Capture expression-dependent AU relations for expression recognition

Jun Wang, Shangfei Wang, Mengdi Huai, Chongliang Wu, Zhen Gao, Yue Liu, Qiang Ji

Abstract: To date, there is only limited research that explicitly exploits the relationships among Action Units and expressions for facial expression recognition. In this paper, we propose an facial expression recognition method through modeling the expression-dependent AU relations. First, the incremental association Markov blanket algorithm is adopted to select crucial action units for a certain expression. Second, a Bayesian Network (BN) is constructed to capture the relationships between a certain expression and its crucial action units. Given the learned BNs and measurements of AUs and expression, we can then perform expression recognition within the BN through a probabilistic inference. Experimental results on the CK+ and MMI databases demonstrate the effectiveness and generalization ability of our method.