Abstract: Considering the low efficency and mismatching in SIFT\|based multi\|source remote sensing image registration, we improve the SIFT algorithm through two aspects of point feature extraction and point feature matching. Firstly, we acquire an appropriate number of well-distributed point features by controlling their number and distribution. Secondly, an optimized strategy is adopted to realize a coarse\|to\|fine feature point matching approach by applying the initial affine matching, precise matching and mismatching elimination processes. Experiments on several multi\|source data sets show that the proposed algorithm performs better than the SIFT algorithm on both efficiency and accuracy.