Lecture 2: Recognizing others

Question 1

why is recognising identities so important

Answer 1

recognising the identities of people we know is fundamental to being able to interact with them in terms of our past experience of them as individuals, so recognition from the face is an ability at which we become very skilled as we grow up.

Question 2

What is thus puzzling about recognising identities?

Answer 2

Puzzling how we achieve this, given the constraints on the underlying biological structure of a human face- if all faces essentially fit a common overall template, how do they also convey so accurately our individual identities?

Question 3

Importance of face perception in real world settings

Answer 3

Because faces are the most important key to identity, scientific understanding of face recognition has applications in several forensic contexts e.g. eyewitness testimony

Question 4

What would be a starting point for understanding how people recognise faces

Answer 4

• Starting point would be to understand how people describe faces when remembering them.

Question 5

What happens when people are asked to describe a face/ speculate on how individual faces may be represented in memory

Answer 5

When asked to describe faces, people list separate features (e.g. “large eyes” “hooked nose”)

Question 6

What are the neural correlates of known and unknown faces

Answer 6

Time course of face recognition
• Bentin & Deouell (2000) compared ERPs to familiar and unfamiliar faces.
• Bentin original study No difference in N170 (N170 can distinguish between faces and objects clear negative N170 for faces but not cars). Within 170ms the brain is able to recognise that it is a face but not something else but not able to recognise in that time frame that is it is a face of a particular individual.
• Presented unfamiliar faces and pics of celebrities.
• Found that there is no difference in N170 for familiar and unfamiliar. But at later time range later stages there are differences but the study was only focused on the N170.
• But: figures suggest that familiar faces elicit more negative amplitudes in a subsequent time range.]]

N250 familiarity effect- e.g. Gosling and Eimer, 2011
• Celebrity faces elicit more negative amplitudes in the N250 time range than unfamiliar faces.
• Within 250 ms brain can say that it is the face of someone you know.
It could be in principle be about any differences in the pics e.g. luminance difference. Importance of face recognition is that you recognise a face from almost any image.

Question 7

Why do we describe faces in list of features?

Answer 7

probably because our language has vocabulary items for functional parts of faces- different features have different functions- the eyes see, the jaws chew- it doesn’t necessarily follow that these are the best way to represent identity

Question 8

What happens when faces move from unfamiliar to familiar

Answer 8

When unfamiliar faces must be recognised, the external features of hairstyle and head shape dominate our memory, perhaps because these occupy a large and relatively high contrast part of the visual image.

When faces are familiar there is a shift in memory so that the internal face features become more salient (Young et al., 1985).

Question 9

Why does a shift to relying on more internal features take place

Answer 9

Hairstyles can vary across encounters with familiar faces while internal features do not, and because internal features must be attended in in face-to-face communication allowing us to create a more robust representation.

Question 10

What has been suggested that we use instead of features

Answer 10

Might not be features per se but the interrelationships between features

Question 11

What did Bruce and Young 1998 do/ show

Answer 11

Showed that when keeping the same features but altering distance between these features produces striking differences in appearance

Question 12

What is strong experimental evidence that we do not process features independently from each other?

Answer 12

Composite effect- Young, Hellawell & Hay 1987

Question 13

Outline composite effect

Answer 13

They divided faces horizontally into upper and lower halves. Although people may be quite accurate at identifying the isolated top half of a face when it is seen on its own, when it is combined with a wrong lower half it becomes harder to recognise to whom the upper features belong.
Effect is found only when the two face halves are aligned into an overall face shape. If the two halves are both present but misaligned then identification of each half is unaffected by the presence of the other one.

Question 14

What do holistic processing findings fit?

Answer 14

A distinction introduced by Carey and Diamond (1977) between Piecemeal representation of face parts versus configural representations of properties of the whole face.

Question 15

what three forms of configural processing have been proposed and by who?

Answer 15

• Maurer et al., (2002): three forms of configural processing (three ideas):
o 1st order relational configuration: two eyes, above a mouth, above a nose- overall idea that people see faces everywhere (e.g. veg arranged as a face
o Holistic or gestalt configuration: combination of different parts into Gestalt (see composite face effect)
o Second-order relational configuration: metric distances between individual parts (their positions within an overall first-order configuration) e.g. what is the distance between the eyes/ the eyes and the nose.
• Authors emphasise the value of keeping these separate where possible

Question 16

Evidence that a whole face is more than the sum of its parts using second-order relational configuration

Answer 16

Tanaka and Farah (1993) asked volunteers to learn the names of a set of faces constructed from a ‘kit’ of face features so that each face had different features. Later the volunteers had to try to identify which face feature belonged to a particular target character.

Individuals much better at doing this when the e.g. nose was shown in the context of the whole face than when it was shown in isolation. (question of which is Larry as opposed to which is Larry’s nose- same faces minus the feature being tested)

This is persuasive evidence of holistic configural processing, because second-order relational configuration was not substantially changed- parts of the face were swapped but kept in the same positions.

Consistent with this when the faces had been learned with scrambled features (violation of first-order configuration) so that they could not be learned holistically as faces, the effects reversed and participants were then better when tested with isolated features than a whole (scrambled) face context.

Question 17

What face areas are of interest in terms of face identity according to what model

Answer 17

In particular the FFA is of interest because it is identified by Haxby et al., (2000) as central to the perception of invariant characteristics of faces, which of course includes face identity.

Question 18

Importance of first-order configural representations in FFA

Answer 18

Kanwisher et al., 1997 classic study that identified the FFA as a functional brain region- the FFA is defined as the part of the fusiform gyrus that shows a stronger BOLD response to faces than other visual stimuli in an fMRI scan.

However, it can also be identified as the part of the fusiform gyrus that shows a stronger response to faces than to scrambled faces,

This fMRI contrast reveals essentially the same region. So FFA is highly sensitive to first-order configuration and many later studies bear this out.

Question 19

Talk about human reactivity to second order characteristics

Answer 19

The visual system is highly sensitive to metric distances between features. This sensitivity is much reduced when face images are rotated by 180 degrees (Maurer et al., 2002). People notice changes in distances in faces very well- led researchers to believe that the second order relational configuration was key to recognising faces- this is wrong.

We are good at these measurements but not when face is presented upside down.

Question 20

So what is the problem with metric distances?

Answer 20

o Orientation of a face changes and thus the distance between features changes e.g. a face front on or sideways.
o Change of expression e..g if a face is smiling mouth will be closer to edge of face
o (but could probably still say- we have a complicated mechanism that compensates for viewpoint by computing orientation direction or for expression by knowledge on face changes.)
o But it is complicated In reality faces are seen in 3D whereas in these presentations and manipulations of eyes moving for example is on 2D.

Question 21

What can be said in conclusion about metric distances

Answer 21

people can’t tell metric distances very well when faces are inverted, nor can they recognise people very well, which could lead to the conclusion that metric distances are non-important in facial recognition. So recognising identity might depend configural processing

Question 22

What is evidence for the inversion effect

Answer 22

Yin- 1969

Bruce and Langton (1994)

Question 23

Classic demonstration of inversion effect

Answer 23

Yin- 1969- showed that recognition memory for unfamiliar faces was more severely affected by inversion than the recognition of various other kinds of pictured item.

Question 24

Further evidence for inversion effect

Answer 24

o Bruce & Langton (1994): Accuracy for famous face recognition drops from 95% in upright condition to 70% in inverted condition. Doesn’t mean that people can’t do it at all but it requires more effort and more errors are made.
o Perceiving subtle aspects of face configuration depends on normal, upright, orientation- despite inverted faces carrying the same physical information than upright faces
o Difficulties in recognising inverted faces may result from the relative insensitivity for configural properties.

Question 25

What is consistent with inversion effect? And adds to evidence?

Answer 25

Composite effect disappears when the face is inverted- Young Hellawell and Hay, (1987) showed that people are actually better at identifying one of the half faces in an inverted composite than in an upright one.

Question 26

Explanation for disappearance of composite effect when inverted

Answer 26

Whilst this may seem paradoxical, given that upside-down faces are harder to identify, the paradox can be resolved as follows- The reason that half a face within an upright composite is difficult to identify is because a new face identity arises from the combination of upper and lower face features. If the perception of configural information is particularly impaired by inversion and an inverted composite does not get seen as a novel configuration, making it relatively easier to access the identity belonging to the top or bottom features.

Question 27

Real life evidence for expertise in faces

Answer 27

Prosopagnosic indiviudals
May be unable to recognise familiar people via their faces but able to recognise the same people from their voices or names. Despite severely impaired ability to recognise faces, people with prosopagnosia may have relatedly spared abilities to read other messages from faces, such as emotions or facial speech. This suggests that their problem is not a generalised deficit in perceiving faces but rather linked to some more specific aspect of face perception needed to recognise individual identity.

Question 28

What do people with prosopagnosia have difficulty with and evidence for this

Answer 28

Particular issue processing configural information, and perhaps as a result are relatively unaffected by the inversion of faces.
Busigny and Rossion (2010)- inversion effects for PS woman with prosopagnosia- across a range of tasks her performance did not differ substantially between upright and inverted faces. They reviews the literature on acquired prosopagnosia and found that this lack of face inversion effect is the typical pattern.

Question 29

What are those with prosopagnosia also less sensitive to? And evidence for this

Answer 29

Configural properties of upright faces than are controls. Busigny et al., 2010) report further work with PS looking at her holistic configural processing abilities.
Experiment 1- carried out a whole-part face task similar to Tanaka and Farah- Participants saw an intact whole target face for 1500 ms and after a short delay were shown the target alongside a distractor that varied only in a single feature. The test pair could be whole faces or isolated features.
Controlled participants shows on average higher accuracy and faster responses to whole than in part test pairs
PS was more accurate on the part than the whole trials
ALSO
Same experimenters
Composite effect and showed that while control participants were generally less accurate and slower when the two halves of composite faces were aligned, or when the irrelevant bottom half of the face showed a different identity to that of the relevant top half, this was not observed for PS

Question 30

What do prosopagnosic individuals show

Answer 30

That clearly being able to holistically process faces is to do with recognition

Question 31

Evidence that spacing changes hugely with angle

Answer 31

Burton et al., 2015
o Different photographs of politicians all front on so not a difference in angle change.
o In different images of the same person these metric differences are quite different and this is not due to orientation changes.
o Sometimes in the photos there was a smile but even just looking at distance between eyes and nose changes quite substantially.
o Standardised the image size by the size of the pupil and measured distance between outer edge of one pupil and outer edge of others.
o There was a huge variability between images of the same person.
o If something like this changes so much from pic to pic it probably isn’t stable enough information to be able to recognise someone and use it as a recognition tool.
o Do not have recognition problems even though metric differences between faces changes a lot.

Question 32

Evidence for distortions against second-order configurations

Answer 32

Hole et al., 2002
o Faces remain recognisable despite severe distortions of second-order configurations.
o Produced these by laterally or vertically stretching or squashing the face.
o These manipulations dramatically change 2nd order configurations yet the person can still be recognised.
o Such stretched faces can yield behavioural results that are not significantly different from the unstretched originals.

Question 33

electrophysiological response for distorted faces evidence

Answer 33

Burton et al., 2008
Showed that such distorted faces can yield electrophysiological as well as behavioural responses indistinguishable from the original photographs of these people.

Question 34

But how can stretching studies not even be explained by holistic processing?

Answer 34

Simple 2D relationships are not preserved in stretched faces, but nor are ‘holistic’ patterns obviously preserved either. Should a ‘half-distorted’ face not match a simple holistic template better than a fully stretched one?

Question 35

What do Bruce and Young, 2012 suggest are issues with our understanding of face representation and what do they suggest instead?

Answer 35

Potential implications for our understanding of face processing are that either the spatial relationships coded from faces must be based on a very sophisticated set of coordinates, which encompass the coding of the shape of individual features as well as more distant spacing, or that faces are not decomposed via these kinds of explicit measurements at all –
for example, they might be analysed as something like arrays of pixel intensities, though some kind of pre-processing would still be needed to accommodate the stretching effects we have just discussed

Question 36

where have we seen shape and surface pigmentation being important?

Answer 36

We have seen in earlier chapters the importance of 3D and surface pigmentation for the perception of such things as the sex or age of faces.

Question 37

What is the 3D shape of the human face revealed from?

Answer 37

The 3D shape of the face is revealed from a combination of the surface features which are visible and the pattern of light and shade which can be used to derive surface information

Question 38

Preliminary evidence that 3D shape might be important in face recognition

Answer 38

comes from the advantage sometimes found for ¾ views rather than full-face images in face recognition. An angled view reveals more about the way that a face is structured in depth. The shape of the nose in particular is very difficult to see in a full-face image and much clearer from an angle

Question 39

evidence that it is not just 3-D information and what does this imply

Answer 39

Bruce et al. (1991): presentation of laser scans (3D shape only) and photos to friends of depicted people.
• Identification rates remarkably lower for laser scans than photographs of individuals with their eyes closed and wearing bathing caps
• Highlights need for no colour/ pigment info. Would recognise all friends in photos.
•  people can use shape info BUT skin pigmentation contributes to recognition

Question 40

Difference between 3-D and second order processing

Answer 40

Different is now talking about shape info in 3D whereas stimuli for second order studies was 2D stimuli

Question 41

What is a relatively easy/ one of the most disruptive transformations that can be made to face image?

Answer 41

put it into photographic negative. Upside-down faces are diffi cult to recognise, but photographic negatives are even harder.

Question 42

Evidence that photographic negatives very hard to negate

Answer 42

Bruce and Langton (1994) compared the effects of inverting and negating famous faces and found negation had an even more detrimental effect than inverting the image.
• Upright faces were named correctly 95% of the time,
• inverted 70%.
• negated images were named only 55% on average
• inverted negatives only 25%.

Question 43

What does Bruce and Langton’s study suggest

Answer 43

The effects of inversion and negation were additive, suggesting they affect distinct aspects of the face or distinct processing stages. If inversion prevents confi gural processing, negation must hurt something quite different (skin pigmentation)

Question 44

what do normal photo negatives reverse? And what is it possible to do?

Answer 44

two separable aspects of the image – the brightness and the colour values. However, it is possible to manipulate these separately, reversing only brightness (making the light regions dark, and dark regions light) or colour (making red hues green, blue hues yellow, and so on).

Question 45

Study on luminance and colour effect

Answer 45

Kemp et al., 1996
• Can manipulate just colour and also just reverse luminance.
• Normal photographic negation reverses brightness and the colour values.
• Hue and luminance negation can be manipulated separately
• Colour negation has little effect on identification- colour info has probably not a lot to do with face recognition e.g. can recognise people in black and white or when in different coloured light.
• Brightness negation is very disruptive.- very difficult to tell individual when brightness is inverted.

Question 46

why has it been argued that brightness negation is so disruptive

Answer 46

o Why? Two potential arguments:
o 1) Is it difficult to derive 3D shape from shading? E.g. shape of the nose throws shade onto the face so can tell about shape of the face from shading information. But… Negation does not additionally reduce identification of “shape-only” laser scans (Bruce and Langton 1994) these have no pigmentation (these are pure shape) people are not super but they can do it. If you then make an negative of this bust then performance doesn’t change- if no pigmentation info in positive then none in the negative so probably not about cues to shading.
o 2) so it more likely seems that reversing brightness of important pigmented areas “disguises” faces. E.g. The reversal of brightness in the eyes may make it difficult to encode the face

Question 47

Study on the importance of eyes in negation

Answer 47

Gilad, et al., (2009): It seems that some parts of the faces are more important for this pigmentation disguise effect
Presented negative photos where either everything was negative or where the face was negative and the eyes were positive.
Full positive= 100%, full negative= roughly 55% so dramatic drop in performance, contrast camera (positive eye region)= only slight drop 92% ish.
• If eye region is positive within otherwise negative images (contrast chimeras), detrimental effect of negation largely eliminated!
• Positive eyes in blank silhouette is not sufficient
• Positive eyes allow coding of useful information from the remainder of the negative face!

Question 48

what does Gilad et al.’s 2009 study suggest?

Answer 48

means that face representations must significantly favour, or be built around, 2D ordinal contrast relationships around the eyes

Question 49

so is it all about the eyes?

Answer 49

No as we know people are very bad at recognising different eyes when they are presented as a whole face so it is not that we are purely looking/ using eyes.

Question 50

Evidence that it is not all about the eyes

Answer 50

Gilad et al., 2009 when only positive eyes were presented with grey filling out the rest of the face performance dropped to around 12-15% correct.
In fMRI also showing just the eyes in an otherwise blank silhouette evoked much less strong FFA activation to full images/ contrast chimeras of eyes

Question 51

Evidence that not just about positive eyes

Answer 51

Sormaz, Andrews & Young, 2013:
Showed full positive, full negative and then different chimeras e.g. forehead/ mouth/ nose/ eyes in +contrast rest in negative
Chimera effect is not about positive eyes per se
• Information from the negated part of the face “becomes useable” when eyes are positive
• When eyes are positive and presented alone then recognition accuracy becomes poor
• Thus seems it has to be positive eyes in a negative face
• Are representations of familiar faces centred around the eyes?

Question 52

So what overall can be said about the eyes?

Answer 52

So is not about eyes per say but is about positive eyes in a negative image. It seems to be that the eyes allow us to make use of the rest of the face. Thus one idea is that face recognition is centred around the eyes and are the most important part.

Seems to be the case that surface information from the eye region is of importance… so from this we would conclude that the eye region of a face would be better remembered and that we hold more info about eye regions than we do about the rest of the face.

Question 53

Intro to distinctiveness and caricaturing

Answer 53

Although all faces are built to the same basic template, some faces deviate more from the average or prototype face, while other faces have a more average or ‘typical’ appearance

Question 54

Evidence that more distinctive faces are recognised better

Answer 54

More distinctive faces are recognised better than more typical faces- known in daily life but also in studies. (face in crowd tasks have shown this)
• Famous faces recognised faster when distinctive (Valentine and Bruce, 1986)
• Unfamiliar distinctive faces more likely remembered in memory tasks (Bartlett et al., 1984)

Question 55

what is one way to understand the effects of distinctiveness?

Answer 55

Multidimensional Face-space framework introduced by valentine (1991)

Question 56

Outline the Multidimensional Face Space (MDFS) framework

Answer 56

• Any face can be described by values along a number of dimensions of facial variation these dimensions describe particular characteristics of a face.

e.g. have a dimension on eye form and nose length then would say there is an average value on these dimensions and probably most people would be close to this average and a few people would have extremes of these dimensions. Have a normal distribution around these dimensions.

• Because the model is from 1990’s this is why it argues about dimensions as nose length as same time as 2nd order studies, but could also think of these dimensions as colour pigmentation or contrast around the eyes.

Question 57

In the multi-dimensional face space model what can we say about distinctive faces?

Answer 57

If model like this then can think that distinctive faces are the faces which do have extreme values and maybe this is why they are easier to recognise.
Some people have more extreme variations to average so these will be further from the centre thus as most near to centre there would not be people in these more outer areas

Question 58

How is it proposed we recognise faces using the face space model?

Answer 58

When we see a face and want to recognise it from the stimulus would extract these values and project the memorised face onto the seen face and see if face stored representation matches face presented. And have a margin of error so that if there is a match and you can see the face in the area then you will match it and accept ok this is the face.
• Identification require comparison of the dimensions of to-be-recognised face with dimensional descriptions of already stored faces.
o In case of single match the face is recognised

Question 59

Explain using face space model why then recognising a more distinct individual Is easier

Answer 59

If the face is more typical and more average it will be more easily confused with someone else. If a more extreme face is easier to say ok it is that person.
• Typical to-be-recognised face falls into regions where many descriptions are stored more difficult to get unambiguous match.
o More precision on some dimensions needed, which takes more time and is more error prone.
• Distinctive faces lie in less densely clustered regions  easier match.

Question 60

In the face space model where are typical faces

Answer 60

• Typical faces cluster in the centre, distinctive faces are scattered around the periphery

Question 61

What can the face space model do well as a model

Answer 61

Thus, is a model that can well explain why distinctive faces are far easier to recognise and unambiguously say it is a certain person. Distinct faces lie in less densely populated areas of facial representations.

Question 62

What may distinctiveness/ face space model help to explain and why?

Answer 62

This may explain why we recognise characatires

• Caricatures exaggerate individual’s idiosyncratic features –> exploit distinctiveness.
• In terms of the face-space framework, the caricature produces a description which is more extreme (further out in space periphery) than the actual individual.
• Thus in consequence this representation becomes less likely to be confused with any face other than the individuals

Question 63

Computer technique for caricatures

Answer 63

Brennan (1985) first reported a technique for generating caricatures automatically by computer, and her technique rests on the suggestion that a caricature ‘is a symbol that exaggerates measurements relative to individuating norms’

Question 64

What did Brennan do?

Answer 64

digitised photographs of faces and located a set of 186 key points which described the major face features.
For example, points within the set of 186 included the inner and outer corner of each eye and a small set of additional points along the upper and lower eye-lids.
These points could be linked together to form a line-drawing of the original face- put lots of these together and have a line drawing of an average face
To caricature a specifi c face, the locations of the individual points for that face are compared with the average and exaggerated by multiplication.

Question 65

Talk about caricaturing and anti-caricaturing

Answer 65

• You can make a face more distinctive by caricaturing.
• Anti-caricaturing reduces the differences between an individual and an average face, and thus creates less distinctive faces.

Question 66

Support that caricatures enhance recognition first study

Answer 66

Rhodes et al., 1987
produced line-drawn caricatures like this, using smooth curves rather than straight lines to join up the points. The stimulus faces were personally familiar to the participants in their experiments, and Rhodes et al. (1987) were able to confi rm that positive caricaturing of this kind made line drawings of faces more recognisable.
Also found faces were made less recognisable by making ‘anticaricatures’ that shift the features towards the average face – in effect making a face more typical in appearance.

Question 67

Support that caricatures enhance recognition second study

Answer 67

• Rhodes and Tremewan (1994): Caricatures of line drawings recognized better than undistorted drawings. This makes sense as you are making the individual parts of a person more individual.

Question 68

But what are issues with traditional caricatures

Answer 68

• But: line drawings are impoverished stimuli (e.g., no texture information).

Question 69

What can be done to overcome issues with line drawn caricatures?

Answer 69

the same techniques can be applied to full photographic-quality images as well as with line-drawings, to produce morphs and caricatures of face images.
All that is needed for photorealistic image manipulation is a way to combine the surface textures of the images with the set of tesselated regions that results from the positions of the fiducial points.

Question 70

What would photographic caricatures in theory give?

Answer 70

More scope for recognition by the exaggeration of distinctive features.

Question 71

What is surprising about findings on photo caricatures

Answer 71

• Photo-realistic (shape) caricatures not easier to recognise then veridical photos (e.g., Kaufmann & Schweinberger, 2010).
• the recognition advantages of full photographic caricatures are found only infrequently and under circumstances where recognition is made more diffi cult (e.g. brief presentation – see Rhodes, 1996 for a review).

Question 72

Explanation as to why line drawings beneficial effects so much better?

Answer 72

This is because a line drawing hugely reduces the information available to the perceiver, leaving more scope for recognition to be aided by the exaggeration of distinctive features
 Under impoverished conditions, exaggerating spatial information can help; under most everyday conditions, face recognition is provided by a multiplicity of cues; enhancing one probably does not help much.- this is again an argument for the importance and particular relevance of of surface area and pigmentation information. Doesn’t make much of a difference because on the basis of a photo can recognise surface information also which we know is important for face recognition and you will recognise the face based on this info also.

Question 73

What is overall conclusion from line drawing studies

Answer 73

The point is that it can not be that simple and just be about shape but is about dimensions but these are probably more about pigmentation.