Facial recognition fails at forensic reliability


A 2009 paper by the US National Academy of Sciences found that among forensic methods only DNA can reliably and consistency match evidence to specific individuals or sources. While it's commonly understood that techniques such as analysis of blood spatter patterns are up for debate, other types of visual evidence have been more readily accepted. In 2015 the FBI announced that virtually all of its hair analysis testing was scientifically indefensible, and in 2016 the Texas Forensic Science Commission recommended banning bite mark evidence from courts. More recent research finds that comparing facial images is equally unreliable, even though most people think they're good at recognising faces. There is no standard for the number of points of similarity needed to constitute a match, and photographic conditions can radically alter how a face appears. In Denver, local resident Steven Talley was accused of bank robberies he did not commit based on CCTV images; the case was dismissed after more than a year, during which time Talley was unable to find work and his professional licences expired. Further studies have found that even passport officers perform poorly in tests. In a 2014 paper, the National Institute of Standards and Technology suggested that automated facial recognition systems are not the solution, particularly since the escalating size of the datasets increases the potential for false matches - and therefore place myriad citizens in the position of having to prove the technology is wrong.


writer: Ava Kofman
Publication: The Intercept
Publication date: 2016-10-13