Describing Faces

Abstract

Over the last several decades, facial recognition technology has become increasingly accurate. However, despite advances in technology, these systems are not yet as good as what was envisioned in crime scene investigation (CSI) films. For example, facial recognition technology failed in the Boston Marathon bombing manhunt in 2013 (Gallagher S (2013) Why facial recognition tech failed in the Boston bombing manhunt. http://arstechnica.com/information-technology/2013/05/why-facial-recognition-tech-failed-in-the-boston-bombing-manhunt/). The two bombers, Dzhokhar and Tamerlan Tsarnaev, were both in the facial image database. There were photos of the suspects, but the system could not find a match, or at least could not come up with one before the suspects were identified by humans. Under the best circumstances, facial recognition can be extremely accurate, returning the right person as a potential match within ideal conditions, e.g. front-view faces where all photos are shot from the same angle and with the same lighting. To reach that level of accuracy in real-world footage, which is often blurry, with different poses and lighting, computers almost always require a degree of skilled human guidance. According to the NIST report on the evaluation of 2D still-image facial recognition algorithms, the facial recognition accuracy rate decreases linearly with the logarithm of the population size of the image database. In all cases, human adjudication is ultimately necessary for verification (Grother P, Quinn GW, and Philips PJ (2011) Report on the evaluation of 2D still-image face recognition algorithms. NIST Interagency Report 7709, August 24, 2011).