U.S. flag

An official website of the United States government, Department of Justice.

Face Recognition Accuracy of Forensic Examiners, Superrecognizers, and Face Recognition Algorithms

NCJ Number
251746
Date Published
May 2018
Length
10 pages
Annotation
Findings and methodology are reported for a study that measured face identification accuracy for an international group of professional forensic facial examiners who worked under circumstances of real-world casework.
Abstract
The study’s goal was to achieve the most accurate face identification, using people and/or machines working alone or in collaboration. The assigned task for participants was to determine whether pairs of face images showed the same person or different people. Image pairs were pre-screened to be challenging based on data from humans and computer algorithms. Images were taken with limited control of illumination, expression, and appearance. So as to provide a comprehensive assessment of human accuracy, the study tested three face specialist groups (forensic facial examiners, forensic facial reviewers, and “superrecognizers”). The latter group consisted of members of the general public who performed well on laboratory-based face recognition studies. These three groups’ performance in facial recognition was compared with that of a control group whose members did not qualify as “superrecognizers.” The study found that the two professionally trained groups and the untrained superrecognizers were more accurate in their identifications than the control group. Data that compared state-of-the-art face-recognition technology with that of the best human face identifiers showed that the best machine identifier performed in the range of the best professional facial examiners. The study found that optimal face identification was achieved only when trained humans and machines worked collaboratively. 5 figures and 25 references

Date Published: May 1, 2018