Enhanced facial feature tracking of spontaneous and continuous expressions.
Amr Goneid, Rana el Kaliouby · 2001
The integration of multimedia technologies into mainstream computing have both raised the user’s expectations of computer interfaces, and made possible the development of multi-modal emotionally intelligent systems. The true strength of facial expression recognition (FER) shows when seamlessly integrated into emotionally intelligent systems enabling applications to add facial expressions to traditional input modalities. FER systems must operate on natural human expressions in a real time environment. This paper presents efficient methodologies for tracking spontaneous facial expressions that are continuous in time. A model for natural facial expressions is developed, within which we propose three feature point tracking mechanisms that operate efficiently within that model; the adaptive facial feature tracking methodology developed to improve tracking results while decreasing the total number of computations, the predictive feature point tracking mechanism well-suited to large motions, and the nonexhaustive sum of squared differences (SSD) technique, an optimization on the SSD algorithm on which many tracking methodologies are based. Performance analysis carried out on different image sequences showed that adaptive and predictive mechanisms used in conjunction with the non-exhaustive version of the SSD yield up to a 1-10 increase in time efficiency, whilst improving tracking accuracy. These results help further the development of real-time natural facial expression recognition systems. 1.