Abstract: This paper presents a method of selective coding of still images using an semi-automatic extraction of Region of Interest (ROI). To facilitate the extraction of ROI which are characterized by complicated arbitrary shapes, the image is rst segmented in regions using photometric clustering of the pixels based on the Competitive Agglomeration method. These regions are dilated to eliminate the small and nonsignicant areas. Extracted ROIs are thereafter coded losslessly using JPEG 2000 algorithm while the rest of image (background BG) is coded with losses. Performances of the approach in terms of ROI extraction eectiveness and compression ratio are compared through its applications to a set of color images containing ROIs of various natures.