Emotion Research Based On Facial Expression Recognition

With the continuous development of Artificial Intelligence, It is growing hope that the computer can have emotional capacity in the human-computer interaction process. The facial expression recognition is an important part of affective computing and intelligent human-computer interaction, related to computer vision, pattern recognition, psychology and other challenging field. This paper studies some difficulties about automatic facial expression recognition, and puts forward the emotional model described by the theoretical equation of self-organization. Main work is as follows:(1)We study uniform model LBP operator and rotation invariant LBP operator, the different characteristics of uniform model LBP, and compare with the facial expression performance characteristics based on two different LBP operators. Improve the sign function in LBP algorithm, and propose a method that disposes the image from coarse to fine, and combines with the local and overall histogram. That method can contain expression information.(2) We study preservation of the local projection (LPP) algorithm that reduces the extracted histogram to the LPP space dimension. The recognition process is simple and only calculated in low-dimensional space, reducing the computation time. Comparing with the SVM method of different inner product kernel function, we know the radial basis function is best. Also we construct fifteen one by one SVM classifies for face expressions classification. The method together LBP and LPP is better than Gabor filtering and LBP histogram intersection.(3) Emotional model is the key component to achieve effective human-computer interaction. In the process of modeling to study model, we study the basic emotions theory and the self-organization structure theory, putting forward the method with personality, mood and needs. Then we simulate the emotional decay and the relationship among basic emotions, complex emotions and internal driving force. (4) We design a program of human-computer emotional interaction. It is verified through the ASR robot in different facial expression databases. And we integrate the emotion model to simulate the change of emotional state.