Fraudulent ID using face morphs: Experiments on human and automatic recognition.

Area of Science:

• Cognitive psychology
• Computer vision
• Forensic science

Background:

• Face matching is crucial for security but challenging.
• Identity fraud can exploit difficulties in recognizing unfamiliar faces.
• Graphical face morphs on identification documents present a novel fraud method.

Purpose of the Study:

• To investigate the vulnerability of human viewers and smartphone systems to face morphs in photo identification.
• To assess the impact of brief training on human morph detection capabilities.
• To compare the error rates of humans and a smartphone face recognition system when encountering morphed IDs.

Main Methods:

• Two experiments with human participants matching face pairs, with and without prior morph detection training.
• A third experiment using a smartphone face recognition system to assess morph acceptance.
• Analysis of acceptance rates for morphed versus genuine faces.

Main Results:

• Without training, human viewers frequently accepted morphed faces as valid identification.
• Short training significantly reduced morph acceptance by humans, but large individual differences remained.
• Smartphone face recognition systems exhibited error rates comparable to trained human viewers, accepting some morphs.

Conclusions:

• Face morphs pose a significant security risk for identity verification using photo IDs.
• While training enhances human detection, inherent individual variability and system errors necessitate robust security protocols.
• Current face recognition technology, including smartphone systems, may not reliably distinguish morphed identities.