Yan Zhang, Xiangyang Feng and Ming Zhu, Donghua University, China

Facial expressions can express different emotions. Similar facial expressions usually correspond to the same emotions, and the changing process of emotional states is reflected in the dynamic changes of facial expressions. However, existing studies mainly focus on instantaneous emotional states, which cannot reflect the intensity of emotions. This paper proposes a method to study the process of emotional change based on dynamic expression similarity, which can assess not only the change of emotional state but also the change of emotional intensity. First, the features of dynamic expressions are extracted based on the VGG16 network model. Then, the cosine similarity of the expression features is calculated to match the corresponding emotions. At the same time, the expression intensity of each frame is calculated to evaluate the change in emotional intensity. The experimental results show that the similarity calculated in this paper is increased by 9.7% on average, which can be used for the study of emotional states.

Dynamic Expression Similarity, The Emotional State Change, Emotional Intensity.