Deliberate Occlusion: Asymmetry in Facial Expression

Facial expression recognition (FER) has become increasingly common in research. However, region feature technologies that play an important role in this field have not been well developed. In this study, a psychology-based solution to address problems in FER of regional features was proposed. This method employs the use of deliberate occlusion for enhancement and attention module to find meaningful and important features in the regions processing features with high inter-class similarity and variation. The result of this study shows that the approach has higher accuracy than recent models and, through the use of feature heatmap visualization, the possibility of approaching FER from human facial asymmetry is raised.