Toward Believable Emotions: Evaluating FACS Coding for Virtual Human Expressions

In interactive computer graphics, the Facial Action Coding System (FACS) is widely used to enhance the emotional expressiveness of Virtual Humans (VHs). By linking specific Action Units (AUs) with facial blendshapes, animators can theoretically reproduce a wide range of human emotions. However, achieving realistic and believable emotional expressions remains challenging, as the same AU intensities do not work equally well across all VHs. This paper explores whether optimal sets of AU intensities can be defined for specific subgroups of VHs, such as those differentiated by gender and visual fidelity, rather than pursuing a one-size-fits-all approach. Through a focused analysis of happiness, sadness, and disgust, we demonstrate that visual fidelity plays a critical role in emotional clarity, while certain emotions require gender-specific adjustments. The findings emphasize the limitations of uniformly maximizing AU intensities across all VHs and offer practical insights for animators, providing a nuanced framework for creating believable emotional expressions in various types of VHs and enhancing realism in interactive applications.