Verification of face identities from images captured on video.

Humans are notoriously poor at identifying unfamiliar faces.

“This lack of empirical validation for facial image comparison training courses is particularly concerning because humans are notoriously poor at identifying unfamiliar faces.”

Error-rates in tests of unfamiliar face identification typically average 20-30% for both novices and trained practitioners.

“Error-rates in tests typically average 20–30% for both novices and trained practitioners [5–9].”

Bruce and colleagues found mean error rates of 30% on a face matching task with high-resolution images of young adult male subjects taken minutes apart.

“Despite target and gallery images being taken only minutes apart, in full frontal pose and under similar lighting conditions, mean error rates on this task were 30%.”

People make large numbers of errors when matching photos of unfamiliar faces.

“This is an important limitation because psychological research has shown people make large numbers of errors when matching photos of unfamiliar faces.”

Previous research has shown that face aging makes unfamiliar face matching more difficult, increasing error by 20% when images are separated by more than a year.

“Previous research has shown that face aging makes unfamiliar face matching more difficult, increasing error by 20% when images are separated by more than a year [17, 22, 23] and so we expected error rates to be higher than in similar test using images taken on the same day [12].”

Face aging caused poorer performance compared to similar tasks where images were taken on the same day.

“To account for this, and consistent with recent work [17, 23] we suggest that face aging caused poorer performance compared to similar tasks where images were taken on the same day [12].”

Comparable error rates when targets did not appear in candidate lists suggests that foil candidates selected by the computerised FR engine from a database of millions of passport images were more easily mistaken for the target identity than foils selected by humans from much smaller databases in previous studies.

“Comparable error rates when targets did not appear in candidate lists suggests that ‘foil candidates’–selected by the computerised FR engine from a database of millions of passport images—were more easily mistaken for the target identity than foils selected by humans from much smaller databases in previous studies [12].”

Error rates were substantially higher than in previous studies of face matching performance using one-in-ten face matching arrays.

“Despite these constraints, error rates were substantially higher than in previous studies of face matching performance using one-in-ten face matching arrays [12].”