Yes, you really CAN spot a wrong ’un by their face
WE all reckon we can spot a wrong ’un from time to time, though it might not always be prudent to say it aloud.
But now scientists say a controversial computer program can pick out potential criminals – just by looking at their faces.
The Chinese software was said to have identified law-breakers with up to 89.5 per cent accuracy.
And in our picture quiz on this page, you can try to beat the boffins by sniffing out the ne’er do wells among the saints.
The program found criminals have upper lips which curve around 23 per cent more than law-abiding citizens – and their eyes are closer together.
This contentious type of science is linked to discredited 19th century Italian criminologist Cesare Lombroso, who argued that criminals were more primitive than the rest of us, with differently shaped skulls and faces.
There is no real evidence of why criminals might look any different, but academics at Shanghai Jiao Tong University say their program can be trusted.
Their study states: ‘Unlike a human examiner/judge, a computer vision algorithm or classifier has absolutely no subjective baggage, having no emotions, no biases whatsoever due to past experience, race, religion, politi-
‘Computer algorithm has no baggage’
cal doctrine, gender, age, etc, no mental fatigue, no preconditioning of a bad sleep or meal.’ Dr Leandro Minku, a computer science lecturer at the University of Leicester, said the difference in facial features, including a narrower angle from criminals’ nose to the corners of their mouth, might be explained by non-criminals being more likely to smile slightly in photographs.
But he said the research was well-conducted, adding: ‘This is a very controversial topic, but one of the advantages of using machines is that they “learn” based on the data given. They can analyse a very large amount of information which we would struggle to do.’
The Chinese researchers collected 1,856 ID photographs, rather than police mugshots, of men aged 18 to 55 with no facial hair and no facial scars or other markings for the study.
Half were convicted criminals and all were fed into a computer system ‘taught’ to tell the difference. The best classifier achieved 89.51 per cent accuracy, according to the report on the research website arXiv.
The computer system found there is no one typical criminal face, but many, with law-abiding citizens actually having more facial similarity to each other than the crooks.