sparse representation for computer vision and patt