麻豆传媒

AT NEARLY 2 metres tall, Idris Bada is a big man with a big reputation. His colleagues in the Metropolitan Police Service call him Idris the Jailer, thanks to his remarkable track record in catching criminals.

On arriving for our meeting at a London police station, I am unnerved by an angry youth at the entrance, who I suspect has more contact with the police force than I usually do. Matters aren鈥檛 helped when I call my contact to let them know I鈥檓 here. 鈥淚鈥檒l send Idris to get you,鈥� says the voice on the other end of the line. 鈥淒on鈥檛 be scared鈥︹��

Bada certainly towers over me but he turns out to be a friendly giant with a deep, infectious laugh. He is one of the Met鈥檚 鈥渟uper-recognisers鈥�, officers who have been singled out for their astonishing ability to put names to faces. Their talents went unnoticed until 2010, when Detective Chief Inspector Mick Neville started keeping track of the number of people that officers had identified from CCTV images.

There are 35,000 officers in the Met, all with access to the images, but Neville found that the same few officers were topping the league time and again. Their identification rates were so impressive that Neville started to wonder if his top cops had special powers of recognition. So he called in a forensic psychologist to investigate.

What Neville didn鈥檛 know is that face recognition is a hot topic in neuroscience. For one thing, it appears to be a deep-seated skill. Babies who are just a few days old appear more interested in pictures of faces than any other images, and brain scans show that the neural networks we use to process faces are separate from those we use for recognising other complex objects, such as animals or cars. Because we know roughly which areas of the brain are involved, facial recognition is seen as an ideal model for understanding visual processing more generally.

And on a practical level, working out what is different about the way that people like Bada recognise faces could allow us to harness their powers more effectively, and perhaps even help train the rest of us to improve our skills. 鈥淲hatever it is that they鈥檙e doing, can we polish it?鈥� asks Neville.

The first clues to the special way that we recognise faces came from people whose ability in this area is impaired. Brain damage caused by a stroke or head injury occasionally leaves people unable to recognise close family and friends, sometimes even their own face in the mirror. Yet they have no problem identifying other kinds of objects.

These people tend to have damage to the fusiform gyrus or the occipital face area, both regions at the back of the brain. Studying uninjured people in brain scanners showed these two regions respond more strongly to pictures of faces than anything else.

It emerged that this condition, known as face blindness, or prosopagnosia, can also be present from birth in people with no sign of brain damage. As many as 1 in 40 people may be naturally face-blind, although some may be unaware of it because they have developed strategies for recognising people by their hair, clothing or gait.

Neuroscientists were starting to get to grips with what goes wrong in face blindness when along came people at the other end of the ability scale, claiming to be able to recognise individuals they had met years ago and then only fleetingly.

Secret powers

Four such people contacted Richard Russell, then at Harvard University, and Brad Duchaine, then at University College London, in the mid-2000s after reading press reports of their research into face blindness (麻豆传媒, 25 November 2006, p 34). All said they sometimes hid their exceptional ability for fear of making people uncomfortable. 鈥淚 do have to pretend that I don鈥檛 remember [people],鈥� said one, 鈥渨hen I recall that we saw each other once on campus four years ago.鈥�

Such anecdotes are hard to verify, but when Duchaine and Russell made the four perform standard face memory tests, they got perfect or near-perfect scores. A more difficult test had to be devised, on which the four duly scored far higher than a control group; in fact, the super-recognisers turned out to be about as good at recognising people as those with face blindness are bad 鈥� in the top 2 per cent of the ability range, compared with the bottom 2 per cent for prosopagnosics (, vol 16, p 252).

That suggests there is a broad spectrum of face-recognition abilities within the population, say the researchers. Since then, Duchaine and Russell have amassed more than 50 similarly talented individuals, who they are still studying.

Super-recognisers do not seem to be especially rare people. An interactive exhibit at London鈥檚 Science Museum that tested face-recognition abilities classified about 1 in 50 of people who tried it as super-recognisers.

An obvious question is whether this kind of talent is specific to faces, or whether it applies to other objects as well. Perhaps these people just have an exceptionally good memory for anything they see, not just faces.

Josh Davis, a forensic psychologist at the University of Greenwich, UK, has been studying the Met鈥檚 super-recognisers, and he discovered that their abilities were nowhere near as good when it came to other objects 鈥� for example, similar-looking flowers. 鈥淲e found that [their ability] did seem to be quite specific to faces,鈥� he says.

Alongside people who are face-blind, the super-recognisers provide researchers with a variety of ways to explore the special way the brain processes faces. For instance, most of us find an upside-down face harder to recognise than one that is the right way up. We have less trouble recognising inanimate objects that have been turned upside down. In people who are face-blind, this 鈥渋nversion effect鈥� is small or non-existent. In Duchaine and Russell鈥檚 tests of their initial group of four super-recognisers, the inversion effect for faces is larger.

Ashok Jansari, a cognitive psychologist at the University of East London is investigating the unique way that the brain processes a face as a whole, rather than the sum of its parts. 鈥淲hen looking at an object like a pair of glasses, you work out what it is by adding up the different parts: two bits of glass, a frame, and two bits to go over the ears,鈥� he says. 鈥淏ut in face recognition, you don鈥檛 see left eye, right eye, nose, mouth, you see the whole.鈥�

Jansari is trying to get to the bottom of this with a test involving not faces, but letters. Volunteers look at a large letter, say an H, whose outline is built up from a smaller letter (see 鈥淣ame that letter鈥�). Subjects have to name the building-block letter as quickly as possible. When the letter is different from the one it forms, most of us are slowed down by the clash between the two. 鈥淲e find it very difficult to suppress the large letter because we see the whole before we see the sum of the parts,鈥� says Jansari.

As with the inversion effect, face-blind people do not exhibit this slowdown. Jansari is investigating what happens with super-recognisers 鈥� perhaps they will show the opposite, a much larger slowdown. Or they might turn out to be good at processing both details and the whole.

As well as visual tests, Jansari uses electroencephalograms, tracking brain activity with electrodes placed on the scalp. Russell鈥檚 group is using an fMRI brain scanner to do something similar. His team has a paper due to published later this year, so at the moment all Russell will say is that the super-recognisers 鈥渟eem to be using their brains somewhat differently鈥�.

While these lab tests continue, the Met is already investigating if it can harness the powers of its own super-recognisers more effectively. 鈥淐an we make them better?鈥� asks Neville. 鈥淲hat we want is a practical use of the science.鈥�

Neville had already started tracking officers鈥� identification rates when, in August last year, riots and looting took place in London over several days. Thousands of CCTV images of suspects were put on an online database for viewing by the Met鈥檚 35,000 officers. So far there have been over 3000 arrests. An identification made from a CCTV image does not necessarily lead to a conviction of course, and police may still need to go through a range of other evidence-gathering procedures like searching the suspect鈥檚 home for stolen goods; but it is a good start.

Most officers were able to contribute one or two identifications, if any at all. But a few had a hit rate way above average. The top scorer was Constable Gary Collins, who has accounted for 185 arrests so far, and was named the Met鈥檚 鈥淐op of the Year鈥� for his efforts. In all, there were 20 officers who had a hit rate of 20 or more.

Neville called in Davis, who began testing this top 20, plus other officers who had already come to notice for their high identification rates, such as Bada. Although the testing is ongoing, Davis says fewer than half of the top performers would be classed as super-recognisers according to his results so far. One explanation, he points out, is that you don鈥檛 have to be a super-recogniser to identify a lot of people from CCTV footage. 鈥淚t might be that you work in a very small neighbourhood, and you get the same people caught on camera over and over again,鈥� he says. 鈥淵ou don鈥檛 have to have any super-recognition powers to be able recognise someone in that circumstance.鈥� And, he says, the best identifiers may just be very good cops. 鈥淪ome are exceptionally meticulous and spend hours, sometimes in their spare time, going through images.鈥�

Flash of recognition

Collins and Bada may well be good cops, but they both turned out to be genuine super-recognisers too. Down at the police station, they try to tell me how they do it. Collins shows me a series of fuzzy CCTV images of some of the rioters, most with their faces partly obscured, alongside police mugshots taken after their arrest (see images here). He is in no doubt that the pairs of faces match, although he struggles to convey what it is that he sees. 鈥淚t鈥檚 little things, the eyes or鈥� It鈥檚 an instant thing really, it鈥檚 hard to explain,鈥� he apologises.

鈥淐ollins is in no doubt that the pairs of faces match but he struggles to explain why: 鈥業t鈥檚 little things, the eyes or鈥� it鈥檚 hard to explain鈥︹�欌��

Bada seems to have a more systematic approach. 鈥淚 usually go from the side, the forehead to the brow, to get a definite ID,鈥� he says. 鈥淪ometimes a prisoner comes in and I鈥檓 like: 鈥業 know you as Captain America and now you鈥檙e telling me you鈥檙e Spiderman?鈥� All it needs is for him to turn sideways: it clicks, and I say, 鈥業 remember you鈥�.鈥�

Even with his special talent, Collins agrees that dedication pays off. 鈥淚鈥檓 forever printing off pictures at work, and I look at YouTube videos of gang members. People do think I鈥檓 a bit obsessive. I just try and cram as many faces as I can.鈥�

It may be that there are different types of super-recogniser 鈥� those that are born, and those that achieve their skills by practice and hard work, and combinations of the two. Whatever the case, Neville is keen to find more of them, perhaps by testing the abilities of recruits at police training college. He also wants there to be a formal qualification that super-recognisers can be awarded so that their evidence is taken more seriously in court: 鈥淭hey almost need a badge of honour that says, if he tells you it鈥檚 Father Christmas, it is.鈥�

There is no shortage of roles for which it would be useful to assess someone鈥檚 face-recognition skills: not only police officers, but also border control staff 鈥� indeed, any job that involves keeping track of people 鈥� as well as witnesses in court.

So far, face-recognition tests have not been brought to bear on these roles, although Duchaine says that the tests he uses in the lab could easily be deployed elsewhere. Perhaps the discovery of super-recognisers will provide the incentive that organisations need to try the tests out.

鈥淪upers do make the point, strongly, that there is a real continuum of face-recognition ability, so I do hope that it鈥檒l push some of these organisations to test,鈥� says Duchaine. He is trying to get the US Department of Homeland Security interested.

Could we ever learn to enhance our facial recognition powers? Joe DeGutis, a psychologist at Harvard University, has found that people who are born with face blindness can improve their abilities somewhat if they practise putting computer-generated faces into groups, based on shared features. He has also found that people with average face-recognition abilities get a little more skilful by doing the same exercise 鈥� by up to 10 per cent.

DeGutis thinks he might be able to achieve more impressive gains by refining the training, although it isn鈥檛 yet clear exactly how the technique works. Even so, this routine would only be for the committed. 鈥淵ou have to do [it] every day for several weeks,鈥� he says. The benefits of the training last just a few months.

Would that justify the investment in time and effort? Unsurprisingly, the officers I spoke to felt that super-recognising is a satisfying skill to have at their disposal. Bada recalls a particularly good day at the office back in 2005. 鈥淚 opened a cell door and I saw my bully from primary school, nearly 27 years after I last clapped eyes on him,鈥� he recalls. 鈥淚 just looked at him and said: 鈥業 remember you鈥�.鈥�