Shape-invariant encoding of dynamic primate facial expressions in human perception

Area of Science:

• Primate communication
• Comparative psychology
• Neuroscience

Background:

• Dynamic facial expressions are vital for primate communication.
• The interaction between facial shape and expression dynamics in perception is not well understood.
• Neural network models often assume joint encoding, but slower evolution of facial motor control suggests separate encoding.

Purpose of the Study:

• To investigate how species-specific facial expressions are perceptually encoded.
• To determine if facial shape and dynamics are encoded jointly or separately.
• To test hypotheses regarding the co-evolution of facial shape and motor control.

Main Methods:

• Developed photo-realistic human and monkey heads for animation.
• Utilized motion capture data from humans and monkeys for realistic expressions.
• Employed a Bayesian machine-learning technique for precise control over expression dynamics.

Main Results:

• Human observers rapidly learned to recognize cross-species facial expressions.
• Facial dynamics were represented largely independently of facial shape during perception.
• Findings challenge neural network models predicting joint encoding of shape and dynamics.

Conclusions:

• Supports the hypothesis of co-evolution between visual processing and motor control of facial expressions.
• Suggests that facial dynamics are encoded separately from facial shape.
• Challenges current appearance-based neural network theories for dynamic facial expression recognition.