Abstract: By analyzing the major image registration techniques at present, a new image registration method based on image entropy on account of the distribution issue of feature point and the registration of corresponding points was introduced. First, the image was divided into blocks to a certain extent and the image entropy of each block, which reflected the texture transformation within the block, was computed according to the information theory. The rough-match was then made on the basis of the computed image entropy. After that, a certain number of feather points were extracted from each block. The more information content the block had, the more abundant the texture became and so the larger extraction number we got. The precise match was made with these typical corresponding points. To demonstrate the validity of the proposed method, the improved image registration technique was compared to conventional methods on same images.