Face morphing attacks: Investigating detection with humans and computers.

Area of Science:

• Biometrics
• Computer Vision
• Forensic Science

Background:

• Face morphing attacks utilize digital manipulation for fraudulent identification.
• High-quality morphs, averaging two faces, pose a significant security threat for official documents like passports.
• Previous research on low-quality morphs indicated poor human detection rates, though training showed potential improvement.

Purpose of the Study:

• To investigate human and computer performance in detecting high-quality face morphs.
• To assess the effectiveness of training on human morph detection capabilities.
• To compare human detection accuracy against automated systems for biometric security.

Main Methods:

• Conducted four experiments involving human participants and a computer model.
• Utilized high-quality morph images comparable to those used in criminal activities.
• Performed live matching tasks to evaluate real-world detection scenarios.

Main Results:

• Human participants demonstrated high error rates in detecting face morphs, with training proving ineffective.
• Face morphs were frequently accepted in live matching tasks, indicating a substantial security concern.
• A basic computer model significantly outperformed human participants in identifying morphs.

Conclusions:

• Advanced computational methods are more reliable than human training for combating face morphing attacks.
• The findings highlight critical implications for security agencies globally in addressing biometric fraud.
• Automated detection systems offer a promising solution to the growing threat of sophisticated identity manipulation.