A fully-automated system for facial muscle action

Typically, facial expressions convey conversational signals and signs of attitudinal states—such as interest and boredom—that are usually displayed as one or few facial muscle actions, such as raising the eyebrows in disbelief. Instead of classifying facial expressions into a few basic emotion categories, we attempt to measure a large range of facial behaviour by recognizing facial muscle action units (AUs) that produce expressions. As described in the FACS,2 all visually-distinguishable facial activity can be described in terms of 44 AUs. Examples include the inner brow raise (AU1), outer brow raise (AU2), brows pulling together (AU4), and so forth. Thus, if a computer system were able to detect these 44 AUs automatically, it could identify each and every facial expression that the human face can possibly display. Figure 1 outlines our system for recognition of 27 In this issue: