using features automatically derived from the video content