A study compares the accuracy of human forensic examiners and computer algorithms in face recognition tasks. Minimizing errors in facial identification has implications for forensic science. However, tests of accuracy of professional face identifiers are rare, and methods to maximize the accuracy of face identification are lacking. P. Jonathon Phillips and colleagues compared face identification in humans and computers. The authors tested human face recognition ability, and demonstrated that forensic professionals, including forensic facial examiners with extensive training, were more accurate than control groups at identifying faces in a face identification test. Subsequently, the authors administered the face identification test to four computer algorithms trained for facial recognition. The algorithms, which were developed between 2015 and 2017, performed within the range of human accuracy, with the most recent algorithm scoring higher than the median of the forensic facial examiners. However, the authors found that combining a single forensic facial examiner with the top-performing algorithm produced the most accurate face identification, resulting in improved accuracy, compared with the combination of two forensic face examiners. The results may help improve accuracy in face identification, according to the authors.
Article #17-21355: "Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms," by P. Jonathon Phillips et al.
MEDIA CONTACT: P. Jonathon Phillips, National Institute of Standards and Technology, Gaithersburg, MD; tel: 301-975-5348, 240-364-4586; e-mail: <firstname.lastname@example.org>