Quantifying human sensitivity to spatio-temporal information in dynamic faces

Area of Science:

• Cognitive Psychology
• Neuroscience
• Computer Vision

Background:

• Facial motion conveys significant perceptual and social information.
• Understanding how humans process complex spatio-temporal facial expression dynamics is crucial for human-computer interaction and artificial intelligence.

Purpose of the Study:

• To investigate human sensitivity to spatio-temporal information in facial motion.
• To identify the specific cues observers use to judge the similarity of dynamic facial expressions.
• To develop a quantitative method for explaining perceived similarity in facial motion.

Main Methods:

• Motion capture of four facial expressions.
• Decomposition of expressions into time courses of local facial actions.
• Generation of approximated facial motion time courses with varying information content.
• Animation of an avatar head using original and approximated time courses.
• Observer tasks to rate similarity between animations.
• Development of objective stimulus similarity measures.

Main Results:

• Observers preferred animations that preserved more information about natural facial motion dynamics.
• The temporal dynamics of facial actions (e.g., onset, peak timing) better explained similarity judgments than image-based measures like optic flow.
• Sensitivity to natural facial dynamics was demonstrated.

Conclusions:

• Human perception of dynamic facial expressions relies on sensitivity to natural motion dynamics.
• Sparse, meaningful spatio-temporal cues are critical for processing facial motion.
• The developed method provides a quantitative explanation for perceived similarity in dynamic facial expressions.