Human Face Recognition Using Generalized Kernel Fisher Discriminant and Wavelet Transform

In this paper the generalized kernel Fisher discriminant (GKFD) method is used to do pattern feature extraction for human face image. First, we extend the KFD originally used in pattern classification problems to the generalized KFD (GKFD), which will be used in feature extraction problems. Compared to several commonly used feature extraction methods, the GKFD can not only reduce the dimension of input pattern, but also provide the useful information for pattern classification. Further, this GKFD also performs well for linearly nonseparable pattern classification problems for it possesses a nonlinear transformation capability. To reduce the computation complexity, the original face images are pre-processed by wavelet transform. Finally, the experimental results on human face recognition problems demonstrate the effectiveness and efficiency of our approach.