Towards real time facial expression recognition

Facial expression recognition is a key aspect in the synthesis of adaptive human machine interfaces. If advanced expression recognition techniques are developed, machines can tailor their response to the feelings of their users. In this paper, we describe an algorithm which integrates Optical flow analysis and Support Vector Machines (SVM) to classify eight facial expressions with accuracies up to 98.73%. Colour images fed into the system are pre-processed to accentuate the subject's eyes, eyebrows and mouth. An Optical flow analysis on two successive frames from a video sequence enables us to identify the dominant feature on the face, for a given expression. Further, feature extraction is carried out on the selected feature and the data from this is input into the trained SVM classifiers. The results obtained from the tests carried out, suggest that the system is robust in dealing with subjects of a variety of races and both genders. In the near future, the proposed technique will be incorporated in a real time system.