Multi-scale joint feature network for micro-expression recognition

Abstract Micro-expression recognition is a substantive cross-study of psychology and computer science, it has wide range applications (e.g., psychological clinical diagnosis, emotional analysis, criminal investigation, etc.). However, the subtle diverse changes in facial muscles make difficult for existing methods to extract effective features, which limits improvement micro-expression accuracy. Therefore, we propose multi-scale joint feature network based on optical flow images recognition. First, generate an image that reflects motion information. The then fed into extraction classification. proposed module (JFM) integrates features from different layers, beneficial capture with amplitudes. To improve ability model, also adopt strategy fusing prediction results three JFMs backbone network. Our experimental show our method superior state-of-the-art benchmark datasets (SMIC, CASME II, SAMM) combined dataset (3DB).