The proposed research is aimed at constructing an image representation that improves performances of current image compression algorithms and organizes the information in a form suitable for fast patter discrimination and search algorithms. The approach relies on recently discovered bandlelet orthogonal bases, which can be adapted to curves and efficiently represent image profiles. The work will also build on and incorporate previous work in such areas as graphs matching. Developing new techniques in this area can greatly improve the efficiency of future image search engines.