Real-time facial expression recognition based on features' positions and dimensions

This paper describes a method of real-time facial expression recognition which is based on automatic measurement of the facial features' dimension and the positional relationship between them. The method is composed of two parts, the facial feature extraction using matching techniques and the facial expression recognition using statistics of position and dimension of the features. The method is implemented in an experimental hardware system and the performance is evaluated. The extraction rates of the facial-area, the mouth and the eyes are about 100%, 96% and 90%, respectively, and the recognition rates of facial expression such as normal, angry, surprise, smile and sad expression are 54%, 89%, 86%, 53% and 71%, respectively, for a specific person. The whole processing speed is about 15 frames/second. Finally, we touch on some applications such as man-machine interface, automatic generation of facial graphic animation and sign language translation using facial expression recognition techniques.