Abstract: Facial pose under depth rotation is a tough problem in human face recognition. In this paper, it is firstly discussed that orientation and size of Gabor filter have the optimal capability to discriminate face pose. Secondly, a Gabor-based KFDA(kernel Fisher discriminant analysis) method coupled with fractional power polynomial model is proposed to serve as the face pose classifier. Finally, an integrity human face synthesis method is presented based on a new full-correspondence algorithm and linear object classes. The experiment results show that the proposed method is efficient.