SIFT algorithm is a classic method of image matching, but there are large amount of calculation and high time complexity. To solve these problems, we put forward an improved SIFT algorithm in this study. According to the eight gradient directions, we divided the 128 dimensional data of the SIFT algorithm into eight groups, and redefined the key point information. According to the new key point information, it generates new order descriptors. In this way, it will reduce the amount of calculated quantities, so as to improve the efficiency of the algorithm. The experiment shows that the improved algorithm keeps the advantages of the original algorithm, and greatly improves the efficiency of the algorithm without reducing the precision of the original algorithm.