Computing image translations
		What do we do about the “bad” matches?