Too real for comfort? Uncanny responses to computer generated faces.

Area of Science:

• Human-Computer Interaction
• Computer Graphics
• Psychology

Background:

• Virtual humans are increasingly realistic, yet can evoke discomfort in human observers.
• The 'uncanny valley' describes this negative emotional response to near-human replicas.
• The underlying perceptual and cognitive mechanisms of the uncanny valley remain poorly understood.

Purpose of the Study:

• To investigate the perceptual, cognitive, and social mechanisms underlying the uncanny valley.
• To identify specific visual features that contribute to the uncanny valley effect in virtual humans.
• To propose design principles for mitigating the uncanny valley in virtual human creation.

Main Methods:

• Four studies systematically varied facial proportions, skin texture, and level of detail in computer-generated characters.
• Participants rated characters on perceived eeriness, human likeness, and attractiveness.
• Specific manipulations included texture photorealism, polygon count, facial proportion accuracy, and eye-face size/texture consistency.

Main Results:

• Increased texture photorealism and polygon count enhanced human likeness.
• Photorealism improved judgments of ideal facial proportions, but atypical proportions became more disturbing.
• Mismatches in eye size and texture relative to the face significantly increased perceived eeriness.

Conclusions:

• The uncanny valley is not solely a function of human likeness but influenced by specific perceptual discrepancies.
• Facial proportion accuracy and feature consistency (especially eyes) are critical for avoiding the uncanny valley.
• Design principles derived from these findings can help create more comfortable and acceptable virtual humans.