A Method for Generating Facial Expressions Using Action Units

Facial expressions play an important role in emotional communication and impression formation. However, existing studies have often focused on a limited set of expressions, overlooking the complexity and variety of human facial displays. This study proposes a method to generate diverse facial expressions using the facial action system, focusing on the synergistic and mutually exclusive relationships between action units (AUs). Based on Russell's Circumplex Model, we synthesized 49 different expressions and conducted a validation experiment, achieving an 83% accuracy rate and providing valuable insights into the AU-emotion relationship. Our approach captures a broader range of expressions than those used in previous studies. In future work, we will investigate how these 49 expressions affect users' impressions, explore which expressions are considered "favorable" or "approachable" in specific contexts, and identify the defining characteristics required to produce optimal facial expressions in different scenarios.