person. The key issue of a successful face recognition approach is how to extract discriminated features from a face image. Within the past three decades, major advances have occurred in face recognition; numerous algorithms have been proposed for face recognition. Two of the most classical methods for this purpose are Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). More and more business face recognition systems appeared. Bur the recently FERET and FRVT indicate, that even if the best face recognition system, it performance reduces quickly when the database is larger, photography condition is uncontrolled and the user is non-cooperation. Illumination, background, pose, expression affect the performance greatly. So the illumination is the key problem of face recognition yet. It has been proven, both experimentally and theoretically that, in face recognition, variations caused by illumination are more significant than the inherent differences between individuals. Nevertheless, the performance of most existing methods including PCA and LDA is highly sensitive to illumination variation, it will be seriously degraded if the training/testing faces expose under severe lighting variations. Thus illumination variation is the most significant factor affecting the performance of face recognition, and has received much attention in recent years.In our study paper, the main work can be summarized as follows:Studied the two typical methods in face recognition: PCA and LDA, and identified the impact of the first eigenvectors by experiment.Experimented the image preprocessing techniques such as logarithmic transform, centering and gray-scale normalization, and identified how the combined effect face recognition rate.As we all know, face recognition usually cares about the main features of a face, such as the shapes and relative positions of the main facial feature, and ignores the illumination changes on the face. Accordingly, we propose a minimum squares error model to estimate the illumination by adaptive filtering,which can keeps the image edges very well. The proposed method transform images into logarithmical first then use the adaptive Wiener filtering to extract the facial feature invariant, and use PCA directly in the logarithm domain at last.Not only have Multi-scale analysis properties of wavelet transform, but also have multi approximate shift invariance, good directional selectivity, appropriate redundancy and computational efficiency properties. All the advantages of Dual-Tree Complex Wavelet Transform (DTCWT) make it a good candidate for representing the face features. 2D DTCWT can isolate edges with six orientations in different sub-bands, which avails to description of texture features in different orientations. Consequently, we purpose an approach which can exact the illumination invariant from the face images based on DTCWT. In this paper, the traditional de-nosing method of wavelet transform was inducted into the dual-tree complex wavelet transform using to extract the face contours and edges which get rid of the illumination.