Can people tell a hyper-realistic mask from a real face?
Almost nobody noticed that a face was actually a lifelike silicone mask, whether they saw it in a photo or sitting on a bench in front of them.
Source
Hyper-realistic face masks: a new challenge in person identification
Study at a glance
- Design
- Human experiment — Experiments 1-2: one mask photo hidden among 19 real faces, followed by graded detection questions; Experiment 3: between-subjects field experiment with a live confederate (hyper-realistic mask, real face or obvious party mask) viewed at 5 m or 20 m, own- vs other-race
- N
- N=407 · 60 UK students (Experiment 1), 60 Japanese students (Experiment 2) and 407 passers-by in York and Kyoto (Experiment 3; 160 of them saw a hyper-realistic mask). The headline N is the field experiment.
- Population
- University students in York (UK) and Kyoto (Japan)
- Outcome
- Whether viewers mentioned a mask spontaneously or when prompted, forced-choice guesses of being in the mask condition, and picking the mask out of a photo array
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
In two lab experiments, 60 UK and 60 Japanese students rated 20 face photos for age, trustworthiness, dominance and attractiveness; one of the photos was secretly a person wearing a hyper-realistic mask. Afterwards they answered increasingly pointed questions: what they thought of the faces, whether anything was unusual, whether they had been in a mask group, and finally which photo in an array was the mask. In a third experiment, 407 passers-by on two campuses rated a live confederate who wore a hyper-realistic mask, an obvious party mask or no mask, seen from 5 m or 20 m, and whose face was of their own or another race.
What they found
Across all studies, only 2 of the 280 people who saw a hyper-realistic mask mentioned it spontaneously, and 3 more did so after prompting, all of them live viewers at the near distance. In Experiment 1 only 21.7% guessed they were in the mask group, and even when told a mask was present and shown all the photos, 30% missed it while most also flagged real faces as masks. In the field experiment, 42.5% of hyper-realistic-mask viewers guessed 'mask', between the 96.5% for obvious masks and 8.6% for real faces, so subtle cues had some effect, and closer viewing raised mask guesses only for the hyper-realistic mask. Other-race viewers guessed 'mask' more often even when the face was real, which points to a response bias rather than better detection.
The limits
What it doesn't show
Only a handful of commercial masks were tested, and all participants were young university students, so detection by trained staff such as border officers is untested. The closest viewing distance was 5 m, far from the 1-2 m of a passport check or a conversation, and viewers never saw the masked person talk or show strong expressions, which the authors suspect could give masks away. The own- versus other-race effects are hard to interpret because the fully crossed race comparison appears only in the field study, where race is confounded with how common each face type is on campus. Experiments 1 and 2 used a single mask among 19 real faces, so the low hit rates cannot be split cleanly into sensitivity and bias.
Key terms
- Hyper-realistic mask
- A hand-painted silicone mask covering the whole head and neck that moves with the wearer's face and shows their real eyes and mouth.
- Two-alternative forced choice (2AFC)
- A question that makes people choose between two options (here, mask group or no-mask group), which can reveal hunches they would not volunteer on their own.
- False alarm
- Reporting a signal that was not there, such as calling a real face a mask; the real-face condition measured this baseline.
- Other-race effect
- The usual finding that people are worse at telling apart faces from racial groups they have less experience with.
- Response bias
- A general tendency to give one answer more often regardless of the evidence, as opposed to genuinely better discrimination.
Flashcards
0 of 11 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
Across all three experiments, how many of the 280 people who saw a hyper-realistic mask mentioned it spontaneously?
Common questions
Why did the researchers include an obvious party mask and a no-mask condition?
They act as anchors: the party mask shows how often people report a mask when it is obvious, and the real face shows how often they report one when there is nothing to find. The hyper-realistic mask landed in between, much closer to a real face than to a party mask.
Were other-race viewers better at spotting masks?
Not really. They guessed 'mask' more often for hyper-realistic masks, but also for real faces, so the authors read it as a bias, possibly because an unfamiliar-looking face on campus seemed more likely to be an experimental trick.
Does this mean masks can never be detected?
No. Mask guesses rose at 5 m compared with 20 m and people who knew about such masks did better in the photo array, so some cues exist; the point is that ordinary viewers at ordinary distances almost always miss them.
More on Face processing