'He just looked exactly like the man who raped me'

It is well-known in psychology that there's something special about faces – they embody our sense of self, we have dedicated regions in the brain for processing them, and humans are excellent at recognising faces. Or are we? 

Dame Professor Vicki Bruce (University of Newcastle) explained that her interest in faces began with cases of mistaken identity by eyewitnesses which had led to wrongful convictions. She gave the example of Ronald Cotton, convicted of a horrific rape after the victim, Jennifer Thompson, pointed to him in a photo array and subsequent live line up. Cotton spent 10 years in prison before it emerged that another man had committed the rape. Thompson, who now campaigns together with Cotton to spread the word on wrongful convictions based on eyewitness testimony, had said:

'When I first saw the photo of him and I saw the pictures of the men that were in front of me. Ronald Cotton – he just looked exactly like the man who raped me. And not a lot of time had elapsed between the crime and me looking at the pictures, so my memory was still very fresh. And then when I saw him in the physical line up and I was actually able to see him as a person and his demeanor and his postures – it just further convinced me that Ronald Cotton was the man. He looked exactly like the man. He looked like the sketch that I had given to the police.'

Our italics here highlight Bruce's key point: that resemblance is not the same as identity. We excel at recognising resemblance, but we need to understand that there are shifts in representation as faces become more familiar. Seeing a resemblance has seen many people get convicted wrongfully: and, in the case of Jean Charles de Menezes, killed.

In some of her early work Bruce found that people were surprisingly bad at matching a face to the same face shown in an array at a slightly different angle – something she thought people would score 100 per cent in. It appears we are poor at comparing faces, even if we are just shown a pair of faces and asked if they're the same. When experts in the area, in this case passport control officers, are asked to do a similar task they also perform poorly. It seems two pictures of the same person taken a few years apart can look very different while two pictures of different people can look surprisingly similar.

Bruce suggested we may be over-confident in our own ability to recognise faces. It is quite common knowledge from neuroscience that we have processes in the brain for this purpose alone, and face recognition is a staple of popular fiction and in the media. The true picture may be quite different.

When witnesses are asked to identify suspects in a line-up or photo array, Bruce said, they should simply be asked who most resembles the offender rather than pointing the finger to who did it. She moved on to the use of composite sketch programmes in investigations where witnesses are asked to focus on individual facial features. In experiments where participants are asked to make a composite of an unfamiliar face, such as a snooker player, then snooker fans attempt to identify the player, accuracy is 'woeful' according to Bruce. Why is it so hard to create a recognisable composite?

Bruce explained that using memory to recall specific facial features is unnatural and difficult. Witnesses would also be unfamiliar with the faces and in this case they tend to recognise external features such as ears and hair rather than internal facial features. In contrast, we remember internal features of familiar faces. In an experiment subjects were asked to match a photo to one of a group of composite sketches – when internal features were blurred they showed good recognition, but when only internal features were visible performance was poor.

So how could these systems be improved to create more recognisable composites in police investigations? Bruce pointed out four potential ways – the first, in cases where there are multiple witnesses, is to get each to create a composite and morph these. In these morphed faces the features which are agreed-on across witnesses will be more visible and errors will be minimised.

Second, in interviewing witnesses it is useful to use a holistic approach by asking people about traits of the face including attractiveness and intelligence. Evidence has shown this interviewing method results in better composites compared to a usual interview focusing on individual facial features.

Third, it is useful to exploit peoples' natural ability for seeing resemblance across faces. The Evofit system presents groups of faces which are built out of certain features. People pick those which most resemble the target and the array of faces evolves to come closer to the person's actual face. In this case people aren't forced to recall individual features but faces as a whole.

Finally it can be useful to focus first on internal features of a face by using the Evofit system but blurring out external features, and adding on external features later, which results in more accurate composites. Bruce concluded her fascinating talk by suggesting that we use what we know about face recognition to build even more accurate and helpful tools for investigations. 

- Read more about Dame Professor Vicki Bruce's work. More reports from the Society's Annual Conference will appear online over the next fortnight, and then a selection in the July edition. Next year's event is to be held in Brighton from 3-5 May.