Face Expression Recognition using Recurrent Neural Networks.

The development of face expression recognition (FER) technology is a topic of active research, because in the several applications, it is necessary to verify the emotion expression of a given person. As a result, several face expression recognition schemes have been proposed during the last decades. The development of face expression recognition schemes is mainly based on the face image analysis, instead of video sequence analysis. However, considering that a human recognizes an emotion of other person using his/her time video sequence data, it is natural to use video analysis instead of image analysis. This paper proposes a face expression recognition system using video sequences, in which temporal feature vectors are extracted from the face landmarks, which are relevant characteristic points on the face. These feature vectors are then introduced into the different configurations of recurrent neural networks (RNN), such as Long Short Time Memory (LSTM) and Bidirectional Long Short Time Memory (BLSTM) to obtain better results.