Representing shapes for visual recongnition