Spatial and temporal relations between different facial muscles are very important in the facial expression recognition process.However

these implicit relations have not been widely used due to the limitation of the current models.In order to make full use of spatial and temporal information

we model the facial expression as a complex activity consisting of different facial events.Furthermore

we introduce a special Bayesian network to capture the temporal relations among facial events and develop the corresponding algorithm for facial expression modeling and recognition.We only use the features based on tracking results and this method does not require the peak frames

which can improve the speed of training and recognition.Experimental results on the benchmark databases CK+ and MMI show that the proposed method is feasible in facial expression recognition and considerably improves the recognition accuracy.