Dynamic Facial Expression Recognition Based On K-order Emotional Intensity Model

With the development of artificial intelligence and pattern recognition, facial expression recognition plays a more and more important role in intelligent human-computer interaction. Actually, facial displays of emotion are highly dynamic phenomenon, which evolve over time from the onset, the apex and the offset can be more effectively reflected that the essence of facial expression process by the dynamic features which are extracted from dynamic expression sequences. Moreover, dynamic expression sequence contains more information about facial expresstions. Consequently, the application of dynamic facial expression recognition is extensive, and it has more practical significance. As a kind of property of facial expression, emotional intensity has been thought an effective way for dynamic facial expression recognition. However, how to measure the value of emotional intensity is a problem. The former reseachers commonly used the anthropogenic emotional intensity values in their work. In this thesis, we design a model named K-order emotional intensity model. Different from other related works, the proposed approach can divide emotional intensity into several segments in an unsupervised way. And then the output from K-EIM is encoded by the coding rule which is used for the dynamic facial expression recognition. As the number of frames that the dynamic facial expression sequence contains is not considered, the proposed approach is convenient in practical use. The experiments are conducted on Cohn-Kanade facial expression database and the support vector machine classifier is used for facial expression classification. The experiments contain closed experiment and open experiment. This method achieved a dynamic facial expression recognition accuracy of 88.32% which suggest that the proposed method shows better performance and proves its validity. Moreover, effect of coding and no coding is also discussed in the paper.