Although research show that human recognition performance for unfamiliar faces is relatively poor, when the sample is always available for analysis and becomes ”familiar”, people are able to recognize a previous unknown face from single sample. In this paper, a method is proposed to deal with the one sample per person face recognition problem based on the process how unfamiliar faces become familiar to people. Particularly, quantized local features which learnt from generic face dataset are used in the proposed method to mimic the prototype effect of human face recognition. Furthermore, a landmark-based scheme is introduced to quantify the distinctiveness of each facial component for the sample face, then the difference between the sample and the average face is emphasized by weighting face regions according to the gained distinctiveness. The experiments on ORL and FERET face databases demonstrate the efficiency of the proposed method.