Summary: Performance comparison of human and automatic face
School of Information Technology and Engineering, University of Ottawa, Ontario, Canada
3M Security Systems Division, 1545 Carling Avenue, Suite 700, Ottawa, Ontario, Canada
Of all biometric modalities, the capabilities of automatic face recognition software are most often compared to the
human ability to do the same task. Unlike other biometric modalities, such as fingerprints, where a small group of
trained people are experts, most people use their face recognition abilities every day. There is a significant body of
literature analysing the ability of humans to recognize faces, including a number of recent works which analyse the
ability of people to perform security related face recognition tasks. For example, Kemp et al.  analysed the ability
of supermarket cashiers to identify shoppers from photos on credit cards, and discovered overall poor performance.
Chang Hong et al.  analyzed the ability of people to match poor-quality video footage against high-quality
photographs, and showed a high level of performance. Such discrepancy in results illustrates the importance of the
task context for human face recognition. People focussing on face recognition as their primary task will outperform
those, such as supermarket cashiers, whose primary task is customer service.
There are few reports in the scientific literature of comparisons of human face recognition performance to that