Optical flow for dynamic facial expression recognition.

Abstract

Facial expressions are essential in human face-to-face communication because they reflect emotion and are a form of nonverbal communication. The face of a human has the most relevant visual characteristics of personality and emotion. We developed and analyzed a real-time fully automatic facial expression classification system that uses sequences of frames describing the dynamics of the expressions. This includes face detection, face registration using landmark points of the face, feature extraction using optical flow and finally classification between six basic expressions. Extensive experiments on the Cohn-Kanade database illustrate that this approach is effective for facial expression analysis. The contributions of this paper are twofold: (1) to have a fully automated way to measure spontaneous facial expressions and (2) to incorporate optical flow to cope with temporal information to enhance the overall recognition results. The results of the classification indicate that the correct alignment has a significant effect on the accuracy of the system.