Computing mutual similarity of 3D human faces in nearly linear time

Area of Science:

• Anthropometry
• Computer Vision
• Biometrics

Background:

• Three-dimensional (3D) facial scanning is increasingly used in human variation studies.
• Quantitative facial similarity assessment complements traditional somatic trait measurements.
• Existing automated registration and similarity algorithms face computational limitations in batch processing.

Purpose of the Study:

• To develop a rapid and accurate approach for batch processing of 3D facial scans.
• To overcome the quadratic complexity limitations of N:N analyses.
• To enable efficient quantitative assessment of facial similarity in large datasets.

Main Methods:

• Utilized properties of facial scan geometry for optimized registration and similarity measurement.
• Developed an algorithm with nearly linear time complexity.
• Implemented automatic handling of mesh artifacts such as holes.

Main Results:

• Achieved a significant reduction in computation time compared to quadratic-time methods.
• Demonstrated high accuracy comparable to precise baseline approaches.
• Successfully processed large datasets efficiently.

Conclusions:

• The proposed method offers a practical solution for large-scale 3D facial scan analysis.
• Enables more efficient quantitative studies of human facial variation.
• Addresses computational bottlenecks in biometric and anthropometric research.