Recognition of unfamiliar faces across viewpoints: a beneficial effect of the

crumcasteAI and Robotics

Nov 17, 2013 (3 years and 8 months ago)

113 views

Exam number 4193800


1





Recognition of unfamiliar faces across viewpoints: a beneficial effect of the
mid
-
profile view on recognition.


Exam number
:

4193800


MSc Human Cognitive Neuropsychology


The University of Edinburgh


2010






















Exam number 4193800


2

Abstract


The present stud
y investigated two main areas of face recognition. The first aim was
to provide supportive evidence for the distinctiveness/fluency account of memory
dissociation. Specifically, it hoped to find that: 1) rating faces for distinctiveness
elicited more remem
ber responses, and 2) sorting faces into categories elicited more
know responses. Results showed a trend consistent with these predictions but this was
not significant (p=0.288), and the null hypothesis was accepted. The second aim was
to find supportive e
vidence for the so
-
called ‘mid
-
profile’ advantage in face
recognition. The mid
-
profile was investigated in both same
-
angle (i.e., same angle at
study and test) and different
-
angle (i.e., frontal at test and mid
-
profi
le at study or vice
versa) conditions
. T
his effect was not found in the same
-
angle data although the
results showed the expected trend (p=0.08). In the different
-
angle data, recognition
performance was found to be significantly better when faces were studied at frontal
and tested at mid
-
profile
view than when studied at mid
-
profile and tested at frontal
view (p<0.05). This last effect has been interpreted as showing a test
-
view advantage
fo
r the mid
-
profile perspective.


















Exam number 4193800


3






Introduction



The information we derive from faces

is

essential for effective social
communication

(
Gobbini & Haxby, 2007)
, and humans have evolved specialised
abilities to recognise and discriminate between faces (Goffaux, Rossion, Sorger,
Schiltz & Goebel, 2009)
.

Furthermore, r
esearch into the processing o
f unfamiliar
human faces has a wide range and reach of implications; findings affect areas such as
research into artificial face processing, the study of human social interaction, and
eyewitness testimony (Hancock, Bruce & Burton, 2000; Bindemann, Scheeper
s &
Burton, 2009).

This introductory section will first establish some of the most important
concepts in face recognition and the methods most commonly used to investigate
them. The main intentions of the study will be explained fairly briefly before the
t
heoretical considerations that stimulated them are discussed.
D
ual
-
process models of
memory will be summarise
d and the relevant

theories regarding dissociations in facial
recognition
discussed. Some aspects of the methods used in the study will be outlined

along with the theoretical and/or practical justifications for these, before the main
aims are summarised along with their corresponding major hypotheses. Evidence will
be provided from neuropsychological and brain
-
imaging
studies

and from
performance by

healthy subjects in cognitive tests
of face recognition, as well as other
areas, to help illustrate the points made throughout the text.


Main A
ims

The aims of this study were twofold. The first primary aim was to add
evidence to the growing body of resea
rch showing that different types of cognitive
processes employed when viewing a face during encoding elicit different forms of
memory retrieval at test. More specifically, that judging faces for distinctiveness
(typically using a four
-

or five
-
item rating
scale) affects one type of memory retrieval
(remembering), whereas sorting faces into categories based on their similarity to
Exam number 4193800


4

others results in another (knowing). The second primary aim was to find evidence for
the mid
-
profile advantage in two sets of view
ing conditions: the first condition
investigating recognition of faces at frontal/mid
-
profile views when they are presented
at the same angle at study and test (same
-
angle condition), and the second condition
comparing recognition of faces presented at dif
ferent views at study and test
(different
-
angle condition). As Troje and Bülthoff (1996) note, the ‘mid
-
profile’ view
is not always specifically defined throughout the face recognition literature, but for
the stimuli in this study at least, mid
-
profile ref
ers to an angular rotation of the face
that is approximately 45 degrees from full frontal and equidistant from the profile
view. The terms view, angle, perspective and pose are used interchangeably to refer to
the angular rotation of the face (here, either

frontal or mid
-
profile).

The text will also
occasionally refer to

test
-
view


and

study
-
view


advantages. To clarify, a test
-
view
advantage is said to exist when the beneficial effects of a particular view become
apparent only at test, while a study
-
view

advantage is declared when better
performance is recorded after a certain view has been presented at study.


Some studies into recognit
ion of unfamiliar faces (i.e.
, faces that have o
nly
been viewed once by

subjects) have found that the
cognitive processe
s employed
during encoding has an effect on the type of memory retrieval that takes place during
te
st; more specifically,

rating

faces for distinctiveness promotes episodic recollection,
whereas sorting faces into categories based on the similarities betwe
en them has been
found to result

predominantly

in
the

form of

recognition that is known as a ‘feeling of
familiarity’ (
Mäntylä, 1997;
Mäntylä & Holm, 2006).

Mäntylä & Holm (2006) appeal
to a particular theoretical framework, known as the distinctiveness
-
fl
uency model,
which makes two
qualitatively different
prediction
s

about memory retrieval. One
prediction is that
remembering will be enhanced by cognitive processes that focus on
distinctive or salient aspects of stimuli (Rajaram, 1996).

Distinctive process
ing might
take place, for example, when the subject views a face with a particularly long nose,
which the subject can then recollect at a later point.
The second prediction is that
familiarity memories will be increased by cognitive processes that emphasis
e ‘fluent
processing’ of stimuli (Rajaram, 1996).
Fluent processing is said to take place when
the conditions at study and test (e.g., the size of the stimulus) are very similar.
It is
hoped that the results of the present study will add useful informatio
n to the literature
on dual
-
process models of memory.

Exam number 4193800


5


Over several decades if not longer, face recognition researchers have been
interested in the question of whether there is a canonical view of the face

(Troje &
Bülthoff, 1996)
. One much
-
disputed
theor
y

is that the mid
-
profile

(or portrait)

view
provides the canonical or optimal view for face encoding and recognition (Liu &
Chaudhuri, 2002).
Canonical perspectives have been defined by Palmer, Rosch and
Chase as

“the view

that reveals the most informatio
n of greatest salience” (1981,
p147)
.
The present study compared recognition of faces at frontal and mid
-
profile
views in order to
find supportive evidence for a

mid
-
profile advantage.

Were such an
optimal view to be proven (to a significant degree) to exi
st, the implications would be
especially relevant to forensic psychology. Although many factors associated with
witnessing a crime


such as changes in the suspect’s appearance, the trauma of
witnessing the crime, and interference from viewing other photog
raphs/suspects
before seeing the correct one (Laughery, Alexander & Lane, 1971)


can reduce the
probability of a correct identification, when photographs of suspects are presented at
the view most conducive to accurate recognition, correct identification
of criminals
may be improved. Furthermore, evidence of a mid
-
profile advantage should
encourage greater use of this view in

computational models of face recognition
, which
predominantly
operate exclusively on

the frontal view of faces
(O’Toole
, Edelman &
B
ülthoff, 1998).




The remember/know paradigm and its relation to theories of consciousness
.
The present study employed the remember/know paradigm to investigate both of the
main aims of the study.
The remember/know paradigm has been used in numerous
studi
es to ascertain the type of recognition participants experience, where ‘remember’
and ‘know’ refer to two distinct types of memory retrieval

(Rajaram, 1996;

Mäntylä
& Holm, 2006)
. Converging lines of evidence have been found to support this
distinction (Mä
ntylä, 1997; Yonelinas, 2002
; Rajaram, 1996
), which

is one that
subjects can consistently and accurately recognise and put into practise (Rajaram &
Roediger, 1997)
.
The technique
-

of educating subjects about the nature of the
remember/know distinction in
order to retrieve useful measures of the type of memory
they experience


has sometimes proven problematic but sizeable quantities of data
revealing reliable effects have still been produced (Rajaram & Roediger, 1997).
Remembering

(R)
has been described as

an experience of explicit recollection of
certain details of the item, such as

its meaning,

physical properties or ideas associated
Exam number 4193800


6

with the item (Rajaram, 1996). Knowing
(K)
has been described as a feeling of
familiarity that lacks explicit recollection
of any particular details of the item
(Rajaram, 1996).
The distinction between remembering and knowing (or recollection
and familiarity, respectively) has found support in areas such as
electroencephalography (EEG) research

(Tsvilis, Otten & Rugg, 2001)
,
n
europsychological studies

(Yonelinas, 2002)

and eye
-
tracking research (Mäntylä &
Holm, 2006).

For example, it has been discovered that,
unlike know responses,
subjects’ probability of a remember response is negatively

affected by divided
attention at study

(Mäntylä, 1997
)
.

Evidence for both structural
(Wheeler & Stuss,
2003; Rajaram & Roediger, 1997)
and functional
(Rajaram, 1996)
dissociations has
been found,
which has been instrumental in the construction
of several dual
-
process
models of memory (Yonelina
s, 2002).

Furthermore, it is assumed throughout

much of

the literature that remembering is consciously controlled and knowing is automatic
and unconscious (Jacoby, Yonelinas & Jennings, 1997). It is necessary to
consider

some of the main arguments concerni
ng the relationship between conscious and
unconscious processes in order to fully comprehend the theories concerning the
remember/know distinction
, as the distinction is based in substantial part on the
relationship of conscious to unconscious influences
.
The arguments regarding the
relationship between the conscious/unconscious

will be introduced and discussed in
the
context of the remember/know paradigm
,

in the
paragraphs that follow.


The problem of defining conscious and unconscious processes is centuri
es old
(Rajaram & Roediger, 1997; Jacoby et al., 1997) and it is not possible to

meaningfully
summarise it let alone

have a full discussion here. Nevertheless, some important
points

regarding their relationship to each other
should be covered, and examples

will
be used to illustrate.
There appear to be three main
theories concerning

the
relationship between conscious and unconscious processes in the literature
(Kopelman, 2002).

Jacoby, Yonelinas and Jennings (1997) describe

the three main
theories regarding

the relation of co
n
sciousness to unconsciousness

as follows
. The
first is that

that
consciousness

can be fully differentiated from
unconsciousness

although the two m
ay overlap (i.e., independence). The second is that consciousness
and unconsciousness are
distinct and never overlap (exclusivity). The third is that
consciousness cannot be present without unconscious processes (known as
redundancy). Since remember memories are identified with conscious control, and
Exam number 4193800


7

know memories with automatic (or unconscious
) control, the relationship between the
two types of memory retrieval can be seen to mirror that of consciousness to
unconsciousness. Most theorists seem to accept that remembering and knowing are
related in some way (i.e., independence or redundancy relat
ionshi
p)

and that they
operate in parallel (K
opelman, 2002; Yonelinas,
2002).

The unconscious is commonly
related to performance in tests of implicit memory (such as word
-
fragment
completion,) but, problematically, conscious processes may still affect this

(Jacoby et
al., 1997).

Gardiner, Java and Richardson
-
Klavehn,
advise

that conscious control on
the one hand and awareness on the other (and unconscious control and unawareness)
should not be thought of as identical in nature, though there may be many case
s where
the two terms are interchangeable; for example, a subject may be aware of the
unconscious recollection of a word (1996).

One avenue of research that has been used to investigate the relationship
between conscious and unconscious influences is the p
rocess
-
dissociation model. This
involves using tasks where the instructions
-

it is assumed


constrain the s
ubject to
complete the task with

either inclusion or exclusion of conscious control

(of the
relevant processes)
, hence the use of term ‘process
-
dis
sociation’
.

One such task
described by Jacoby et al. (1997) is

referred to here as word
-
stem completion
.
Subjects are presented with two lists of words


one verbal, the other written.
Afterwards they are presented with word
-
stems which they have to try t
o complete.
They are given either ‘inclusive’ or ‘exclusive’ instructions: in the inclusive
condition, subjects are

instructed
that they may use words that they recall from earlier
in the list, and

that

if they cannot recall a word then they should use the

first word that
comes to mind; in the exclusive condition, subjects are instructed not to use the words
that were presented earlier in the experiment, but to use
the first new

word that comes
to mind (Jacoby et al., 1997).

To put it into other words
: in t
he inclusive condition,
subjects are ‘allowed’ to exercise conscious control (by using words that they
explicitly recall from the study section), but in the exclusive condition it is assumed
that the words subjects use to complete the word stems are uncons
ciously produced.

The first assumption is that, whenever conscious and unconscious processes overlap,
the conscious processes dominate. The second assumption is that, following
instruction set 1), subjects will recall words from the study list via both con
scious and
unconscious processing.

Exam number 4193800


8

Gardiner

and colleagues

warn against conflating an independence model of
remember/know memories with the pattern of results obtained in process
-
dissociation
(i.e., inclusion/exclusion) paradigms

(1996)
. Instead, they sugg
est that it is better to
conceptualise
models of consciousness and the inclusion/exclusion paradigm

as
“…different, but complementary,” (1996, p120).

Joordens and Merickle point out in a
similar vein, that different assumptions underlying the results of th
e above
-
described
process
-
dissociation procedure can be used to validate either redundancy or
independence models of consciousness.

The semantic versus episodic processing account of the remember/know
distinction
.
Endel Tulving
(1985)

is attributed with
fi
rst proposing

the
remember/know distinction
(based on
episodic and semantic forms of

memory
,
respectively
)
that

has been used so extensively (Kopelman, 2002). T
he

cognitive

mechanisms t
hat

underlie
the two types of recognition
and how they relate to each
o
ther
have been extensively disputed and alternatives put forward

(Yonelinas, 2002;
Rajaram, 1996)
. Tulving’s original conception of remembering was that i
t is enabled
by episodic memory. That is

to say
, that remembering is autobiographical
,

in that it
pert
ains to memories that
include
the awareness of one’s own presence

(Wheeler &
Stuss, 2003)
. For example, a subject might say they remember a certain face because
they distinctly recall thinking to themselves that it was a pa
rticularly attractive face.
This

conception of

the phenomenal experience of

remembering

appears largely to
have been adhered to in the literature (Rajaram & Roediger, 1997)
, although opinions
differ on certain details (Yonelinas, 2002)
.
Semantic memory includes knowledge
‘that’


for exa
mple, the knowledge that Edinburgh is the capital of Scotland.
Semantic memory is said to process meanings of things like words and pictures, and
more abstract items such as superordinate and subordinate categories (Kopelman,
2002).

Semantic and episodic
memory
processes
are
thought

to be related, but the
nature of this relationship is uncertain

(Kopelman, 2002)
.

Tulving’s conception of the
two
asserts

that information must pass through the semantic system before it can be
encoded episodically

(i.e., a red
undancy model)

(Yonelinas, 2002).

Kopelman
describes MRI research on patients with semantic dementia, which found that the
brain structures found to ‘accommodate’ semantic processes include the pole of the
Exam number 4193800


9

left temporal lobe, the left posterior temporal g
yrus, and the inferior and lateral
temporal gyri. The right temporal lobe is also implicated to varying degrees
(Kopelman, 2002).
Behavioural evidence from another study was taken to suggest
that, “…current autobiographical experience interacts with and fa
cilitates the
maintenance and/or retrieval of residual semantic knowledge in semantic dementia
patients,” (2002, p2165). This is particularly relevant to face
-
processing research, as
it supports Bruce’s assertion that identification of faces is to some ex
tent context
-
dependent (Bruce, 2009)



on which there will be more discussion later
. Tulving’s

idea that

knowing


is supported by the semantic system has been
adapted more
recently and characterised as ‘familiarity’ (Rajaram & Roediger, 1997
; Jacoby et al
.,
1997
).

That remembering (unlike knowing) is enabled by episodic processes is
supported in the neuropsychological literature. Wheeler and Stuss (2003) refer to
findings from patients with frontal lobe injuries, for instance, who have been found to
be sel
ectively impaired at remembering, even though their recognition rates were
similar to those of healthy controls.
This finding is consistent with a redundancy
model of memory. Wheeler and Stuss, 2003,

connect the episodic processes that
support remembering
with mental time travel (consistent with Tulving’s original
conception) and emphasise the importance of (frontally mediated) functions that
require awareness of self and self
-
continuity. They cite anecdotal evidence from
patients who had undergone frontal
lobotomies, who were said to describe past events
“as if they were describing someone else,” (2003, p828). Quantitative evidence
comes from research with patients with frontal lobe lesions; for example, source
memory (i.e., memory of where or when an item

was previously encountered) is
impaired in these patients even though recognition rates are close to normal (Wheeler
& Stuss, 2003).

Further neuropsychological e
vidence for remember/know distinction comes
from studies on people with either left
-

or right
-
hemisphere temporal lobe epilepsy.
Rajaram and Roediger (1997) cite a study, which found that patients with left
temporal lobe epilepsy (L
-
TLE) relied more on know responses in a test of picture
recognition. Conversely, patients with right temporal lobe e
pilepsy (R
-
TLE) gave
more remember responses. It was inferred from the combination of these results and
Exam number 4193800


10

the authors’ understanding of the cognitive basis of the r/k distinction, that the right
temporal lobe is more specialised for perceptual processing whe
reas the left temporal
lobe is more specialised for conceptual processing (Rajaram and Roediger, 1997).
When normal subjects either processed the stimuli semantically or superficially
,

more
remember or know responses (respectively) were elicited. However,
when the L
-
TLE
patients were instructed to encode the stimuli semantically, they still produced a high
number of know responses, and the R
-
TLE patients were correspondingly more likely
to give remember responses after superficial encoding (Rajaram and Roed
iger, 1997).
Quite aside from consideration of the exact nature of remember versus know
responses, it is clear that the two types of recognition are different in a significant
sense, and mediated at least partially by different structures in the brain.

Th
e evidence cited here
not only
supports t
he remember/know distinction, but

is

also

consistent with the idea that remember and know judgements are based on
conceptual and perceptual processes respectively.
New dual
-
process models of
memory have been propose
d and are summarised briefly below. Some adhere to the
conceptual/perceptual distinction

(e.g., Mandler, 1980)
. Some conceptualise the
distinction differently
,

since

further

evidence has been found that

seems to be
inconsistent with the conceptual/perceptu
al explanation

(Rajaram, 1996). For
example, knowing has failed to make up a larger proportion of results despite
increasing the perceptual similarity between verbal study (auditory versus visual
words) and test (visual) items (Rajaram, 1993).

Alternative

dual
-
process models of memory
.
In his review paper, Yonelinas
outlines several of the dominant models of dual
-
processing that encompass the
remember/know distinction (2002). He notes that there are several avenues of
agreement, such as the statement that
knowing (or familiarity) is the faster of the two
processes; but also that there are several areas of disagreement. Yonelinas’ own
model characterises judgements of familiarity as arising from a quantitative
assessment

(i.e., signal detection criterion)
, w
hereby a level of familiarity that rises
above a certain threshold is characterised (and felt) as a feeling of familiarity
.
Alternatively, recollection is a qualitative assessment, so that even one recollected
item can lead to a judgement of ‘remembering’.


Exam number 4193800


11

A more recent explanation of the remember/know distinction, based on the
transfer
-
appropriate processing (TAP) principle, asserts that remembering is
associated with conceptual processing whereas knowing is related to more
perceptually
-
driven processes (
Mäntylä, 1997). The transfer
-
appropriate processing
account states that, when there is a greater match between the cognitive processes that
take effect at study and at test, memory performance (either accuracy or speed or
both) will improve

(Rajaram, 1996
)
. This advantage is

illustrated

in the finding that
performance in a test of
word
-
stem completion is better when there is a match in the
mode of presentation of study and test stimuli: for instance, in perceptual priming
experiments, when words are used i
n both parts of the experiment, as opposed to
pictures at study and word
-
stems at test (Rajaram & Roediger, 1997). Althoff and
Cohen (1999) found evidence to support this in a study of eye
-
movements to famous
and unfamiliar faces; eye
-
movements to famous f
aces were qualitatively different to
those of unfamiliar faces, showing a ‘re
-
processing’ effect.

In another popular model credited to Mandler

(1980
)
, recollection is thought
to constitute a ‘search p
rocess’ that is activated when familiarity does not rea
ch a
certain threshold. On the other hand,

a judgement of familiarity is said to be enabled
by the “intra
-
item integration of the perceptual aspects of that item in memory,”
(Yonelinas, 2002, p445).
That is to say, that familiarity memories are formed on t
he
basis of the integration of the sensory/perceptual aspects of an item.
This last
definition of familiarity is similar to that
put forward by

Jacoby

et al. (1997
), who
equates familiarity roughly with judgements of processing fluency: Jacoby contend
s

tha
t, when processing fluency of a stimulus is recognised as resulting from prior
experience with that stimulus, then a ‘feeling of familiarity’ ensues. The Jacoby
model differs from the Mandler model in that processing flue
ncy may act on both or
either

perce
ptual and conceptual features (Yonelinas, 2002). The Jacoby model also
holds that processing fluency is connected with performance in implicit memory tasks
(such as word
-
fragment completion)

in that both are unconscious (Jacoby et al.,
1997)
.

Furthermore,
recollection is said to o
ccur when contextual or ‘elaborative’
information
that was associated with an item at study is retrieved at test

(
Jacoby et al.,
1997
)
.

An example of

elaboration is found in Laughery, Alexander and Lane (1971)


it is reported that

when people create (or generate) mental images of to
-
be
-
remembered stimuli, recall is significantly better than when this technique is not used

Exam number 4193800


12



this mental image that was created at study may be ‘remembered’ at test

and would
qualify as a form of elabor
ative information.


Rajaram and Roediger’s
(1997)
proposed conception is de
rived from that of
Jacoby

et al.
, and they emphasise a model of memory that accounts for the
remember/know distinction through a suggested dissociation between distinctive
processin
g and fluent processing (1997).

This is partly based on observations of
phenomena such as the level of processing effect, where it has been found that
‘remember’ memories are affected by manipulating level of processing,

but ‘know’
memories are not (Ra
jara
m & Roediger, 1997).

The effect of ‘distinctive’ processing
is held to arise when elaborative encoding takes place.

Rajaram (1996) disputes the
long
-
held theoretical argument that holds that remember responses are only affected
by changes in conceptual var
iables and know responses by changes in perceptual
variables

(Mandler, 1980)
. In three experiments using pictorial stimuli, Rajaram
showed that remember responses could be altered when perceptual factors were
manipulated. Firstly, it was found that more R
responses were produced when a
picture was presented at study and test, than when a word was presented at study and
a picture at test. Secondly, using line drawings, he showed that the proportion of R
responses was negatively affected by changes in the siz
e of stimuli. Thirdly, it was
discovered that mirroring images between study and test (as opposed to leaving the
picture at the same orientation) significantly reduced the proportion of R responses.
He does suggest that the pictorial materials used do lend

themselves towards
perceptual processing, perhaps more than written or verbal materials would (1996).

Additionally, it is noted that the instructions given to the
subjects included

several
examples of memories consistent with R memories: “…such as aspects

of the physical
appearance of the words (or pictures), a particular association or image, something
about its appearance or position, and so forth,” (1996, p373), but the author does point
out that examples of R memories involving conceptual aspects such
as associations,
wer
e also given.

Rajaram asserts that the previous findings regarding differential retrieval of
remember/know memories (e.g., that divided attention at study reduces R responses)
can be reinterpreted not as conceptual versus perceptual eff
ects but in terms of
distinctiveness versus fluency processing; that in fact they exhibit the effects of
Exam number 4193800


13

increased (or decreased) opportunity to encode distinctive or salient aspects of stimuli
(1996). The present study, by allowing subjects a comparativel
y long time to encode
the stimuli (five seconds for each face), should therefore allow more opportunity for
distinctive processing and result in a high percentage of remember responses.
Unfortunately since the methodology has not been repeated exactly else
where
(using
similar stimuli and presentation rates, for example)
t
his prediction must remain
hypothetical.


Criticisms and queries concerning the remember/know paradigm
.
A criticism
of the remember/know paradigm is that dividing results in terms of rememb
er and
know responses is based on an assump
tion that the different types of recognition are
independent and mutually exclusive (Rajaram & Roediger, 1997).
However, some
contend that this differentiation between remember and know memories does not
necessari
ly entail exclusivity, as Gardiner, Java and Richardson
-
Klavehn explain: “…
at the level of the responses the difference between exclusivity and inclusivity is only
a matter of definition,” (1996, p114). Remember and know responses may be treated
independe
ntly becau
se, according to Tulving’s (
1985) theory, either recollection
occurs along with a feeling of familiarity, or the feeling of familiarity is experienced
alone.
That is to say that, although at the theoretical level the remember/know
relationship is

one of redundancy, knowing is behaviourally

(and purportedly
phenomenologically)

different to remembering (Gardiner et al., 1996
).

Another criticism of the paradigm was that measuring memory using
remember/know responses might only measure the subjects’ l
evel of certainty

(Rajaram, 1996)
.

This criticism holds that remember responses simply signify greater
confidence in memory retrieval, and that know responses signify lesser confidence.

Rajaram and Roediger cite
findings from
four studies that show that pa
tterns of
certainty judgements differed significantly from the patterns of remember/know
responses observed (1997). Whilst this does not rule out the

likelihood

that level of
certainty is somehow related to the type of recognition experienced


certainly t
hey
are correlated (Rajaram, 1996),
-

it does at least provide some reassurance that the
these measures of mem
ory retrieval are not identical, and that the remember/know
paradigm can be used to find out useful information about the nature of memory
retriev
al.

Exam number 4193800


14

Discussion continues over the relationship between remembering and
knowing, with particular focus on whether the two are independent or redundant. A
statement that the two are independent

implies that a subject can
be said to
either

remember or know a
n

item
. On the other hand, a statement of redundancy refers to the
position that accedes that a subject may both know and remember an item
simultaneously (Rajaram & Roediger, 1997). If remembering and knowing are
redundant, this would help to account for t
he large percentage of ‘guess’ responses
obtained in the present study, as it would be more likely that subjects were not sure
whether remembering or knowing was the dominant experience.


Aim One: Summary

and methodological considerations
.
One of the main

aims
of the present study was to replicate the findings of Mäntylä and Holm, 2006, under
altered conditions. Mäntylä and Holm (and also Mäntylä, 1997) found that specific
types of judgements made about faces at encoding resulted in different proportions o
f
R/K responses. Regardless of how distinctive the faces were individually, it was
discovered that rating faces for distinctiveness at study elicited more Remember
responses at test. Conversely, when faces were sorted into one of four categories
(‘sporty
-
t
ype’, ‘intelligent’, ‘home
-
body’, ‘party
-
goer) at study, a higher proportion of
Know responses were obtained. The authors’ suggestions regarding the relative
differences between the two types of judgement and the effect they will have on
memory are based o
n the distinctiveness/fluency processing distinction described by
Rajaram and Roediger (1997) and Mäntylä (1997). Mäntylä and Holm attribute the
higher proportion of Remember memories following ‘rating for distinctiveness’ to the
demand placed on the subje
ct to look for distinctive features in the faces they looked
at. The other encoding condition, ‘sorting faces into categories’ was understood to
encourage subjects to look instead for similarities between the faces


in other words,
fluency of processing.

The present study

used a similar procedure to that of
Mäntylä and Holm
(
2006
)

when describing to participants how to respond to the question of the type of
recognition they had experienced. Brief and clear explanations of the distinction
between ‘remember
ing’ and ‘knowing’ were given along with examples of each, and
it was established that the participant had experienced both of these types of
recognition
;

if
participants

expressed doubt or seemed not to understand, the
distinction was re
-
iterated and disc
ussed with the
m

until they explicitly expressed
Exam number 4193800


15

comprehension
.

In the Mäntylä & Holm (2006) study, the authors decided not to force
the subjects to choose either remember or know, but also allowed subjects to respond
that they ‘guessed’ if they were not su
re which type of recognition they had
experienced.

The intention here was to reduce participants’ reliance on ‘know’
responses when they were not sure if they ‘remembered’ (2006). Participants in the
present study were similarly advised that, were they uns
ure as to whether they had
experienced ‘remembering’ or ‘knowing’, they should respond that they had
‘guessed’.

This usage of the ‘guess’ option is distinctly different to the use in other
studies such as Gardiner et al., 1996. Gardiner and others instruct
ed subjects to use
the ‘guess’ option when they felt that their judgment of recognition was not based on
‘remember’ or ‘know’ memories


that is, that their judgement of recognition itself
was a guess (1996). This methodological difference is substantial a
nd its effects are
not fully understood. Unlike in the present study and the

Mäntylä & Holm (2006)
study, the use of ‘guess’

by Gardiner et al., (1993)

would appear to lead to a possible
dependency between remember and know responses
.

On the other hand, an

advantage
of the use of this response option by Gardiner and others is that it reduces the
dependency between recognise/don’t recognise judgements (which was unavailable to
subjects in the current study).

As such, the any comparison between these studies
should be undertaken with caution, and this area may warrant specific investigation.


Main Aim Two: The Mid
-
Profile A
dvantage

The present study investigated two aspects of the effect (if any) of viewpoint
on recognition of unfamiliar faces. The first iss
ue being addressed was that of the so
-
called ‘same
-
view’ advantage of the three
-
quarter view over other views of a face.
Proponents of the same view advantage believe that, when faces are viewed at the
same angle during study and test, the mid
-
profile angl
e (usually as opposed to the
full
-
frontal or profile view) is associated with the best chance of recognition. This
effect has not been consistently supported by the evidence (Liu & Chaudhuri, 2002),
which will be discussed in greater detail later in the te
xt. The second issue to be
addressed was the ‘different
-
view’ advantage of the face at mid
-
profile. This
advantage is held to manifest itself in higher probability of recognition of a face
presented at a different angle when it was originally studied at a
mid
-
profile view. Its
supporters claim that the mid
-
profile perspective ‘generalises better’ to a different
Exam number 4193800


16

view (Hancock, Bruce & Burton, 2000; Liu & Chaudhuri, 2002). For example, Hill,
Schyns and Akamatsu (1997), found that, when the face was mirrored o
r ‘flipped’
between study and test, faces at mid
-
profile were recognised better than those at
profile view. This theory has also been inconsistently supported. Liu and Chaudhuri
contend that, of the articles they cite that reported some form of mid
-
profile
, different
-

angle advantage, several may in fact have only been showing an effect of angular
rotation. That is, that the increased hit rates for mid
-
profile view faces were due to its
intermediary position, 45 degrees from both full
-
frontal and profile vi
ews, whereas
the reduced hit rate of full
-
frontal to profile views (and vice versa) was due to the
larger degree of rotation involved. They report, however, that the limited amount of
relevant data available in the articles cited means that this possibilit
y could be nei
ther
confirmed nor denied. Another study

found the converse effect: a mid
-
profile view
advantage at

test
.
Bruce, Valentine and Baddeley (1987) used faces in frontal, mid
-
profile and profile views and conducted their experiment as a series of
individual
trials, where study and test views were presented successively. They discovered a
mid
-
profile test
-
view advantage.

These findings will first be discussed in light of the
different variables (such as type of stimuli) that affect them.


Factors th
at influence/negate the viewpoint effect
.
The conflicting findings in
the paragraph above show how transient the mid
-
profile advantage is. Several
different factors have been found to affect the manifestation of
viewpoint effects
.
Viewpoint has been found
to interact with distinctiveness of faces (as rated by people)
in interesting ways

(Newell, Chiroro & Valentine, 1999); i
n general terms,
recognition of objects across viewpoints is affected by how similar they are to one
another, so that when objects are
very similar, viewpoint becomes more important to
accurate recognition. Typical faces are more easily and quickly identified as
belonging to the class of faces, but distinctive faces are more likely to be accurately
discrimi
nated from other faces (Newell e
t al., 1999). Newell and colleagues
(1999)
describe a model known as the ‘face
-
space’ model that has been created to account for
these findings. The face
-
space is proposed to be an internal representation of the
entire spectrum of face shapes, with more ty
pical faces densely populating the area
near the centre (since there are more typical than unusual faces). Distinctive face
shapes therefore occupy the sparser outer regions. Each face can be mapped on to the
face
-
space such that it can be coded and retrie
ved by the co
-
ordinates it occupies. If a
Exam number 4193800


17

face is more typical, it is identified as a face more quickly (according to the model)
because it has such a central (and immediately accessible) position. If a face is
distinctive and is represented in the less po
pulous outer regions of face space, it is not
so quickly identified as a face as such, but is more easily discriminated from other
faces (Newell et al., 1999).
Based on their findings in this area, O’Toole and
colleagues (1998) theorise that there is a tra
de
-
off that takes place between two
different factors in different
-
angle face recognition: typicality and discriminability.
O’Toole et al say that, although more typical faces are easier to recognise across
different view
-
points, this is ameliorated by the

fact that more typical faces are harder
to discriminate between

(1998)
. Distinctive faces will not pose this problem, although
they may prove harder to recognise depending on the nature of the distinctiveness.

More generally, numerous studies of face rec
ognition have revealed the
significant effect that the viewing environment has on likelihood of recognition. For
instance, it has been found that changes in lighting conditions (such as direction of
light source) and contrast patterns (such as negative ima
ge) have strong effects on
subjects’ ability to recognise faces (Bruce, 2009).

Bruce argues from these effects and
others that structural encoding of faces is at least partially dependent on context
(2009).

Furthermore, it has been reliably found that inve
rsion of faces results in
markedly poorer recognition rates, in addition to affecting judgements of age,
attractiveness, and mood (Bruce, 2009). Of particular relevance to the question of
view
-
specific advantage is the finding that inversion interacts with

change in angle:
inverted faces are much harder to recognise when they undergo a change in angle,
than when upright faces undergo the same degree of angle change (Hancock, Bruce &
Burton, 2000). Results from studies of the inversion effect have been taken

to suggest
that inversion upsets the configural processing of faces (Hancock et al., 2000), rather
than the componential processing.
The terms configural and componential are used
here to refer to two related aspects of the processing of the physical stru
cture
of faces;
in the current text,

‘configural’ processing
refers to the encoding of the relationships
between different features of the face, such as the distance between the eyes.
‘Componential’ processing is used to refer to encoding of the individual

components
of the face, such as the nose. The definition of the two types of processing is not well
defined in the literature

(Mäntylä & Holm, 2006)
, and this issue is covered later in the
text.

Exam number 4193800


18

An overview of the literature seems to show that effects of
pose change
disappear in more naturalistic settings. Logie, Baddeley and Woodhead (1987) set up
an experiment to test the ecological validity of the pose change effect found
previously, where volunteers were asked to try to identify a person in a busy town

centre from memory of a photograph studied earlier


this method led to poor
recognition rates, a high false alarm rate and no effect of pose. An effect of pose was
found only under highly constrained conditions (as in the present study), where, out of
th
ree poses


frontal, profile and mid
-
profile,
-

the mid
-
profile face was the “most
helpful,” (1987, abstract). Effects of change in angle can also be altered by the
presence or absence of texture (Troje & Bülthoff, 1996). In their 1996 study, Troje
and Bül
thoff
compared recognition of textured versus untextured faces, using
shading
alone to render face shape in the untextured faces. In addition to finding an advantage
for mid
-
profile faces with both types of stimuli, Troje and Bülthoff report that the
range

of the mid
-
profile angle change in which facial recognition was facilitated was
wider


20 to 70 degrees


for untextured than it was for textured faces, where the
advantage was reduced to a 25 to 40 degree range (1996).
These latter findings imply
that,
in the more naturalistic condition using textured faces, subjects’ ability to
recognise
faces across changes in angle was

reduced.

This is consistent with the
findings of Logie et al., (1987), and it can be inferred from the combined findings that
the mid
-
profile advantage, if it exists at all, is to be found only in highly controlled
circumstances.

There are several lines of evidence that suggest


if not a mid
-
profile
advantage


but certainly that different views of a face elicit different levels of
reco
gnition. Neurophysiological evidence from single unit recording studies on
primates is consistent with view
-
dependent processing: certain cells were found to
respond selectively to perspective view and others to perspective view and identity
(Newell, Chir
oro & Valentine, 1999). O’Toole, Edelman & Bülthoff found that
recognition of faces at test depended on the angle at study (1998), and further, that
faces studied at mid
-
profile were the easiest to recognise when tested at the same
angle, although Newell e
t al. were not able to replicate this effect (1999). Bindemann,
Scheepers and Burton (2009) found that eye movements to faces show qualitatively
different patterns depending on the angle of view. Krouse (1981) contends that, of the
21 facial features liste
d by Goldstein, Harman and Lesk (1971) as the most useful in
discriminating faces, all are visible in the mid
-
profile pose


unlike other viewpoints.
Exam number 4193800


19

Hancock et al. agree with this to a small degree, saying that recognition of faces at
profile could be poo
r because, “it obscures the configural information that underlies
normal face processing” (2000, p331). Additionally, Hancock and colleagues note
that recognition rates appear to be higher for same
-
angle faces set at 20, 30, or 45
degrees (2000).

Lastly,

Newell et al speculate that, “the three
-
quarter view and the
profile view may be more similar in psychological face space than the three
-
quarter
and the full
-
face view,” (1999, p526)
, based on their finding that responses to full
-
face views were significan
tly slower than for three
-
quarter and profile views.

The evidence and theories above seem to show that it is possible that the mid
-
profile view is the canonical perspective, although this is affected by many variables
,

both of the stimuli used and the test
ing environment.
Palmer and colleagues note that
canonical views should: allow most potential for ordinary interaction with the object;
be as alike as possible to different perspectives of the same object whilst being most
unlike views of other objects; an
d should provide the best access to internal
information about the object. In a series of experiments they showed that it was
possible to operationalise and test for manifestations of the ‘canonical’ view. The
canonical perspective was a perspective that w
as consistently found to provide the
best subjective ratings of goodness of view, to be the initial perspective people
imagined an object from, and to be identical with the perspective chosen for choosing
the ‘best’ photograph (Palmer et al., 1981). The pr
esent study did not utilise any of the
above methods for testing whether the mid
-
profile view is the canonical view, but
these considerations are helpful for trying to work out what or why the mid
-
profile
advantage might consist in. One suggestion might be

that the mid
-
profile view
contains the most configural information about the face


that is, that is allows
greatest access to the overall shape of the face and the relations of the different
features to one another.


Alternatively, i
t may be that the mid
-
profile view offers the best access to
distinctive information about a face.
O’Toole et al. (1998)
conclude

that qualitatively
different aspects of a face can be distinctive


for example, a face may have a
particularly distinctive nose (componential dist
inctiveness), or may have eyes that are
unusually far apart (configural distinctiveness). That is not to say that
configural/componential processing are distinct entities


it is far more plausible that
there is a degree of overlap between them. In fact, a
lthough there is general agreement
Exam number 4193800


20

in the literature about certain aspects of the nature of what is here described
configural/componential
processing,

different terms have been used to refer to them.
For instance, Diamond and Carey (1986) speak of ‘relati
onal’ versus ‘isolated’
features, while Bruce and Young (1986) refer to ‘internal’ and ‘external’ features, and
the terms ‘variant’ and ‘invariant’ are used by Haxby, Hoffman & Gobbini (2000).
There is yet to be put forward a definitive definition of these

types of processing that
includes the relationship of one to the other, although this may not occur until more
research has been done that can reveal more about their differential effects. A large
part of this research is likely to come from studies of th
e face inversion effect. For
example, in two recent experiments using functional magnetic resonance imaging
(fMRI), Goffaux, Rossion, Sorger, Schiltz and Goebel (2009) found evidence to
support the previous finding that face inversion mostly affects proces
sing of vertical
spatial relations. Horizontal relations (or facial symmetry) and componential aspects
were found to be only moderately affected by face inversion.

Eye
-
movement research has provided some intriguing suggestions as to the
nature of facial p
rocessing and the different ways in which a face can be processed.
Bindemann, Scheepers and Burton (2009) recently found that observers’ eye
movements differ qualitatively according to the angle a face is presented at (2009).
Their presentation of faces at

three different views


frontal, mid
-
profile and profile


allowed them to find that facial processing is subject to a ‘centre of gravity effect’,
whereby the eye is initially drawn to the centre of the picture before moving to other
areas. Previous stud
ies that had used only frontally viewed faces failed to find this
effect since it coincided with a well
-
documented tendency of viewers to focus on the
(centrally located) eyes and nose of a face (Bindemann et al., 2009; Mäntylä & Holm,
2006). In viewing m
id
-
profile and profile faces, the authors found that observers
tended to look most at the innermost (or only visible) eye. For profile faces observers
took longer to fixate upon the eye (2009), which may be explained by the ‘centre of
gravity effect. This
latter finding appears to account for the relatively poorer
recognition of faces studied at profile when compared to the other two views
(O’Toole, Edelman & Bülthoff, 1998). These eye
-
movement
-
related findings together
indicate that processing of unfamilia
r faces is affected by the viewing angle.

Mäntylä and Holm also looked at the relation of eye movements to recognition
and its component measures of recollection and familiarity (2006). In their first
experiment, they investigated the effect of preclusion
of eye
-
movements during
Exam number 4193800


21

study/test on later ability to recall/recognise faces. They found that ‘remember’
responses (experiences of recollection) were significantly reduced when eye
movements to faces were not permitted during study and/or test, but ‘know’

and
‘guess’ responses were not affected (Experiment 1). In the second experiment, the
authors compared eye movements at study and test to see if differences were apparent
in the pattern of eye movements when they were grouped by ‘remember’ or ‘know’
respo
nse. They argued from the transfer appropriate processing account (Roediger,
Weldon & Challis, 1989), and Tulving’s encoding specificity principle (1983), that it
would be reasonable to expect that an experience of recollection would be
accompanied by grea
ter similarity in eye movements between study and test than
would an experience of a feeling of familiarity. The authors predicted that recollection
would be associated with a higher frequency of refixations on parts of the face that
were fixated at study;

that is, that at test the subject would fixate more often on (or
within two degrees of) areas that had been fixated at study, when an experience of
recollection was indicated (2006).

It was felt that decomposing recognition performance into measures of
re
collection and familiarity might help to reveal the nature of the mid
-
profile
advantage, if one exists, and perhaps also prove instructive as to the reason(s) why it
has not been consistently found. For example, might it be that equal recognition
performan
ces could obscure a difference in the
type

of recognition elicited by mid
-
profile and other views? It is possible

that the different views elicit

differences in the
componential or configural processing of a face. That is to say, the different
perspectives

afforded might affect how the observer perceives the componential
aspects of a face, such as the shape of the nose; on the other hand, the different views
might impact differentially on the holistic or configural processing of the face, such
that in one v
iew the features might seem more squashed together than in another
view.


To sum up
, i
f recognition depends on view at encoding

(O’Toole et al., 1998;
Hancock et al., 2000)
, and
processing of faces

is

categorically different depending on
view

(Bindemann e
t al., 2009)
, this suggests that research using

different facial
perspectives c
ould be informative to understanding of faci
al encoding/retrieval
processes
.
Additionally, because other recent research has shown that qualitative
differences imposed at encodi
ng result in differential effects on remember/know
responses

(Mäntylä & Holm, 2006), it was felt that dividing subjects’ responses using
Exam number 4193800


22

the remember/know paradigm might uncover underlying differences in the way mid
-
profile versus frontal views are process
ed.


Overall Summary


In summary, the main aims of the present study were to replicate the findings
of Mäntylä and Holm (2006; experiment 2), and to find support for and explore the
nature of the mid
-
profile advantage in face recognition.
The following are

the formal
hypotheses of the present study:
Hypothesis o
ne had two parts: it predicted that rating
faces for distinctiveness would result in a higher percentage of ‘remember’ responses,
and also that sorting faces into categories would result in more ‘kno
w’ responses.

Hypothesis t
wo also had two parts: it predicted that, when faces were viewed at the
same angles at study and test, mid
-
profile faces would be associated with a higher
percentage of hits than frontal faces; its second prediction was that, when

faces were
viewed at different angles between study and test, faces studied at mid
-
profile would
be associated with a higher percentage of hits than those studied at frontal view.


Due to the complex design the results will be divided into three main sect
ions:
an overall summary of the results including hits and false alarms; a section with the
descriptive statistics, statistical analyses and brief discussion pertaining to the first
hypothesis; and a third section that presents the descriptive statistics a
nd statistical
analyses along with brief discussion of these as they apply to the second hypothesis.















Exam number 4193800


23










Method


Subjects and Stimuli

Twenty
-
four University of Edinburgh students took part in the study. Their
ages ranged from 20 to 33 ye
ars with a mean of 27 years 6 months.

Five of the
subjects were male. A
ll

subjects reported having

normal or corrected
-
to
-
normal
vision. Subjects were recruited via e
-
mails sent to student e
-
mail lists (i.e., self
-
selected) and paid £5 for their participat
ion.

The face images used in this work have been provided by the Computer
Vision Laboratory, University of Ljubljana, Slovenia. The 142 photographs used
were of young men all said to be aged approximately 18 years. Practi
c
e images
consisted of a selecti
on of 14 images of 8 different individuals. 128 photos of 64
individuals were used in the experiment proper
,
with

two photos of each individual
-

one at frontal and one at mid
-
profile angle. The photographs were manipulated using

editing tool Adobe

Photoshop to remove head hair, ears and background, leaving only
the
face centrally positioned
on a white background. Ears were removed because in
many photos these were obscured by hair and so were not consistently apparent.

Indivi
duals

wearing glasses

and/or more than a light growth of facial hair were
similarly excluded since these were not regular features of all faces, and it was
intended that faces be distinguishable (as much as possible) by face shape and
features alone
. Counterbalancing was implemented
to ensure that each face was
shown in each condition: old and new, frontal and mid
-
profile, rated for
distinctiveness and sorted into categories.
Subjects were tested individually.

Subjects were seated in front of the computer so that images appeared at eye
-
level, with a viewing distance of approximately 55cm from a 17 inch screen. Images
Exam number 4193800


24

had a resolution of 71dpi and b
it depth of 24. Images varied slightly in size, at 128
-
130 pixels in width and 155
-

170 pixels in height; this equates to a height/width ratio
of around 1.2 to 1.3. Mid
-
profile images were all presented facing to the subject’s
right. It should be noted t
hat images were not all exactly frontal or mid
-
profile (i.e.,
45° from frontal) but sometimes varied a few degrees to the left or right of these
positions.




Design and Procedure

The study f
ollowed a repeated measures

design with two levels of judgement
t
ype (rating for distinctiveness vs. sorting into categories,) as a within
-
subjects factor,
and two levels of viewpoint type (full face vs. mid
-
profile) as a second within
-
subjects factor. Dependent variables were:
percentage of
hits, correct rej
ections, false
alarms and misses. Due to a small number of test trial responses that had to be
excluded, data could not be compared directly but were converted to percentages.

After reading the initial information sheet summarising the procedure (see
Appen
dix A), subjects were informed that they would complete practise sessions for
each of the experimental phases before the experiment proper. They were also told
that after the two study sections they would be tested on their ability to recognise the
faces w
hen they were mixed in with new faces; learning was therefore not incidental,
unlike in the Mäntylä and Holm

study (2006). The study phase consisted of two
sections where each contained 16 randomly ordered stimuli for appraisal (either rating
for distinc
tiveness or sorting into categories). The test phase consisted of the 32
studied items mixed in with 32 distractor items, also randomly ordered.

Prior to the practise session for ‘sorting faces into categories’, it was
emphasised that subjects should try

to ensure that the faces within each category were
as similar to each other as possible; the purpose of this was to encourage them to
search for similarities (i.e., think about whether the face was consistent with a certain
‘type’ such as ‘intelligent
-
loo
king people’). It was also made clear that responses in
the study section were not timed and that subjects should take as long as was needed
to come to a decision. Before
subjects began to practise

rating
faces for
distinctiveness, the followin
g clarification of the use of the term ‘distinctiveness’ was
provided: subjects were told that, for each face they saw, they should imagine that
Exam number 4193800


25

they had to meet that person at a busy railway station and consider how easy it would
be to pick the face out i
n the crowd. If the face was easy to pick out then it was very
distinctive, and if it was hard to pick out then it was not distinctive. Subjects were
informed that these two sections might run in a different order in the experiment
proper and so they shoul
d pay attention to the type of judgement they were being
asked to make at the beginning of each section.

Subjects used a button box for all
responses during the study and test phases.

Numbering of responses (such as the four
rating levels used in the judge
ment condition, rating for distinctiveness) was
counterbalanced across subjects
to avoid

any effects of button position.

Following completion of the
practice
sessions for the study phase subjects
were introduced to the test phase procedure. They w
ere advised that, of the faces that
they had studied earlier some would be at a different angle whereas some would be at
the same angle as before. Subjects indicated whether or not they recognised the face
by responding th
at it was ‘old’ or ‘new’ using the

button box. Participants were
informed that the face would remain on screen until they responded, and that a ‘new’
response would result in a new face being presented, whereas an ‘old’ response would
result in the appearance of a screen asking about the t
ype of recognition s/he had
experienced.
There were three options

presented onscreen
:

subjects could
press
buttons

to indicate either

‘remember’, ‘know’
or

‘guess’

judgements of
recognit
i
on

type
.

Participants were educated about the distinction betwee
n ‘remembering’ and
‘knowing’ as types of recognition using examples (see Appendix B) and the subject
was required to indicate their understanding before they could continue.

Individual stimulus presentation in the study phase comprised: centrally
position
ed fixation circle 500 milliseconds (msec) in duration, followed by centrally
positioned stimulus (face) 5000msec in duration. Stimuli were followed by the
judgement
-
appropriate ‘rating screen’ that gave the response options; this remained
onscreen until t
he subject responded

by pressing a button on the button box to indicate
their judgment
.

In the test phase the fixation circle was followed by the test image which
remained onscreen until the subject responded. Subjects were asked to respond as
soon as th
ey had decided whether or not they recognised the face.
If subjects
responded that a face was new, the next stimulus would be presented. If subjects
respon
d
ed that a face was old, a screen was presented asking what type of recognition
had taken place and g
iving the three options (
‘remember’,


know’, ‘guess’
,
) the
Exam number 4193800


26

subject should decide between.

The testing phase immediately followed the two study
sections.








Results


Overall Summary



Hits

F
alse
alarms

Mean percent of
total responses

64.3%
(7.35)

34.4%
(18.98)

Table 1. Standard deviations in brackets.


It can be seen in Table 1 that the mean hit rate (i.e., the rate of correct
identifications of studied faces) is at just over 64% with a sta
ndard deviation of 7.35.
Respondents falsely identified new items as ‘old’ approximately one third of the time
on average. Hits and false alarms were reduced to their component
remember/know/guess responses (see Table 2 below).



Hits

False alarms

Remem
ber

Know

Guess

Remember

Know

Guess

Mean

41%

39%

20%

20%

37%

43%

Std. dev.

21.11

15.01

16.567

22.24

21.475

27.366

Table 2.

Hits and false alarms broken down into r/k/g as a percentage of overall score
available.


Table 2 shows a clear difference in the d
istribution of remember/know/guess
responses according to subjects’ accuracy. ‘Remember’ responses were made half as
often in false alarms, whereas ‘guess’ responses were made twice as often. In the false
alarm data, ‘G’ responses were made more often than

‘R’ or ‘K’ responses. These
differences between accurate and false alarm responses are consistent with the view
Exam number 4193800


27

that subjects were using the ‘remember’, ‘know’ and ‘guess’ options as instructed and
not simply pressing buttons at random.



Hypothesis One



Descriptive Statistics and Analysis




The average hit rates of faces studied whilst rating for distinctiveness versus
sorting into categories were 61% and 65% respectively. Responses in each condition
were decomposed into ‘remember’, ‘know’ and ‘guess’

responses (see Figure 1
below). For simplicity, the term ‘Distinctive’ will be used to refer to the study
condition, ‘rating for distinctiveness’, and ‘Similar’ will be used to refer to the study
condition, ‘sorting into categories’.


Graph 1: R/K/G Accuracy for Distinctive and
Similar Study Conditions
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
Remember
Know
Guess
Study condition grouped by R/K/G
Mean percent accurate
Distinctive
Similar

Fig. 1. Percentag
es of hits grouped by study condition (distinctive/similar) and
response type (r/k/g).


Numerically, the ‘remember’ and ‘know’ scores are consistent with the first
hypothesis


that faces encoded in the ‘Distinctive’ condition elicit more ‘R’
responses and

faces studied in the ‘Similar’ condition elicit more ‘K’ responses.
However, the difference in ‘know’ responses does not seem very large.

A two
-
way repeated measures ANOVA, with judgement type
(‘Similar’/’Distinctive’) and response type (remember/know/gu
ess) as factors was
performed to ascertain the presence of any main effects or interactions. A main effect
of response type was found, F(2,46)=7; p<0.005, indicating that remember responses
were given more often than know, and know more than guess. No inte
raction was
found between response type and judgement type (p=0.288). Thus, although the
Exam number 4193800


28

expected numerical trend was found, statistical differences were not present to support
this, and the null hypothesis had to be accepted.


Hypothesis One
-

Discussion

In the false alarm data, more know than remember responses were obtained, in
conformance with previous findings (Gardiner et al., 1996). There is a higher
percentage of guess responses than was seen in Mäntylä & Holm, 2006 (theirs was
less than 3%)
-

this

may be attributable to change in angle of face, the homogeneity of
the stimuli and possibly other factors, which will be discussed later in the text.

The lack of statistical significance for the difference between remember/know
responses may be explained
by the lack of contextual detail provided by the stimuli.
Mäntylä and Holm (2006) used much richer stimuli, using both males and females,
leaving in head hair, glasses, make
-
up and facial hair. This added detail may have
provided a stronger basis for a dif
ferential effect of judgement type. For example, the
increased level of ‘R’ responses following ‘rating for distinctiveness’ might be
facilitated by the presence and combination of hair colour, facial hair and glasses


all
of which are known to improve re
cognition rates (Bruce, 2009). This supports
Mäntylä’s (1997) argument that richer pictorial stimuli enable differences to be
brought out between ‘remember’ and ‘know’ type memories.

In the correct response data, the large percentage of guess responses in
dicates
that subjects were often uncertain as to whether they had experienced ‘remembering’
or ‘knowing’


or that they had misunderstood the meaning of ‘guess’, thinking that it
indicated lack of certainty about their old/new judgement. It is possible tha
t better
education about the distinction between remembering and knowing, and specific
questioning to test participants’ understanding would have resulted in more
remember/know responses. In addition, it seems very likely that the change in angle
made it d
ifficult for subjects to be sure whether they ‘remembered’ or ‘knew’ the
face.

It may also be that the large number and homogeneity of faces meant that
subjects struggled to be certain as to whether a certain test face was a face they saw
earlier, or was
merely strongly reminiscent of another, very similar face. A remember
response may be elicited by a face that has a similar nose, for example


and whilst
one may certainly be said to be ‘remembering’ a particular face, the stimulus that
elicited the memor
y may be of another, similar face. Furthermore, the nature of the
Exam number 4193800


29

stimuli may be another cause of the large percentage of ‘G’ responses, as the lack of
contextual details such as hair and glasses were not present to provide cues that might
have elicited a
strong episodic recollection.


Hypothesis Two


Descriptive Statistics and Analysis



Same angle hits

Different angle hits

Mean percent accurate

81.4%

42.73%

Standard deviation

13.04

19.17

Table 3
.


Firstly, as expected, items seen at the same angle at
study and test were
consistently better recognised than items studied and tested at different angles (Table
3), in addition to exhibiting lower standard deviations. Breakdown of the ‘same angle’
scores into frontal and mid
-
profile (see Table 4. and Fig. 2
below) shows that mid
-
profile faces were better recognised. Of the ‘different angle’ data, it appears that faces
studied at frontal angle are more easily recognised when tested at a different angle.
The higher standard deviations seen for different
-
angle h
it rates indicate that subjects
varied more in their ability to recognise faces than they did for same
-
angle faces
(table 4.).



Same angle

Different angle (angle at study)

Frontal

Mid
-
profile

Frontal

Mid
-
profile

Mean percent accurate

78.5%

85.19%

51.0
4%

39.14%

Standard deviation

18.28

12.73

21.69

23.16

Table 4.


Exam number 4193800


30

Face Recognition Performance
0
20
40
60
80
100
Same-angle
Different-angle
View (frontal/mid-profile) as a function of view
change (same/different)
Mean percent accurate
Frontal
Mid-profile


Fig. 2. Percentage of hits grouped by angle at study (frontal/mid
-
profile)
and view change (same
-
angle/different
-
angle).


A two
-
way repeated measures ANOVA with view change (same or differ
ent)
and angle at study (frontal or mid
-
profile) as factors revealed one main effect and one
interaction: an effect of view change, F(1,23)=84, p<0.005, and an interaction
between view change and angle at study, F(1,23)=5.86, p<0.05. Post
-
hoc tests
confir
med the same
-
angle advantage to be significant, t(23)=9; p<0.001. Further tests
showed that the difference between same
-
angle frontal and same
-
angle mid
-
profile
performance was not quite statistically significant, t(23)=
-
1.817; p=0.08. However, a
significa
nt difference was found between performance levels at different
-
view study
angles t(23)=
-
2.27; p<0.05, where studying faces at frontal view and testing at mid
-
profile view led to a significantly higher hit rate than vice versa.


Exam number 4193800


31

Face Recognition Performance
0
10
20
30
40
50
60
Same -
Frontal
Same - Mid-
profile
Different -
Frontal
Different -
Mid-profile
R/K/G grouped by view change and angle
Mean percentage (%)
Remember
Know
Guess

Fig. 3. Percentage of c
orrect responses grouped by view change
(same/different) and response type (r/k/g).


Fig. 3 above shows the same
-

and different
-
angle scores for both views broken
down into r/k/g. ‘Remember’ responses are made twice as frequently to same angle
items


at
means of 45% and 52% compared to different angle ‘remember’ scores of
24% and 22%. On the other hand, ‘know’ responses are made 10
-
15% more often to
different angle items. Additionally, ‘guess’ responses are made roughly twice as often
to a different
-
angle

face.

A three by two by two repeated measures ANOVA was used to look for main
effects, with response type (r/k/g), view change (same/different) and angle at study
(frontal/mid
-
profile) as factors. One main effect and one interaction effect were
found. S
ince the data for these were non
-
spherical, results from the Greenhouse
-
Geisser estimator are reported. An effect of response type (remember/know/guess)
was found as expected, F(1.4,28.2)=7.622; p=0.005. The interaction was found
between response type and
view change (i.e., whether the angle was the same or
different at study and test); this was also highly significant, F(1.99,39.88)=12.76;
p<0.001.

By reducing the data further, it was established that, within the same
-
angle
data, the percentage of ‘remembe
r’ responses did not differ significantly between
frontal and mid
-
profile views (t=1.373, df=23, p=0.183). Similarly it was found that,
within the same condition, ‘guess’ responses did not differ significantly between the
two views (t=
-
1.283, df=23, p=0.21
2). Next, r/k/g response percentages were
Exam number 4193800


32

compared across the same
-
angle and different
-
angle conditions. First, ‘remember’
responses were compared


the difference was found to be highly significant
(t=5.402, df=47, p<0.001), such that remember responses w
ere given significantly
more often in the same
-
angle condition. ‘Know’ responses were not significantly
different (t=
-
1.749, df=47, p=0.087), and ‘guess’ responses approached statistically
significant difference (t=
-
1.931, df=47, p=0.06) but did not reach
it.


Hypothesis Two
-

Discussion

The difference in hit rates for same
-
angle when compared to different
-
angle is
consistent with previous findings (Krouse, 1981). In the same
-
angle condition,
subjects have the added advantage of pictorial cues not present w
hen the face is seen
at a different angle at test. When the same face is seen at a different angle, it can be
expected that subjects would be less likely to recollect it (fewer ‘r’ responses) or be
uncertain as to whether they do or do not recollect it (i.
e., respond ‘g’), since they are
unable to access many of the pictorial cues available when the same face is seen at the
same angle, although some or many of the structural cues may be present.

This is
supported by the additional finding that higher percen
tages of remember responses
were given in same
-
angle as opposed to different
-
angle conditions.

Interestingly, the test
-
view advantage found for the mid
-
profile face runs
counter to other findings in this area of face research. Many studies have reported a
learning
-
view

advantage for the mid
-
profile view, and it has been contended that the
mid
-
profile angle ‘generalises better’ to a different view, although this has not been
reliably found (Liu & Chaudhuri, 2002). According to this line of reasoning and the
results found here, it is in fact the frontal view that ‘generalises better’ to a different
angle, despite the apparent advantage the mid
-
profile angle has when faces are viewed
at the same angle during encoding and at test. The results may instead be inte
rpreted
as showing a test
-
view advantage for mid
-
profile faces.

General discussion



The work reported here examined the accuracy and nature of face recognition
under three sets of conditions,
put forward

in terms of two different hypotheses. The
data ana
lysed consisted of hit rate (percentage) and percentage of ‘R’ and ‘K’
responses. Hypothesis
one

pertained to the expected differential effect of judgment
type (rating for distinctiveness versus sorting into categories) on type of memory
Exam number 4193800


33

(R/K) retrieved. T
his differential effect was

found to exist numerically but was

not
strong enough to reach significance
, and the null hypothesis was accepted
. Hypothesis
t
wo, in two parts, described an expected mid
-
profile view advantage for same
-
angle
an
d different
-
angle
conditions. The previously found beneficial effect of same
-
angle
as opposed to different
-
angle presentation was replicated: not only was recognition
performance better to same
-
angle faces, but recollection (as measured by the
percentage of remember respons
es) was also improved. No

advantage was found for
the mid
-
profile perspective in the same
-
angle viewing condit
ion
. For the different
-
angle viewing condition,
unexpected results showed a m
id
-
profile test
-
view advantage
(
alternatively
conceptualised as a fro
ntal

study
-
view
)

advantage


this will also be
discussed in more detail. Some of these results confirm previous findings, whereas
others are in conflict with the literature.

The high percentage of guess responses obtained is interesting, and may be
explai
ned by a number of factors. Firstly, the results support the expectancy that faces
viewed at different angles at study and test would receive more guess responses than
same
-
angle faces overall. Although this difference was not statistically significant, it

came very close to significance (p=0.06). This finding indicates that subjects may
have had more difficulty in deciding whether a face was remembered or known. This
explanation also finds support in the related finding that faces were remembered more
ofte
n overall in same
-
angle condition than when faces were presented at different
angles between study and test (p<0.001). A second more worrying explanation is that
subjects misinterpreted the test phase instructions, and used ‘guess’ to indicate that
they we
re not sure if they had recognised the face at all, rather than to signify
uncertainty over the type of recognition experienced. The intended use of ‘guess’ was
clearly stated in the instructions: “If you are not sure whether you ‘Remember’ or
‘Know’ the f
ace, you will be able to respond with ‘Guess’.” However, it is impossible
to verify that this was how subjects actually used the term, since their understanding
was not explicitly tested (e.g., by asking subjects to repeat the instructions back in
their ow
n words). The use of the ‘Guess’ option in this study may indeed be
counterintuitive, and certainly in other studies it has been provided for subjects to
indicate lack of confidence in their judgement of recognition (Gardiner, Java &
Richardson
-
Klavehn, 19
96), rather than the type of recognition experienced.

In their study (and meta
-
analysis of studies) of the effect of levels of
processing on remember/know responses, Gardiner et al. investigated the relation of
Exam number 4193800


34

the ‘guess’ response to remember/know respons
es (1996). They state that subjects
were told to respond ‘guess’ if their judgement of recognition was not based on
remember or know memories, in order to test a theory regarding inconsistent results
from similar studies. From their meta
-
analysis, the auth
ors report finding differences
in the proportion of remember/know responses to
tasks requiring
deep versus shallow
processing. Although it was reliably found that deep processing resulted in higher
proportions of remember responses, some studies reported a

reverse effect on know
responses (Gardiner et al., 1996). Gardiner and colleagues suggest that the
inconsistent levels of know responses obtained can be attributed to the high
proportion of guess responses in some studies, where, following deep processing
,
subjects respond ‘know’ even though it would be more accurate to say that they had
experienced guessing (1996). That is to say, that the higher number of ‘know’
responses in fact obscured a certain number of guesses. Consequently, Gardiner et al.
ran a f
ew experiments investigating the levels
-
of
-
processing effect where ‘guess’ was
included as an option. They found that know responses remained constant between
the two processing conditions when guessing was allowed, consistent with their
theory (1996).

Wh
ether and how much Gardiner et al.’s finding can be applied to the present
study is difficult to say. The main reason for this is that ‘guess’ is used differently
between the studies. In the Mäntylä and Holm study on which the present
investigation is base
d, know responses were found to increase when subjects sorted
the faces into categories even though guessing was included as an option


but
Mäntylä and Holm’s hypotheses are not based on the level of processing effect
(2006). Rather, the eye
-
tracking stud
y hypothesised a differential effect of
emphasising processing
similarities

versus emphasising processing
differences
.
Regardless, the high proportion of guess responses is consistent with a redundancy
model of memory retrieval, because this result suggest
s that subjects were aware of
both know and remember memories.

Regarding the lack of a significant effect of judgement type (rating for
distinctiveness versus sorting into categories) on percentages of remember and know
responses: it may be that the method
ological differences between the present study
and the study on which it is based can explain the different findings. For one thing,
because the stimuli used in the present study were much more homogenous than in the

Mäntylä and Holm study (2006), this ma
y have reduced the potential for observation
Exam number 4193800


35

of a differential effect of judgement (i.e., rating for distinctiveness/ sorting into
categories) on R/K responses. These results support the notion that richer, more
naturalistic non
-
verbal stimuli can provide
more opportunity for investigating dual
-
process theories of memory (Mäntylä & Holm, 2006). Furthermore, there were a few
other methodological differences between the two studies. Unlike in the Mäntylä and
Holm study (2006), subjects knew before beginning t
hat they would be tested on their
ability to recognise the faces among distractor items


this may well have led to a
different kind of processing than that used if learning had been incidental. Another
point of departure was the number of the stimuli


su
bjects studied thirty
-
two faces in
the present study in

comparison to Mäntylä and Holm
s’ twenty
-
four, amounting to an
increase of one third. Furthermore, subjects completed the test phase immediately
following study, whereas in the Mäntylä and Holmes study

subjects returned a week
later. Since longer retention intervals between study and test have a negative effect on
remember responses (Mäntylä, 1997), it can be concluded that a one
-
week interval
between study and test in the present study would probably h
ave decreased the overall
percentage of remember responses


and perhaps revealed a significant difference
between the two study conditions, but not for the know responses.

Nevertheless, the findings support the theory that the mid
-
profile angle
provides t
he optimal viewpoint for
recognition of faces, albeit at test rather than at
study
. If the mid
-
profile viewpoint is reliably found to benefit recognition, this has
further ramifications for forensic psychology. For instance, witnesses of crimes might
be mo
re likely to correctly identify a person when their photographs are presented at
mid
-
profile rather than the full
-
frontal viewpoint. Due to the un
-
naturalistic stimuli
used in this study it is, however, uncertain how generalisable the findings are to such
areas, although they do support claims for an advantage of the mid
-
profile view for
homogenous faces


which may apply to criminal
identification
cases where suspects
are very similar in appearance. It can be said with more certainty that the results
indic
ate that computer models


which currently are mostly programmed to operate
on frontal face views (O’Toole et al., 1998)


may well gain more power if they are
developed to process faces at mid
-
profile (and perhaps other views).

In the same
-
angle viewing c
onditions, the findings were comparable to those
of O’Toole, Edelman and Bülthoff (1998), in that mid
-
profile faces received higher
hit rates than full
-
frontal
, although not significantly so
. In the different
-
angle
condition, however, the two sets of resul
ts differ. In O’Toole et al.’s study, different
-
Exam number 4193800


36

angle hit rates for mid
-
profile and frontal faces are almost equivocal (1998), but in the
present study a significant test
-
view advantage was found for the mid
-
profile face.

To
re
-
capitulate: recognition perf
ormance was better when faces were studied at frontal
and tested at mid
-
profile, than when they were studied at mid
-
profile and tested at
frontal view.
This

different
-
angle
view
-
dependent
finding is in direct contrast to that
of Troje and Bülthoff (1996),
who discovered a mid
-
profile advantage for the
learning

view only. One possible reason for the inconsistent findings is the presence
of shadowing in the stimuli used

in the current study
. The stimuli often had
shadowing on the (subject’s) right side. Altho
ugh shadowing may be useful on the
grounds that it can provide more information abou
t the facial structure (Troje &

Bülthoff, 1996),
shadows may also obscure or exaggerate the perceived shape of
facial features. Humans have been shown to be influenced by c
hanges in the level and
direction of illumination (Bruce, 2009), and although the source of illumination was
held constant (except for a few stimuli), the change in angle
between frontal and mid
-
profile
would nevertheless affect the degree of shadowing pre
sent. This may help to
account for the inconsistency of the current findings with those of other studies.

From a theoretical perspective, the findings in the different
-
angle condition are
hard to reconcile with previous ones. Troje and Bülthoff (1996) asse
rt unequivocally
from their results that only the view at study affects hit rate, but if this is the case then
the current findings require a complex explanation. If it is true that the study view is
important, then further information is required to accou
nt for a study advantage of
mid
-
profile views
in some st
udies (Liu & Chaudhuri, 2002)
, and on the other hand an
advantage for fr
ontal views in the present study
. Perhaps the only other study to find a
test
-
view advantage for the mid
-
profile perspective is
one by Bruce, Valentine &
Baddeley (1987). Bruce et al. report that, out of the three different views used, profile
views showed a consistent detrimental effect on recognition, whereas mid
-
profile
views either produced equivalent performance to frontal vie
ws or were particularly
conducive to recognition (1987). In their discussion of Bruce et al.’s findings, Troje
and Bülthoff dismiss the mid
-
profile advan
tage as an arti
fact of degree change; whilst
the view
-
change from frontal was to profile and vice versa
, the view
-
change from
mid
-
profile was to symmetrical mid
-
profile


and research shows that generalisation
from one view to the symmetrical view is reliably better than any other form of view
change (Troje & Bülthoff, 1996).
Furthermore, the profile view r
eliably elicits poor
recognition performance at study and/or test (Hancock et al., 2000).

Exam number 4193800


37

The
mid
-
profile view
may

to allow some kind of

selective

improvement in
processing of faces.
I
t may be that the advantage of
the
mid
-
profile view in the
different
-
ang
le condition arises from increased ability to process the configuration

or
some aspects of the components

in order to ‘match’ this to the frontal view seen at
study.

It is
equally
possible that, on the other hand, the frontal view facilitates a
different k
ind of processing
.

Further research might look at the effect of inverting
mid
-
profile, frontal and profile images to see whether differential effects of
configuration are elicited.

Can the transfer
-
appropriate processing principle

(TAP)

and/or encoding
sp
ecificity principle be evoked to explain the differential effect of

same
-

versus
different
-
angle conditions
? Rajaram (1996) points out that these accounts
(i.e., TAP
and encoding specificity)
predict a beneficial effect of
similarity between conditions
at
study and

conditions at

test.

Clearly the difference in performance between same
-
angle and different
-
angle conditions can serve as an example of the effect of increased
versus reduced opportunity for transfer
-
appropriate processing. However, neither TAP
no
r the encoding specificity principle seems to have much relevance in explaining the
findings
across the conditions
.

Since the stimuli were presented in random order
within the same study session, the environment did not differ for f
rontal versus mid
-
profil
e views
.
Subjects followed exactly the same procedure in both conditions.
However, the pictorial conditions did change, either from frontal view at study to
mid
-
profile at test or vice versa.
The process or processes that resulted in this
difference in
per
formance

would therefore appear to be unobservable behaviourally,
at least in this study.
It does not appear that consideration of dual
-
process models of
memory can account for the difference in reco
gnition performance, since dual
-
process
models

are based
on observances of remember versus know responses, and these did
not differ significantly within the different
-
angle data.

Consequently, it is concluded that the performance improvement associated
with the mid
-
profile advantage is
only partially
consistent
with characterisations of
this angle as canonical.
Future research must look deeper in order to find out what
cognitive processes differ from frontal to mid
-
profile (and other) views.



Some e
vidence of the tendency of observers to process faces differentl
y

depending on the view presented

comes from the Bindemann, Scheepers and Burton
(2009) paper mentioned
earlier
. Although subjects looked mostly at eyes, nose and
mouth (in that order and magnitude) in all viewing conditions, it was clear that
eye
Exam number 4193800


38

movement
s showed a different pattern between the different viewing conditions.

The
authors found that eye movements are freer to full
-
frontal th
an mid
-
profile or profile
faces; in the two latter conditions

observers tend
ed

to fixate almost solely on the most
visi
ble eye.

They go on to suggest that, regardless of angle of viewing, subjects’ eyes
are predominantly drawn to the visible features closest to the

centre of the face, and
venture to account for this by drawing attention to the relatively larger amount of
v
isual detail contained in this area of the face
.


This argument is consistent with that of Krouse (1991), who contends that, of
the 21 facial features suggested to comprise the most salient aspects for discriminating
faces, all are a
ccessible

in the mid
-
pr
ofile view


unlike frontal and profile views.
H
owever, Laughery et al., (1971)
-

who found no beneficial effect of the mid
-
profile
view

-

suggest that
,

although this view might allow better access to the various
features of a face, this does not necessari
ly have the consequence of improving
recognition. Laughery and colleagues suggest that subjects might well make use of
different aspects of a face depending on the perspective available, but may simply be
unable to encode the additional structural informat
ion due to processing limitations.

The findings of the present study appear to be consistent with this suggestion
. For this
account to have explanatory value, however, it must be defined what these
‘pro
cessing limitations’ consist in, whether they are quan
titative or qualitative, and
etcetera.

Althoff and Cohen report results from two eye
-
tracking experiments that they
interpret as showing a ‘re
-
processing’ effect in eye movements to famous faces
(1999), whereby memory of a face is associated with a specifi
c type of sampling
behaviour. Sampling patterns to famous and unfamiliar faces were significantly
different in several ways, and eye
-
movements to unfamiliar faces were ‘more
efficient’, showed more sampling of internal features, and was less symmetrical th
an
that of famous faces (Althoff & Cohen, 1999). Furthermore, they describe previous
work with two amnesic patients who could not state whether a face was famous or
not, but still exhibited differences in eye movements to the two types of faces; this
seems

to suggest not only that a procedural memory can be established for eye
movements to faces (Althoff & Cohen, 1999), but also that

eye
-
movement research
can
add to research into dual
-
process models of memory using

the

principle of
transfer
-
appropriate proc
essing
.

The growing literature using eye
-
tracking methods
appears to lend itself to the investigation of view
-
dependent processing. Measuring
Exam number 4193800


39

the pattern of eye movements to faces at encoding and retrieval could help to shed
more light on the way in which
differences in processing to different views of faces
affects the
recognition of faces across viewpoints
.

The literature certainly appears to agree that the eyes are the feature
predominantly focused on during face encoding and recognition (Bindemann et a
l.,
2009; Mäntylä & Holm, 2006; Althoff & Cohen, 1999). Several reasons have been
proposed to account for this. One intuitively appealing reason is the unique relevance
of the eyes in social cognition (Langton, Watt & Bruce, 2000). Evidence suggests that
t
he structural make
-
up of the eyes is such that we can infer important information
about eye gaze direction in another face, and that this may in fact be a specialised
neural process (Langton et al., 2000). Another reason for high numbers of fixations
on t
he eye area is the central location of the eyes in the face which, given the tendency
for observers to fixate initially on the centre of a visual stimulus (Bindemann et al.,
2009), seems fairly self
-
explanatory. A third suggestion holds that the relativel
y large
amount of detail contained in the area of the eye, when compared to more simple
features such as the nose and cheeks, demands a larger number of fixations in order to
be processed effectively (Bindemann et al., 2009). Althoff and Cohen refer to res
ults
of an eye
-
tracking study by Henderson, Weeks and Hollingworth (1999), that were
taken to suggest that semantic aspects become more influential to eye
-
movements
following initial superficially
-
related fixations (such as the centre
-
of
-
gravity effect
men
tioned above). In the context of face recognition, this is consistent with the idea of
the eyes as features

rich in semantic information (such as importance to social
interaction, Langton et al., 2000
).

In conclusion, the study has failed to find support f
or Hypothesis One and
found highly qualified support for Hypothesis Two. Although the findings followed
the predicted trends in all areas, only one set
-

the different
-
angle mid
-
profile
advantage


reached significance. In relation to the findings for Hypo
thesis One, it
was suggested early on that the homogenous nature of the stimuli used might cause a
reduction or negation of the expected effect since t
hey were not as rich in detail (i.e.,
little or no facial hair, glasses, etc) as those used in the Mäntyl
ä and Holm (2006)
study on which the present one was based.

The findings were consistent with the
previous suggestion (Mäntylä, 1997) that richer pictorial stimuli provide better
material for investigating dual
-
process models of memory.

Exam number 4193800


40


As for the surpris
ing results obtained for Hypothesis Two, these were difficult
to account for althou
gh several suggestions were put forward.
The same
-
angle
advantage for the mid
-
profile view was not quite reached (p=0.08), but there was
found a significant beneficial effec
t of the mid
-
profile view at test.
This is surprising
because previous studies have located the mid
-
profile advantage in the study view
(Liu & Chaudhuri, 2002).
One tentative
explanation for the unexpected test
-
view
advantage
was that

concerning

proposed

d
ifferences in
configural versus
componential
processing between the mid
-
profile and frontal view,

where the
different views optimise one or the other aspect of processing at study, which can be
accessed for matching/transposing at test. Furthermore, the p
resence of
shadowing

in
the stimuli may

have altered
encoding and/or recognition, possibly contributing to the
high percentage of guess responses obtained.

















References


Althoff, R.R., & Cohen, N.J. (1999). Eye
-
movement
-
based memory effect: a



reprocessing effect in face perception.
Journal of Experimental Psychology:


Learning, Memory, & Cognition, 25
, 997
-
1010.


Exam number 4193800


41

Bindemann, M., Scheepers, C.
,

& Burton, A.M. (2009). Viewpoint and center of


gravity affect eye movements to human faces.
Journal

of Vision, 9
, 1
-
16.


Bruce, V. (2009). Remembering faces. In J., Brockmole, (Ed.).
The Visual World in

Memory

(pp.66
-
88). Hove: Psychology Press.


Bruce, V., &

Young, A.W. (1986). Understanding face recognition.
British Journal of


Psychology, 77
, 305
-
3
28.


Gardiner, J.M., Java, R.I., & Richardson
-
Klavehn, A.R. (1996). How level of



processing really influences awareness in recognition memory.
Canadian


Journal of Experimental Psychology, 50
, 114
-
122.


Gobbini, M.I. and Haxby, J.V. (2007). Neural system
s for recognition of familiar



faces.

Neuropsychologia, 45
, 32
-
41.


Goffaux, V., Rossion, B., Sorger, B., Schiltz, C., & Goebel, R. (2009). Face inversion


disrupts the perception of vertical relations between features in the right



human occipito
-
tempo
ral cortex.
Journal of Neuropsychology, 3
, 45
-
67.


Haxby, J.V., Hoffman, E.A., &

Gobbini, M.I. (2000). The distributed human neural

network for face perception.
Trends in Cognitive Sciences, 4
, 223
-
233.


Hancock, P.J.B., Bruce, V.
,

& Burton, A.M. (2000).
Recognition of unfamiliar faces.

Trends in Cognitive Sciences, 4
, 330
-
337.


Exam number 4193800


42

Hill, H., Schyns, P.G.
,

& Akamatsu, S. (1997). Information and viewpoint




dependence in face recognition.
Cognition, 62
, 201
-
222.


Jacoby, A.L., Yonelinas, A.P., & Jennings, J.M.

(1997)
. The relationship between



conscious and unconscious (automatic) influences: a declaration of



independence. In J.D. Cohen & J.W. Schooler (Eds.)
Scientific Approaches to


Consciousness

(pp13
-
47). New Jersey: Lawrence Erlbaum Associates.


Joorden
, S., & Merickle, P.M. (199
3). Independence or redundancy? Two models of


conscious and unconscious influence.
Journal of Experimental Psychology:


General, 122
, 462
-
467.


Kopelman, M.D. (2002). Disorders of memory.
Brain, 125
, 2152
-
2190.


Krouse, F.L.

(
19
81
)
. Effects of pose, pose change, and delay on face recognition



performance.
Journal of Applied Psychology, 66
, 651
-
654.


Langton, S.R.H., Watt, R.J.
,

& Bruce, V. (2000). Do the eyes have it? Cues to the



direction of social attention.
Trends in Cognit
ive Sciences, 4
, 50
-
58.


Laughery, K.R., Alexander, J.F., & Lane, A.B. (1971). Recognition of human faces:


affects of target exposure time, target position, pose position, and type of



photograph.
Journal of Applied Psychology, 55
, 477
-
483.


Liu, C.H.
,

&

Chaudhuri, A. (2002). Reassessing the ¾ view effect in face recognition.


Cognition, 83
, 31
-
48.


Mandler, G. (1980). Recognizing: the judgment of previous occurrence.




Psychological Review, 87
, 252
-
271.


Mäntylä, T. (1997). Recollections of faces: reme
mbering differences and knowing



similarities.
Journal of Experimental Psychology: Learning, Memory and



Cognition, 23
, 1203
-
1216.


Exam number 4193800


43

Mäntylä, T.
,

& Holm, L. (2006). Gaze control and recollective experience in face



recognition.
Visual Cognition, 13
, 365
-
386.


Newell, F.N., Chiroro, P.
,

& Valentine, T. (1999). Recognizing unfamiliar faces: the


effects of distinctiveness and view.
The Quarterly Journal of Experimental

Psychology, 52A
, 509
-
534.


O’Toole, A.J., Edelman, S.
,

& Bülthoff, H.H. (1998). Stimulu
s
-
specific effects in face


recognition over changes in viewpoint.
Vision Research, 38
, 2351
-
2363.


Palmer, S., Rosch, E.
,

& Chase, P. (1981). Canonical perspective and the perspective

of objects. In: J. Long,

&

A.

Baddeley,

(Eds.).
Attention and




Perfo
rmance, 9
th

Edition

(pp135
-
151)
. New Jersey: Lawrence Erlbaum



Associates.


Peer, P., CVL Face Database:
http://www.lrv.fri.uni
-
lj.si/facedb.html
.


Pike, G.E., Kemp, R.I., Towell, N.A., & Phillips,

K.C. (1997). Recognizing moving


faces: the relative contribution of motion and perspective view information.


Visual Cognition, 4
, 409
-
437.


Rajaram, S.
,

& Roediger, H.L. (III) (1997). Remembering and knowing as states of

consciousness du
ring retrieval.
Chapter 11 in:
J.D. Cohen,

&

J.W. Schooler
,

(Eds.)

Scientific Approaches to Consciousness

(pp
213
-
240
)
. New Jersey:



Lawrence Erlbaum Associates.


Solina, F., Peer, P., Batagelj, B., Juvan,S.,

&

Kovač, J
. (2003).

“Color
-
based face


detection in the ’15 seconds of fame’ art installation”, In:
Mirage 2003
,


Conference on Computer Vision / Computer Graphics Collaboration for

Model
-
based Imaging, Rendering, Image Analysis and Graphical Special

Effe
cts, March 10
-
11
,

(pp38
-
47).

INRIA Rocquencourt, France, Wilfired


Philips,

Rocquencourt, INRIA.


Exam number 4193800


44

Troje, N.F., & Bülthoff, H.H. (1996). Face recognition under varying poses: the role

of texture and shape.
Vision Research, 36
, 1761
-
1771.


Tsvilis, D., Otte
n, L.J., & Rugg, M.D. (2001). Context effects on the neural correlates

of recognition memory: an electrophysiological study.
Neuron, 31
, 497
-
505.


Tulving, E. (1985). Memory and consciousness.
Canadian Journal of Psychology, 26
,

1
-
12.


Wheeler, M.A.
,

& St
uss, D.T. (2003). Remembering and knowing in patients with


frontal lobe injuries.
Cortex, 39
, 827
-
846.


Yonelinas, A.P. (2002). The nature of recollection and familiarity: a review of 30


years of research.
Journal of Memory and Language, 46
, 441
-
517.


















Appendices


Appendix A


Participant Information Sheet

Exam number 4193800


45


What will happen:

In this study, you will be asked to look at a series of
pictures of male faces, and later asked whether you recognise them. You will be
required to sit in front of a c
omputer screen whilst you view the stimuli. For one set of
faces you view, you will be asked to rate how distinctive each face is on a scale from
one to four, by pressing the appropriate button (which will be indicated). Following
completion of the first s
et, you will have 2 minutes to rest while the researcher
explains the next part. For the other set, you will be asked to sort each face according
to whether you think the person looks ‘intelligent’, ‘sporty’, a ‘party
-
goer’ or a
‘home
-
body’, again by press
ing the appropriate button; how you decide to sort the
faces is entirely up to you


the only criterion is that the faces in each category should
be as similar to each other as possible. You will be able to indicate that you are ‘not
sure’ if necessary. Y
ou will have the chance to practise before beginning. The initial
viewing phase should take around 10 minutes. Following this the next phase will be
explained to you.

In the next phase, you will be asked to view more pictures of faces and
indicate whether

you recognise them or not. Some of the faces you will have seen
before, some will be new. For each face, you will be requested to press a key to show
whether you recognise or do not recognise the face. For each face you recognise, you
will be asked to pre
ss a key to indicate whether you explicitly ‘remember’ seeing the
face before or only have a sense of ‘knowing’ the face. These different types of
memory will be explained in more detail, and you will not be required to start this
stage until the researche
r is satisfied that you understand; you will also be able to
indicate that you ‘guessed’ if you are not sure which type of remembering took place.
This phase should take no longer than 20 minutes.

Once you have finished the experiment, you will be free to
make comments
and/or ask the researcher any remaining questions you have.



Time Commitment
.
The study typically takes around 30 minutes.











Appendix B


Instructions given to Subjects


Support Notes
-

Study


Exam number 4193800


46

There are two sections to the first hal
f of this experiment. In each part you will see a
series of faces. In one section you will be asked to make a particular type of
judgement about the faces you see, and in the next section you will be asked to make
a different type of judgement about the fa
ces. In the second half of the experiment you
will be asked whether you recognise the faces when they are mixed in with new ones.


First you will practise each part of the experimental procedure


the two study
sections followed by the test section.


We be
gin with the practise session for the judgement type ‘categorising faces’:


Categorising faces


For each face you see, you should sort it into one of four categories: ‘sporty
-
type’,
‘intellectual’, ‘party
-
goer’ or ‘home
-
body’. If you cannot decide, you ma
y respond
‘not sure’. There are no correct responses here; just try to ensure that the faces in
each category are as similar to each other as possible.


There is no time
-
limit, so please take your time. Begin when ready.


Next we have the practise sessi
on for the judgement section ‘rating faces for
distinctiveness’:


Rating for distinctiveness


For each face you see, imagine that you have to meet that person at a busy railway
station. How easy would it be to pick the face out in a crowd? Faces that are

very
distinctive are easy to pick out, whereas faces that are not distinctive are hard to pick
out.


Again there is no time
-
limit. Begin when ready.


NB: these two sections may run in a different order in the experiment proper, so
please make sure you p
ay attention to the type of judgement you are being asked to
make.











Appendix B Continued:


Support Notes
-

Test


Exam number 4193800


47

In the test section you will see a series of faces that includes faces that you saw earlier
mixed in with new faces. Of the face
s that you have seen before, some will be at a
different angle, but some will be at the same angle as before.


In this last section, each face will stay on the screen until you respond ‘Old’ or ‘New’.
If the face is ‘Old’, press __, if the face is ‘New’,
press __. I will leave a note to
remind you.


Please respond as soon as you have decided whether you recognise the face or not.


If you respond that a face is ‘New’, another face will be presented.
If you respond that
a face is ‘Old’, you will be asked a
nother question about the type of memory that you
had when you recognised the face.


Remember/Know/Guess


There are three options: ‘Remember’, ‘Know’ and ‘Guess’. ‘Remembering’ and
‘Knowing’ refer to different types of recognition.


‘Knowing’ is often defi
ned as a ‘feeling of familiarity’ that lacks specific recollection.
It is the type of recognition you experience when you see a face you feel you know
although you cannot remember anything explicitly about that person. You may have
experienced it upon see
ing someone in the street: you feel you know the face but
cannot remember anything about the person or where you know them from.


‘Remembering’ refers to another type of recognition. You should respond that you
‘Remember’ the face if looking at it brings
an explicit recollection from the study
section of the experiment. For example, you may remember a particular thought or
feeling that you had, or that you associate with that face.


If you are not sure whether you ‘Remember’ or ‘Know’ the face, you will
be able to
respond with ‘Guess’. The options will be provided on the screen for you.