Abstract: The principal of SIFT algorithm is researched in this paper.Due to the descriptor that one feature point needs 128 dimensions, a 60-dimension-square descriptor based on statistic local gradient information is taken forth.Comparing with the orignal one, the new descriptor expands the scope of neighborhood pixels;the ratio between the first and second closest distance is used to match the 60-dimension descriptors.According to the experiment, the results show that matching time is greatly shortened by 30%~60%;and the new descriptor is competitive with SIFT descriptor in effectiveness.Further more, the new descriptor exhibits good performance in more complicated images.Therefore, it is more suitable in real-time applications.