Recently, remote sensing images increase greatly in the repositories. To achieve efficient storage and indexing, we introduce a JPEG2000 compression frame which puts emphasis on selection and designation of regions of interest (ROI). Firstly, SIFT is applied for feature extraction, descriptor generation and point matching to locate ROI. Secondly, an approach is proposed for computing transformation parameters such as rotation, scale and translation separately. To get robust results, we put forward an improved RANSAC algorithm with two cost functions to obtain comparative computation efficiency in our application. At last ROI designated is used in JPEG2000 for remote sensing image compression. The method performs great in a series of experiments and satisfying results can be seen.