Facial expression recognition(FER)is an important means of detecting human emotions and is widely applied in many fields,such as affective computing and human-computer interaction.Currently,several methods for FER heavily rely on large amounts of manually labeled data,which are costly and not available in real-world applications.To address this problem,this paper proposes a semi-supervised method based on the deep difference features.First,a cascaded structure is introduced to the original safe semi-supervised SVM(S4VM)to solve the multi-cl...