In this paper, we propose a new algorithm, feature perception algorithm (FPA), to deal with the problem that a single existing feature extraction algorithm cannot obtain the algebraic and geometric characteristics at the same time, and construct a non-linear feature perception (NLFP) by improving the linear ones, to perceive the feature. Meanwhile, based on the feature analysis model of cognitive psychology, we create the feature perception classification model (FPCM) and put forward a multi-pattern classification algorithm (MPCA), which is applied to human facial expression recognition problems, achieving good recognition rate.