Decoding Emotions from Facial Animations

Facial feature point tracking is used as a primary input in several systems that perform affect recognition using facial expressions. To determine the efficacy of automatically extracted facial feature points in encoding emotional content, we conducted an experiment that compared human raters judgements of emotional expressions between actual video clips and three automatically generated representations of them (point-light displays, stick-figure models and 3D animations, Fig. 1). Although our main objective was to assess the utility of automatically extracted facial feature points in conveying emotions, our results give interesting insights into optimal representation of facial displays in emotion judgments as well as in analysing the perceptual quality and realism of the animations. Here we discuss results from our experiment that we consider are of relevance and interest to this symposium. Preliminary findings and details of the experiment are discussed in [1], while a more detailed description and analysis is in progress.