Is global shape sufficient for automatic object identification?