A comparison of two computer-based face identification systems with human perceptions of faces

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Cognitive Science

Background:

• Facial recognition systems are crucial for human-computer interaction.
• Evaluating computer-based face representation against human perception is key for system development.
• Previous studies highlight the effectiveness of Principal Component Analysis (PCA) in image processing.

Purpose of the Study:

• To compare the performance of a graph-matching system and a PCA-based system for face representation.
• To assess how these systems align with human similarity ratings and memory performance.
• To determine which system better captures specific aspects of facial appearance.

Main Methods:

• Comparison of a graph-matching system and a PCA-based system using a standardized set of face images.
• Evaluation against human similarity ratings, distinctiveness judgments, and memory recall.
• Analysis of system performance under varying conditions, including altered facial expressions and visibility of hair.

Main Results:

• The PCA-based system demonstrated superior face recognition performance and higher correlation with human performance after image standardization.
• Both systems showed comparable ability to recognize faces with changed expressions.
• Correlations were found between the similarity measures of the two computer models.

Conclusions:

• The PCA-based system excels in recognizing faces and correlates well with human performance, particularly with standardized images.
• The graph-matching system appears to capture nuanced facial appearance better than the PCA-based system.
• Both systems have distinct strengths, suggesting complementary roles in advanced facial representation research.