Commercial Face Recognition Doesn't Work

Face recognition plays an increasingly important role in our security and surveillance infrastructure, despite substantial evidence that neither human observers nor automatic systems can reliably match unfamiliar faces in real-world settings. In the present article, we contrast real-world performance with performance in benchmark tests. We then consider human limits in unfamiliar face matching, and offer two discussion points: First, that face photographs may not contain enough information to establish a reliable match in practical settings. Second, that we should abandon photographic snapshots as a means of identifying unfamiliar people. We end by outlining some broad alternatives to relying on face photographs.