Kernel Discriminant Anlaysis (KDA) and Kernel Principal Component Analysis (KPCA) are the nonlinear extensions of Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA) respectively. In this paper, we presented a feature extraction algorithm by combing KDA and KPCA to extract reliable and robust features for recognition. Furthermore, a generalized nearest feature line (GNFL) method was also presented for constructing powerful classifier. The performance of the proposed method was demonstrated through real data.