Leveraging Computer Vision Face Representation to Understand Human Face Representation.

Area of Science:

• Computer Vision
• Psychology
• Human Face Representation

Background:

• Face processing is crucial for social interactions, influencing judgments of trustworthiness, attractiveness, and emotional states.
• Understanding human face representation is key to advancing fields like artificial intelligence and social cognition.

Purpose of the Study:

• To investigate human face representation using computer vision and human similarity assessments.
• To develop a computational model that accurately predicts human judgments of facial similarity and social/affective traits.

Main Methods:

• Utilized computer vision techniques, including shape- and texture-based Active Appearance Models.
• Employed metric learning to enhance facial feature representation.
• Validated model performance against human similarity judgments and social/affective assessments.

Main Results:

• The combined model achieved superior performance in predicting human similarity judgments compared to other algorithms and humans.
• The model demonstrated strong performance in modeling social trait (trustworthiness, attractiveness) and affective (happy, angry, sad) assessments.
• Identified that facial similarity relies on 8-12 features, with race and gender being prominent.

Conclusions:

• Facial similarity perception is influenced by a limited set of features, including race and gender.
• While similarity features are important, they are insufficient alone for complete psychological face representation.
• Affective features, not captured by similarity judgments alone, are necessary for a comprehensive understanding of face representation.