Dynamic recognition of 6 basic facial expressions by discrete-time recurrent neural network

In order to devolop an "Active Human Interface" that realizes emotive interactive communication between machine and human being, we have been investigating the recognition of human emotion from facial expressions. This paper deals with the machine recognition method of 6 basic dynamic facial expressions. First, we carry out psychological recognition tests of 6 basic dynamic facial expressions for 3 subjects to clarify the characteristics of human dynamic recognition of facial expresions. Then, we construct a recurrent neural network (RNN) equipped with a feedback layer and a differential input layer for developing the machine recognition method. Using the dynamic data of each facial expression and the human recognition result for 2 subjects, we train the RNN and recognition tests are performed by inputting the dynamic facial data of one subject not used in training the RNN. By comparing the results obtained with RNN to those of human recognition, the RNN is found to perform very similar dynamic recognition to that exhibited in human recognition of facial expressions.