© The Open University/iStockphoto.com/TAGSTOCK1

Why is it so hard to describe a face?

Did anyone successfully recognise the person you were describing from the comment you posted? Did you recognise anyone described in someone else's comment?

If only the face is described, it is usually the case that no one is able to recognise who is being described unless that person has an unusual, or famously characteristic feature. For example, the British celebrity Bruce Forsyth is often described as having a large chin, making it possible for him to be recognised from a description that includes 'very large chin'.

For most faces though, it is almost impossible to recognise them from a verbal (put into words) description. Can you suggest why this might be?

Generally speaking, describing a face can be difficult because:

Faces tend to look very similar. Saying the face has a nose, two eyes and a mouth does not distinguish it from millions of other faces.It is difficult to bring to mind what an individual feature looks like. Many people find it tricky to imagine a face, and 'zooming' in on an individual feature is particularly difficult, if not impossible.Our vocabulary for describing faces is not very good, probably because describing a face is something we do only rarely. Since most people's features will, by definition, be 'average', describing someone as having an average-looking nose or mouth will not distinguish them from all the other people who have average-looking noses or mouths.

If you were feeling creative or poetic, you might have included a metaphor or simile in your description. For example, Shakespeare's Lady Macbeth described her husband as:

Your face, my thane, is as a book where men
May read strange matters. To beguile the time
Look like the time; bear welcome in your eye,
Your hand, your tongue: look like the innocent flower,
But be the serpent under't.

(from Macbeth, Act 1, Scene 5)

But although such descriptions can be evocative, they tend not to increase the chances of the description being recognised.

Creating a face

Our memory of a face is a 'visual memory' that we must translate into a verbal description when asked to describe a face. Then someone else has to compare our verbal description to their visual memories. Switching between these different 'modalities' is difficult to do.

Given the problems with verbal descriptions, and to assist in the process of translating the witness's visual memory of the perpetrator's face into a visual image that can be used to find a suspect, the police have employed artists to create a pictorial representation.

In the early 1970s, a package known as 'PhotoFIT', which someone without artistic skills could use, became available. This system, shown in the image above, comprised numerous black-and-white prints of facial features (hairlines/ears, eyes/eyebrows, nose, mouth and chin/cheeks). Based on their verbal description, the witnesses would be shown a choice of such features and asked to select the ones that best represented those of the perpetrator. The selected photographed features were then physically blended together, as in a collage.

Research showed that the images produced tended not to be a good likeness of the target face. Psychologists suggested this was because the construction process involved selecting individual features, and the human mind does not remember faces as a collection of features, but instead represents faces 'holistically', i.e. the whole face is stored in memory.

Psychology and technology

The E-FIT system that replaced PhotoFIT represented a step in the right direction by presenting a witness with a whole face.

With improvements in technology, computerised versions of composite systems took over from PhotoFIT. These systems, which included the Electronic Facial Identification Technique (E-FIT), contained a much larger database of facial features. Perhaps more importantly, E-FIT operated by only ever presenting witnesses with a whole face. Although this was a psychological step in the right direction as it was more of a match for the 'holistic' way the mind remembers faces, one key drawback of the system was that the witness was still required to try and improve the face by changing individual features.

Laboratory research on face perception has shown that it is much harder to recognise a facial feature when it is seen on its own than when it is part of the whole face (Tanaka and Farah, 1993), so the E-FIT system was a better match for how the human mind works than PhotoFIT. Moreover, E-FIT also allowed witnesses to easily change the configuration of the features, that is the relative positions of the features, and psychological research has shown that the configuration of the face is a very important component in human face recognition.

Recognising a face

You've just looked at describing faces and will now turn your focus to the psychology of recognising faces, and find out how this knowledge is being incorporated into the latest composite systems being used by the police.

This activity is a demonstration of how our minds work by recognising the whole face, rather than by remembering the individual features.

The face, the whole face and nothing but the face

Forming an image from the halves of two others is known as a 'chimeric' face.

How many of the celebrities did you manage to recognise? Did you find it easier to recognise the halves when they were not aligned?

Obviously you would have found it difficult to recognise any of the halves if they were of people you do not know, but other than that you probably found it much easier to recognise the halves when they were not aligned. This is because when they were aligned, your mind saw the halves as forming a single, whole face and it tried to recognise who that face was. Remember, psychological research has found that the brain recognises faces 'holistically', which means we find it difficult to see a face as being composed of two different halves, and instead see it as a single, whole face - even if the two halves are of very different looking people! When the halves are not aligned, the brain does not see the image as a single whole face, allowing you to concentrate on the two halves separately.

If the facial composite systems used by the police are to create a good likeness of the suspect, then it is vital that they are designed to take account of how the human mind remembers faces.

E-FIT was one of the first computerised composite systems that presented a whole face to the witness rather than collections of separate features. The study by Brace, Pike, Allen and Kemp (2006) evaluated E-FIT using these images of famous faces. This study concluded that although E-FIT did represent an improvement in psychological terms over older systems such as Photofit, there were still aspects of the construction process that were psychologically problematic, including:

that the witness had to recall what the face looked like; our minds seem much better at recognising faces than they do recalling them (picturing them in the mind)the witness had to alter the composite image by selecting individual featuresthe process was very reliant on the witness verbally communicating their memories to a police operator.

These problems were to prove very important in designing the next generation of composite systems. Nonetheless, Brace, Pike, Allen and Kemp (2006) did find that 66% of the 96 E-FITs produced in the study were recognised by at least one person.

Recall and recognition

The police only need one person to recognise a composite to provide them with a lead. How many did you recognise? It's unlikely you recognised them all.

Although E-FIT was based on psychological knowledge of how we remember faces, it still relied on a witness picturing the face in their mind and then describing it to an officer.

In a previous activity you had to recall a face and describe it, and you probably found this a very difficult task. In the activity involving two halves of celebrity faces, you attempted to recognise faces that were either shown in halves or as a composite image. Despite this, you probably found recognising the faces far easier than you did recalling a face previously. The difference between recall and recognition is an issue that has been explored extensively in psychological research. In general, our brains seem to be far better at recognition. Just think how many different objects and faces you can recognise, how quickly you do this and how rare it is that you make a mistake. By contrast, recall is something we find much more difficult and tend to make far more errors when doing.

As you will see in the next video, the difference between recall and recognition was a key element in the design of the next generation of composite systems.

Most of us are familiar with the facial composite images the police use, when searching for the perpetrator of a crime. The police build these faces from witness descriptions. In this film, we're going to be looking at the Photofit system and a demonstration of the E-FIT system, which relies on constructing a face feature by feature. We'll also look at the newer EFIT-V system, which explores the psychological advantage of recalling a whole face rather than individual features. Hello Ollie.

OLLIE: Hello.

GRAHAM PIKE: Ok. What we're gonna do now is to create a facial likeness of the image you've just seen and we're gonna do that by using the E-FIT system on this computer here. So I just need you to start off by telling me the ethnicity of the face you saw and the gender?

OLLIE: White Female.

GRAHAM PIKE: So, white female. What type of face shape did the person have?

OLLIE: Small and round, kind of, yeah round.

GRAHAM PIKE: And quite short.

OLLIE: Mmm.

GRAHAM PIKE: And how old would you say the person was?

OLLIE: Sixty.

GRAHAM PIKE: So...

OLLIE: Old.

GRAHAM PIKE: Old. Alright, ok. Now is there anything you can tell me about the hair?

OLLIE: It was grey.

GRAHAM PIKE: Ok, it was grey.

OLLIE: Like man's short hair.

GRAHAM PIKE: Ok. So, grey and short.

OLLIE: It was messy.

GRAHAM PIKE: It was untidy. So turning to the eyes now, Ollie. First of all, is there anything you can remember about the eyes that you could describe?

OLLIE: They were narrow, kind of slit, and they were old person eyes, and they were dark brown.

GRAHAM PIKE: Now what the system's gonna do is take the information we just provided with those descriptions and it's gonna display a face. What we'll then do is work with the system to alter it, to make it more like the face.

Well faces are actually an odd stimuli. The difference between one face and another is very small. We all have two eyes, a nose, a mouth -so it's very subtle differences between them. So in order to describe you have to be able to bring the face, kind of, into your mind's eye and zoom in on those individual features, and that's very, very difficult to do, and probably the reason that's very hard to do is we don't store faces in our memory as a collection of features. Instead, we tend to store the whole face.

OLLIE: There.

GRAHAM PIKE: There? You sure?

OLLIE: Yeah.

GRAHAM PIKE: Is there anything about the rest of the face? And the eyes? You did say they were narrow and slitty.

OLLIE: Not as much as that.

GRAHAM PIKE: Not as much as that. Well we can make them bigger. Let's have a look for a few other examples. The situation facing an eyewitness is a particularly difficult one when it comes to describing a face, because the eyewitness, by the very nature, will be describing a face of someone they don't know. If they knew who the person was they would simply say to the police, it was Bob, it was Katie, it was whoever. It was the guy who lives down the road; it's the guy that's in the pub. The only circumstances under which you would be needing a description is when you don't know the person.

So therefore, you're describing a face you're not familiar with and we are very bad both at recognising and particularly at recalling and describing unfamiliar faces. Ok, was there anything else about them? Is that quite a good representation of the face?

OLLIE: I think so.

GRAHAM PIKE: Ok Ollie, that looks like quite a good representation of Joe Brand to me. OH! And it isn't...oh dear. Ok so it's Dame Judy Dench. So when you were thinking about trying to recreate the face were you able to bring that picture to mind?

OLLIE: Yeah. I thought the eyes and mouth, and the hair, that's all I remember the most.

GRAHAM PIKE: So how difficult did you find the process of trying to recreate the face?

OLLIE: I found it quite difficult. Some of them weren't exact but they reminded me of them.

GRAHAM PIKE: How easy did you find it to describe the features?

OLLIE: Hard. With the selections.

GRAHAM PIKE: Looking at it now and to describe it - do you think it would be an easier or difficult...

OLLIE: Yeah, it would have been easier.

GRAHAM PIKE: That would have been easier... So looking at it now, and say for instance describe the nose to me.

OLLIE: Umm...yeah that would be hard.

GRAHAM PIKE: It's hard isn't it? We are very bad at describing faces and there are a couple of reasons for that. Perhaps the simplest reason is we don't have the vocabulary to describe individual facial features very well. So if I ask someone to describe a nose, all they're likely to tell me is, well, it kind looked like a nose. They might even say, you know, it's an average nose. But when it comes to hair, we often have a better vocabulary. We can talk about colour, we can talk about length, we can talk about style, but generally when it's individual facial features we just don't have the words available to us to describe the features very well.

The other problem is, to describe something you first have to bring it mind - you have to recall - and we find it very difficult to bring faces to the mind's eye in sufficient detail, in order to describe them. You then have the added problem that you always do with describing face is you then have to communicate somehow, that information, which is visual in nature, which is about how something looks, and to describe it verbally to the police officer. And going from visual memory to the verbal description is quite tricky.

OLLIE: Yeah.

GRAHAM PIKE: What we're going to do now is, once again try to construct a facial likeness of the face that you've just been studying, but we're gonna use a new system rather than the E-FIT. So the first thing is, was it a man or a woman?

OLLIE: A female.

GRAHAM PIKE: And can you tell the ethnicity?

OLLIE: White.

GRAHAM PIKE: And their age?

OLLIE: Fifty-five to sixty.

GRAHAM PIKE: And just an approximate indication there of face shape...? So the problem for the police then, is they need to come up with a visual image of what the perpetrator looked like. So the first attempts to do this, use sketch artists; so you get an artist to talk to the witness, the witness would describe the face to the sketch artist, who would then produce an initial drawing and adapt it in light of what the witness was saying, to produce a sketch of the perpetrator. The problem with this is first of all it relies on the witness describing the face, which we know is difficult.

It also relies on the police having sketch artists that can draw very good images of faces, and that's quite a skill. The initial technological solutions to this problem were the identikit and Photofit systems, and what they comprised were albums of individual features. The person would look through the albums - there's one here, so you can see there's lots of eyes - they would look through the eyes and they would pick one of the eyes out. They would do the same for noses and the police officer then could fetch from a box of all these features, the noses and the eyes, and then put them all together to form an image like that.

Now you can see these are actual photographs of other people's features that have been put together into a composite to try and represent the face of the perpetrator. There are lots of problems with this, it's got lots of lines on it for one thing, the features don't necessarily, kind of, match up particularly well and researchers found that the accuracy of these likenesses that's produced by these systems isn't that brilliant. However, of course, it's arguably still better than simply putting out a verbal description, because verbal descriptions just really don't contain sufficient information to allow someone to identify the perpetrator. What we need to do is to try to work out which is the best face.

We can also tell the computer which are particularly bad matches, and if two or more of the faces look like possibilities we can morph or blend them together. As we go along, the arrays of faces will become more and more similar to one another. There are a new generation of facial compositing systems that are beginning to be used by the police, now. All of these systems work off a similar technique and the idea is that you try to minimise the amount of verbal description that's necessary, that the system has to have in order to be able to create the face, and you don't ever require the witness to work feature by feature.

Instead, the construction process works more holistically - so you're presented with a whole face and you simply look at an array of nine or sixteen depending on the system, and say which of the faces in front of you is a best match, for your memory of the perpetrator. So psychologically this is another big step forward. It's, the process of construction is a much better match for how we think human face recognition and human face memory operates.

OLLIE: This one is not a good one.

GRAHAM PIKE: There is a problem in simply trying to come up with technological solutions because the key limiting factor here, the real problem in the equation is human memory. The technology can still be improved further - I don't think we should be complacent - there's always room to develop it further. OK Ollie, you created the face of Judy Dench again. How did you find the process?

OLLIE: Easier than before. It gave me choices that it had made itself but I think it's good that on the way you could change bits - which I think I should have done - by making her look older and stuff.

GRAHAM PIKE: Before you were saying found it quite difficult. If you remember I asked you to describe the nose in the picture and you found it quite tricky to do?

GRAHAM PIKE: Did you notice there was less description?

OLLIE: Yeah, that made it easier.

GRAHAM PIKE: So how do you think you did? Do you think that's better than the...

OLLIE: It's better than before.

GRAHAM PIKE: Technology cannot completely overcome the limitations of the human mind. Even recreating the face of someone we are familiar with is incredibly difficult. And while the differences between the two systems may seem subtle, psychologically we're aware that humans do not remember a face by its individual features. The newer systems, which primarily use whole faces as opposed to features have proved a much better match for how we think human face recognition and human memory works.

Comparing E-FIT and EFIT-V

Is it possible to design a composite system based on recognition rather than recall?

In the 'Comparing E-FIT and EFIT-V' video Graham worked with a participant to produce a face using E-FIT and then using EFIT-V, which is the latest composite system to be used by the police. As you will see, EFIT-V is designed to try and utilise face recognition, rather than face recall. The design of EFIT-V once again demonstrates how important it is to have psychological knowledge of how the mind works when it comes to police investigations.

Summary

Psychological research has helped create composite systems that make the most of human cognition and get the most from the memories of witnesses.

Composites are used by the police to generate leads when they do not have a suspect and are not intended to be used as identification evidence - that is to say that the similarity of a suspect to a composite image should not be used as evidence that they committed the crime.

Once the police have a suspect, they need to use an alternative method to produce evidence that the person is the perpetrator of the crime, such as asking the witness to attempt to identify the suspect in an identification procedure, such as a line-up. It is these alternative methods that we'll look at next week.

But before you do that, you'll explore face recognition further in the next activity, including hearing from someone with 'prosopagnosia', which is a neurological condition that leaves the person unable to recognise faces.

When face recognition fails

Eyewitness misidentification seems to be the most common cause of miscarriages of justice. As eyewitness identification relies on the ability of the witness to recognise the perpetrator, it is crucial that we understand as much as possible about face recognition.

An eyewitness selecting an innocent person at an identity parade is obviously an example of face recognition somehow going wrong. Of course, there are reasons other than poor face recognition for why witnesses may select the wrong person, including that the officer could suggest they pick a certain face. Alternatively, it is possible that the witness might select the wrong face on purpose, either to protect the suspect or because they are frightened of identifying them. In this course we willI limit our attention to witnesses who, like Jennifer Thompson (the witness you saw in Week 1), think they are identifying the correct person, even though they are selecting an innocent person.

A great deal of psychological research has been conducted on how we recognise faces, and a lot of this work has focused on instances where our face-recognition skills fail us. Some of this research has involved analysing behavioural variables, including how accurately and quickly we recognise images such as the chimeric faces you saw previously. Alternatively, other research has taken a neuropsychological approach and looks at what happens to face recognition when the brain is damaged. We will be looking at both types of research in the rest of this week.

Turning face perception on its head

It is difficult to recognise a face when it is upside down because we have developed face-recognition skills that are based on the 'configuration' of the facial features, that is the relative spatial position of the features in the face.

When a face is upright, these cues provide us with a quick and accurate means of recognising a face. However, when the face is turned upside down, the configuration of the features is completely disrupted, meaning the brain cannot use its normal method of face recognition. Instead, the brain must rely on other methods, such as analysing the individual features, which are slower and tend to be less accurate.

The 'Turning face perception on its head' video provides a very neat demonstration of this phenomenon.

Face-recognition disorders

Faces are an extremely important part of our world. They allow us to distinguish our friends and family from people we don't know, to recognise many hundreds of people even if they change their appearance, and also to determine someone's emotional state.

You have seen that this expertise does not extend to an upside-down face and neither does it mean we can recognise someone with whom we are unfamiliar and have only seen briefly once before. This is a problem for the criminal justice system, as it means that even our extraordinary ability to recognise faces may not be up to the task faced by an eyewitness.

Our highly developed face-recognition skills also mean that faces tend to be recognised in a different way to other objects, using different cognitive processes and different parts of the brain. Psychologists think that the area of the brain known as the 'fusiform gyrus', is activated when recognising a face but not when recognising other objects. That means that if this part of the brain becomes damaged, the person could be left being able to recognise objects but not being able to recognise faces. This condition is known as 'prosopagnosia' or 'face blindness'. It is thought that as well as being caused by damage to the brain, this condition can be congenital and may affect up to 2.5% of the population - including Brad Pitt and the famous neuroscientist Oliver Sacks.

Next, you will hear what it is like to live with prosopagnosia from someone with the condition.

Bias in face recognition

Our ability to recognise faces is highly developed and different to most other objects. The existence of prosopagnosia and the fact that showing faces upside down affects their recognition more than it does other objects, is evidence of that.

Moreover, it seems that we develop skills to recognise the types of faces we come in regular contact with, which is why it is so hard to recognise an upside-down face.

One implication of this is that our face recognition processes are biased towards the faces we see from day to day. That means that these same processes are not so good when it comes to faces that we are not so familiar with. For example, research such as Harrison and Hole's (2009) has shown that we tend to be better at recognising faces from our own age group, or from age groups that we are in regular contact with. In other words, our face recognition skills are biased towards those faces we regularly come into contact with (this explanation is known as the 'contact hypothesis').

The contact hypothesis can also explain why there is a tendency for us to be better at recognising faces from our own race/ethnicity. For example, a large-scale analysis of available studies conducted by Meissner and Brigham (2001) found that own-race faces were 1.4 times more likely to be recognised than faces from another race. In addition, and of particular relevance to eyewitness identification, own-race faces were 1.6 times less likely to be mistakenly identified than faces from a different race.

The amount of contact that someone has with another race will tend to improve their ability to recognise faces from that race, but own-race bias is still a significant problem in countries that have a population of mixed ethnicities, such as the UK and USA. The research certainly suggests, for example, that Caucasian witnesses in the UK and USA will tend to be much more accurate when it comes to recognising white suspects compared to black suspects - and also more likely to incorrectly identify an innocent black suspect.