Most of the existing face recognition algorithms are not ideal when th ey are trained in a single training sample.To solve this problem, we propose a feature extraction method based on Fourier-Mellin transform in the frequency domain invariance and the moment domain invariance.Firstly,the image brightness will be normalized to improve illumina tion variation.Then,the two algorithms are mixed to extract features,according to the analytical Fourier-Mellin transform (AFMT) invariance and the orthogonal Fourier-Mellin moment (OFMM) invariance.Finally,the hybrid algorithm is classified and merged by the neares t neighbor classifier (NNC) and the correlation coefficient method (CCM).Lots of experiments on YALE and ORL face databases show that the pe rformance of our proposed method is superior to that of the traditional face r ecognition algorithms.