Facial recognition by forensic facial experts, superrecognisers, and computer algorithms
Forensic facial experts are individuals who are trained to identify faces from images and video in order to assist in forensic investigations. However, little is known about the accuracy of facial recognition by forensic facial experts relative to other humans and computer algorithms.
To address this issue, a fascinating new study from the National Institute of Standards and Technology (NIST) has investigated facial recognition capabilities of humans and state-of-the-art computer algorithms. In this study, forensic facial experts, superrecognisers (untrained individuals with naturally high facial recognition abilities) and computer algorithms performed a verification task to determine whether pairs of faces were of the same person or different people.
Results showed that forensic facial experts and superrecognisers performed better than control groups (fingerprint experts and undergraduate students). Facial recognition accuracy was improved by averaging the ratings of multiple individuals. However, accuracy was highest when combining (or ‘fusing’) the ratings of human and computer algorithms. These findings suggest that combining human and machine decision-making can improve the accuracy of facial recognition.
Full article at: http://www.pnas.org/content/early/2018/05/22/1721355115.full