Abstract
Face Recognition is one of the most important technology of biometrics. The purposes of this thesis are to develop a robust face image retrieval system and propose a method to accomplish face recognition whose images may suffer from some deformations. A scheme of face image retrieval is proposed for face recognition and facial expression recognition. We utilize a 2D Gabor wavelet representation to be an alternative form of face image. After feature extraction by Gabor Wavelet Transform, two dimensionality reduction methods - PCA and LDA - are sequentially applied on the feature vector. At last, we use a KNN classifier to recognize face image and facial expression in lower dimensional space, respectively. And we also try another approach, which is so called elastic graph matching or dynamic link architecture, to conquer the image deformations of the face recognition system. Finally, some results and discussions are reported between the two approaches.