Learning structured ordinal measures for video based face recognition.