The full-search block-matching (FSBM) algorithm is an ordinary motion estimation method in image sequence stability. In traditional FSBM algorithm, every pair of pixels corresponding to the current and reference frame makes equal contribution to the matching result. Therefore, the performance of the traditional FSBM algorithm will tend to be influenced by various noise in the being stabilized image sequence. Aiming at the shortcoming, an improved FSBM algorithm is proposed. In the new matching rule, every pair of pixels is classified as matching pixel pair (set value as 1) or mismatching pixel pair (set value as 0). A threshold is selected to perform the above classification. The mismatching pixel pairs make no contribution to the result. Now the matching pixel pairs make equal contribution to the matching result as before, then multiply each by a weight that has relation with the position of the matching pixel pair and sum them up. The experiments show that this new method is promising.