Representation for computer vision