Pose and Expression Recognition Using Limited Feature Points Based on a Dynamic Bayesian Network

Abstract : In daily life, language is an important tool during the communications between people. Except the language, facial actions can also provide a lot of information. Therefore, facial actions recognition becomes a popular research topic in Human-Computer Interaction (HCI) field. However, it is always a challenging task because of its complexity. In a literal sense, there are thousands of facial muscular movements many of which have very subtle differences. Moreover, muscular movements always occur spontaneously when the pose is changed.To address this problem, firstly we build a fully automatic facial points detection system based on local Gabor filter bank and Principal Component Analysis (PCA). Then the Dynamic Bayesian networks (DBNs) are proposed to perform facial actions recognition using junction tree algorithm over a limited number of feature points. In order to evaluate the proposed method, we have applied the Korean face database for model training, and CUbiC FacePix, FEED, and our own database for testing. Experiment results clearly demonstrate the feasibility of the proposed approach.