In order to extract more real sample local distribution structure and rationally use samples labels information

a face recognition algorithm based on local Fisher criterion discriminant projections is proposed. The proposed algorithm adaptively choose neighbor parameter through solving sparse representation of samples under totality

which can make the distribution relationship between samples obey the reality. On the basis of obtaining sparse neighbor structure

the customized intra-class local scatter and inter-class scatter matrix are designed by information of sample labels

which can preserve neighbor relationships and improve discriminant power brought by sample label information. The proposed algorithm can maintain their intrinsic neighbor relations of the same class

whereas neighboring structure of samples which are different classes. The experimental results on benchmark face images datasets（Yale

AR and Yale B） show that the proposed algorithm has higher face recognition rate compared with the related face recognition algorithm

and can effectively improve the recognition rate of the face recognition algorithm.