Audio-Visual Compound Expression Recognition Method based on Late Modality Fusion and Rule-based Decision.