The inaccuracy and insincerity of real faces

Since conversation is a central human activity, the synthesis of proper conversational behavior for Virtual Humans will become a critical issue. Facial expressions represent a critical part of interpersonal communication. Even with the most sophisticated, photo-realistic head model, an avatar who's behavior is unbelievable or even uninterpretable will be an inefficient or possibly counterproductive conversational partner. Synthesizing expressions can be greatly aided by a detailed description of which facial motions are perceptually necessary and sufficient. Here, we recorded eight core expressions from six trained individuals using a method-acting approach. We then psychophysically determined how recognizable and believable those expressions were. The results show that people can identify these expressions quite well, although there is some systematic confusion between particular expressions. The results also show that people found the expressions to be less than convincing. The pattern of confusions and believability ratings demonstrates that there is considerable variation in natural expressions and that even real facial expressions are not always understood or believed. Moreover, the results provide the ground work necessary to begin a more fine-grained analysis of the core components of these expressions. As some initial results from a model-based manipulation of the image sequences shows, a detailed description of facial expressions can be an invaluable aid in the synthesis of unambiguous and believable Virtual Humans