Context Based Vision Emotion Recognition in the Wild

With the development of computer science and the increasing demand for personalized human-computer interaction, emotional computing plays an increasingly significant role in human-computer interaction. However, most researches at the present stage focus on facial expression recognition, which makes it difficult to recognize emotions in the wild. In this paper, a method of emotion recognition using both facial expression and the context is proposed to solve the difficulties of facial expression recognition such as occlusion and misalignment in the wild. A multi-head cross attention network (MHCAN) is proposed to distinguish more subtle changes in expression and improve the accuracy of emotion recognition. Our network is tested on CAER-S and RAF-DB datasets, and compared with advanced methods, the effectiveness of our method is proved.