Abstract: A dynamic method for compressing still images is presented in the paper. It decomposes a still image into a series of continuous subimages by a permutation transform, then compresses the subimage sequence by an MPEG encoder. Four types of scan paths are used to sequentialize the subimages, and the Hilbert curve is proved to be the best one due to its perfect local clustering property. The overall performance of the presented method is compared with that of the conventional JPEG method. The quality of the decoded image can be improved by a value of about 0.6dB in PSNR under the same compression ratio 66:1.