SURF (Speeded Up Robust Features) feature extraction is currently more popular image registration method. This paper proposed a improved method based on block strategy on the basis of SURF. Firstly, using Watershed Algorithm to determine the number of image blocks; then the image was divided into blocks and each sub-block extracted a certain amount of feature points to realize uniform feature point extraction; then using sparse feature tree to find the matching feature points and finally using improved RANSAC algorithm to eliminate the error matching feature point pairs, while calculating transformation between the reference image and the image to be registered. Experiments show that this method can efficiently and quickly solve the problem of remote sensing image automatic registration.