On retrieving information from visual images