InnovationLegalRegulationResearchScience

Algorithm Uses Facial Features To Detect Criminals, Claim Chinese Researchers

Tom Jowitt is a leading British tech freelance and long standing contributor to TechWeek Europe

Do you look like a criminal? Chinese researchers believe their facial algorithm can detect your criminal tendencies

Chinese researchers have controversially claimed they have a working algorithm that can identify criminals from their facial features.

The claim (if validated) will add to tensions surrounding the use of facial recognition technology, with privacy campaigners already concerned at its increasing use by intelligence agencies.

Criminal Looks?

The researchers from Shanghai Jiao Tong University claimed in a paper that their computer was able to accurately detect criminals just by using their facial features, after they conducted an experiment in mainland China, using the photos of 1,856 Chinese men.

“In order to conduct our experiments and draw conclusions with strict control of variables, we collected 1856 ID photos that satisfy the following criteria: Chinese, male, between ages of 18 and 55, no facial hair, no facial scars or other markings,” said the researchers.

facebook-facial-recognitionThe researchers used the ID photos of 1,126 non-criminals that were “acquired from Internet using the web spider tool; they are from a wide gamut of professions and social status, including waiters, construction workers, taxi and truck drivers, real estate agents, doctors, lawyers and professors,” the researchers said.

Roughly half of these non-criminals had a degree.

But the researchers also added 730 ID pictures (not police mugshots) of convicted criminals or “wanted suspects by the ministry of public security”.

The researchers then used 90 percent of these photos to train their algorithm. They then tested their algorithm on the remaining 10 percent of photos to see if the computer was able to correctly identify the criminals.

The researchers claimed their algorithm was correctly identified criminals nine times out of ten (actually 89 percent of the time).

The researchers said their algorithm had identified a number of ‘facial landmark points’, such as the curvature of the upper lip and distance between eyes, that is apparently common among those of a criminal bent.

Flawed Science

But the Chinese research has been attacked by a leading crime expert in the UK, who told the BBC that the research had not been peer reviewed.

“This article is not looking at people’s behaviour, it is looking at criminal conviction,” Prof Susan McVie, professor of quantitative criminology at the University of Edinburgh told the BBC.

“There is a huge margin of error around this sort of work and if you were trying to use the algorithm to predict who might commit a crime, you wouldn’t find a high success rate,” she said.

Prof McVie pointed out that in history a 19th Century criminologist (Cesare Lombroso) used phrenology – feeling people’s heads – with a theory that there were lumps and bumps associated with certain personality traits.

Prof McVie said this was now considered to be very old and flawed science and that criminologists have not believed in it for decades, and she also warned that a system which uses looks (rather than behaviour) could have dangerous consequences for society.

Earlier this year Facebook released its ‘Moments’ photo sharing app in Europe – nearly a year after its US launch.

However Moments in Europe came without a facial recognition feature present in the American version following complaints about how Facebook gathers and uses biometric data

Do you know all about biometric technology? Take our quiz!