Choosing a distinctive feature and matching criterion is key to developing a reliable face recognition system. This paper discusses the availability of one of geometric feature invariants

scale invariant feature transform (SIFT) descriptor based face recognition. The SIFT feature description of an image is typically complex. In most cases

the difficulty of feature matching problem is aggravated when the different face expressions and image blur exist. For abovementioned issues

in this paper we proposes a new method that six interest sub regions from the face are selected to be described and later be calculated through different weights according to their distinctiveness. The square of the similarity is used to solve the problem of data deviation. The experimental results demonstrate that our method does effectively moderate the face expression effect. It also successfully reduces the complexity and matching time of SIFT feature sets.