Abstract: As a method to measure the spatial properties of image, the Gaussian kernel with different parameters was used to get the difference in original SIFT algorithm, while the difference in the spectral dimension of optical system was used in the proposed method. Comparing with traditional sift method and a lot of improved methods, counting the image block pixel information only in the neighborhood around the feature points, the new method was arranged to divide the matching process into two steps. Firstly, the image block pixel information got from the neighborhood of the feature points was rough matched. And then four matching pairs with the highest similarity were selected as the benchmark matching pairs, and the feature points were checked twice. The simulation results show that the proposed method can significantly increase the number of detected feature points and effectively eliminate the error matching.