Face
Recognition 1
Running head: EFFECT OF VIEWING ANGLE ON FACE
RECOGNITION
Face Recognition
2
The present study explored the effect of viewing angle on recognition of
face photographs. We predicted that three-quarter views of faces promote better
recognition memory for unfamiliar faces than do full-face views. A total of 60
participants (43 men and 17 women; mean age of 20.15) were randomly selected
from university campus. All participants were presented to a set of 16
different target pictures and then asked to recognise the target pictures from
another set of 32 pictures. The results of the first method suggested that
there was a significant effect on all viewing angles and on weather the
photographs were targets or foils (p<0.001). The results of the
second method indicated that there was a recognition effect but not
presentation effect.
The recognition memory of faces has
been the focus of concentration in the field of eyewitness identification. Some
studies have revealed an apparent advantage in the recognition memory of the ¾
(25 degrees) over full-face and profile representation of faces. Krouse (1981)
presented previously novel faces in either full-face or ¾ view and then tested
them in the same or a different view in a forced-choice recognition test. He
found a significant effect of study pose, which was unaffected by the delay
which intervened between study and test. Three-quarter views at presentation
led to superior levels of recognition memory at test.
Logie, Baddeley and Woodhead (1987)
who compared initial study of faces in full-face, ¾, and profile views found a
similar effect. In the testing stage, the faces were presented in full-face, ¾
or profile and there was always a change in the expression between presentation
and test. Initial study of ¾ views led to superior overall recognition of faces
in the test, compared with study of full-face or profile alone.
Another study by Fargan (1979) has
revealed that babies also show a superior recognition memory for the ¾ view
faces. Infants who were 5 months old could only distinguish faces when they
were shown in the ¾ view. In addition, when infants reach age 7 months old,
they were able to distinguish between two very similar male faces on the ¾ view
angle.
One possible interpretation for the
¾ view effect was offered by Fargan (1979) who suggested that the ¾ views
reveal more of the features which may be useful for recognition memory purposes
than either the full-face or profile. If the ¾ view increase
the number of
encoded features, it is more likely to promote better later memory for faces
presented in the same or in different views. Thus, a face identification task
could require featural information which is best realized by the geometry of a
three-quarter view.
Furthermore,
Perrett and Oram (1991) found that the three-quarter preference could result
from the way our brain is “hardwired” to represent faces. Neurophysiological
evidence suggests that while most face-selective cells are tuned to full-face
and profile views, and the three-quarter angle activates both sets of neurons.
However, Bruce, Valentine and Baddeley (1987) suggest that the advantages
associated with the three-quarter view may be limited to the task of matching
unfamiliar faces and would not reflect more fundamental properties of the
representation of faces.
Hill,
Schyns, and Akamatsu (1996) used three-dimensional facial surface
representation to investigate the effect of rotations in depth on a face
recognition task. They showed by the shape representations that all views used
(full-face, three-quarter and profile) were equally well recognized on the
testing stage. The results suggest that the three-dimensional shape information
is fundamental for recognition across rotations in depth, although superficial
information may also be used to reduce viewpoint dependence.
Another study examined the ability
of participants to recognize a target person (whom they had seen previously) in
a test series of 150 pictures of faces (Laughery, Alexander & Lane, 1971).
They found that the pose position in the test series (front, profile, or
portrait view) and the type of photograph (colour or black and white) did not
affect recognition. These results have implications for procedures used by law
enforcement
agencies in the process of a witness attempting to identify a criminal from a
set of mug shots.
The purpose of the present study was
to determine whether participants are able to recognize better the target
photographs as the angle of view changed from full-face. An additional purpose
was to examine the effect of angle presentation on the face recognition task.
Participants
Participants consisted
of 60 randomly selected participants who volunteered to participate in the
experiment and who were unlikely to be familiar with most faces shown in the
photographs. They were not members of Advanced Experimental Psychology and
Introduction of Psychology classes. The sample consisted of 43 men and 17 women
with a mean of 20.15 years (SD=3.73 years; range from 17 to 24 years).
Materials
Materials contained 48
photographs (black and white) of white male faces which were divided into 2
sets. The presentation set consists of 16 unique faces that include 4 different
faces taken at full-face (0 degree), 4 taken at a viewing angle of 12.5
degrees, 4 taken at a viewing angle of 25 degrees and 4 taken at a viewing
angle of 37.5 degrees. The recognition set contains 32 unique faces, 16 of
which are in the presentation set (targets) and 16 that are not presented
before (foils). The targets and the foils each have 4 faces taken at each
viewing angle (0, 12.5, 25, 37.5 degrees). For each face in the target set,
there existed a matching face in the presentation set, except at a possible
different
Face Recognition 6
viewing angle. Every combination of presentation angle to target
recognition angle (i.e. 0
degrees presentation to 25 degrees target) was present in the experiment.
All the faces in the photographs were taken against a white background and did
not contain any distinguishing features such as earring, scars or excessive
facial hair. A response sheet was used in the recognition set with a
recognition response scale from 1 to 5 for each photograph (1 = ”sure not”, 2 =
“think not”, 3 = “unsure”, 4 = “think so”, 5 = “sure so”).
Procedure
After completing a consent form, participants were
instructed to remember the faces on the presentation set. The photographs were
then presented sequentially for 2 seconds until all the 16 photographs were
shown. In the recognition set, the participants were instructed to rate the
degree of confidence (on the scale sheet) to which they had seen each face in
the presentation set. These instructions were given verbally to them and were
repeated in the written instructions for completing the response sheet which
was given to them prior to the experiment. Presentation ordering effects were
controlled by shuffling the photographs and recording the order for each
participant before the presentation of each set. Participants took
approximately 15 min to complete the experiment. They were given a debriefing
sheet and thanked for their help after completing the study.
All scale scores of the experiment in the recognition
set appeared normally distributed. The scores of individuals were randomly
assigned in two equal groups of 30 participants and a different method was
applied by means of an analysis of variance. The
Face Recognition 7
first method collected the data of the targets and foils with respect to
the viewing angle. This method measured the effect of all viewing angles, the
effect of weather the photographs were targets or foils and the interaction
between the two previous effects.
Using an alpha level of .05, the analysis of variance
revealed a significant effect in all the viewing angles, F(3, 87)=
11.5402, MSE= 8.2768 , p<.0001, such that full-face angle (M
= 10.20) indicated a smaller mean than the 37.5 degrees angle (M =
12.25) in a linear relation (table 1). The mean of the ¾ angle (M
=12.77) was found to be the highest among all the means. There was a
significant effect in the distinction between targets and foils (p <
.0001) and the mean of target (M = 14.87) was higher than the mean of
foils (M = 8.01). The interaction between the two effects was also found
to be significant (p < .0001).
The second method collected the data of all the
presentation angles with respect to the recognition angle. This method looked at
the effect of presentation angle, the effect of recognition angle and the
interaction between them. Using the same alpha level as above, the analysis of
variance revealed no presentation angle effect, F(3, 87)= 1.0887, MSE=
2.1554 (p =.3587), such that full-face angle (M = 3.81) was the
same as the other means (table 2). However, there was a significant effect in
the recognition angle (p < .0001) and the mean of full-face angle (M
= 3.13) was increasing in a linear fashion. The interaction between the two was
also found to be significant (p < .0001).
The findings of the
study showed that there is a significant effect in all the viewing angles and
in the distinction of targets and foils. The results of the participants
Face Recognition 8
indicate that as the angle of viewing change from full-face (0 degrees)
to 37.5 degrees, the level of confidence increase and the target recognition is
more accurate. More precisely, participants demonstrated the ability to
recognize very well a target photograph when it was presented in a viewing
angle of 25 degrees. A close observation of the means suggest that there is a
linear relation between the angle of view and the scores of the mean. However,
this linear relation was not found in the results of the foils because the
participants were able to identify the foils fairly well in the recognition
set.
The results of the
present study were consistent with the findings of Bruce, Valentine and
Baddeley (1987) who used a similar methodology. They presented the participants
familiar and unfamiliar photograph faces in different view angles and asked
them to identify these photographs in the recognition set. They found that
there was a ¾ view effect in the visual matching of unfamiliar faces, but the
same effect was not found in the representations of familiar faces. Thus,
unfamiliar faces are recognized very well at a 25 degrees view angle.
One possible explanation
for these results is that as the angle change from 0 degrees to 37.5 degrees
there is an increase of useful information for depth and for encoding features
of the face. For example, when a face is observed from the front it is hard to
obtain information of the shape of the nose and as a result of that is may be
hard to recognize the face when it is observe from a different angle such as
37.5.
However, there were several participants that
did not show the linear pattern on the target photographs. It might be argued
that some of the photograph faces were familiar to the participants and this
may influence their decision and increase their
Face Recognition 9
accuracy. However, there were only a few participants that recognize a
familiar face and this may have a minor effect on the results of the
experiment.
The
other finding of experiment showed that there was no effect in the presentation
angle and that there was a significant effect in the recognition angle. These
results indicate that the angle of presentation is not important for face
recognition and hence, people can recognize faces regardless to the angle that
they were first presented.
The same results were found by
Laughery et al.(1971) who used the same methodology in the recognition set.
They found that the pose position in the test series (front, profile, or
portrait view) and the type of photograph (colour or black and white) did not
affect on the recognition set. These results may suggest some procedures that
can be taken into consideration by the police in order to identify a criminal
faces from a set of mug shots. However, this is a subject for debate since some
studies have shown that eyewitness testimony is not accurate and it is subject
to different types of influences (Lindsay, Wells, & Rumpel, 1981; Lindsay,
Read, & Sharma, 1998; Terry, 1993).
They
found that one reason eyewitnesses may have difficulty in identifying the
suspect of a crime is due to transformations in regions of the face
(expression, facial hair, or eyeglasses). Thus, they suggested that these
transformations did lower recognition accuracy. This fact was taken into
consideration in our experiment and all the photograph faces did no have any
distinct features.
Furthermore, the interaction between
the effect of presentation angle and recognition angle strongly indicate that
participants are more likely to recognize accurately faces when the angle of
presentation is the same as the angle of recognition.
Face
Recognition 10
For example, If a participant is first presented with a full-face photograph, then he will recognize better the same face in the same angle. The result obtained here consistent with the results of Bruce (1982) who examined the effect on recognition accuracy and latency of changing the view of faces between the presentation and test.
In his studies, he presented the
participants with unfamiliar faces in different angles and on the test set some
of the pictures changed (full-face to ¾) angle and some remained the same
angle. He found that unchanged faces were recognized more quickly and
accurately than faces with a change in angle.
The design of analysis that used for this experiment was the two-way analysis of variance. One of the assumptions underlying this analysis is independent observations and the assumption of homogeneity of variance. The major threat for this design is when there is no homogeneity of variance because the ratio is not distributed as an F distribution. In this design the sample size are the same size in both methods which indicates that the F test robust to the deviation of normality. This design was chosen over other designs because it was predicted in our experiment that there is at least one better viewing angle than the other angles. This assumption was met in our experiment since the means of the angles are not equal to each other.
Future studies will need to check how do faces are stored in the brain in the region that is devoted for faces. This may give us some hints as to how faces are represented in the brain and why there is 25 degrees superiority in face recognition. Continue research on this topic may help to resolve the mystery ¾ effect and will ultimately increase the accuracy for eyewitness identification.
Face Recognition 11
References
Bruce, V. (1982). Changing faces: Visual and non-visual coding processes in face recognition. British Journal of Psychology, 73, 105-116.
Bruce, V., Valentine, T., & Baddeley, A. (1987). The basis of the ¾ view advantage in face recognition. Applied Cognitive Psychology, 1, 109-120.
Hill, H., Schyns, P. G., & Akamatsu, S. (1996). Information and viewpoint dependence in face recognition. Cognition, 62, 201-222.
Krouse, F. L. (1981). Effects of pose, pose change, and delay on face recognition performance. Journal of Applied Psychology, 66, 651-654.
Laughery, K. R., Alexander, J. F., & Lane, A. B. (1971). Effects of target exposure time, target position, pose position, and type of photograph. Journal of Applied Psychology, 55, 477-483.
Lindsay, D. S., Read, J. D., & Sharma, K. (1998). Accuracy and confidence in person identification: The relationship is strong when witnessing conditions vary widely. Psychological Science, 9, 215-218.
Lindsay, R. C. L., Wells, G. L., & Rumpel, C. M.
(1981). Can people detect eyewitness-identification accuracy within and across
situations? Journal of Applied Psychology, 66, 79-89.
Logie, R. H., Baddeley, A. D., & Woodhead, M. M. (1987). Face recognition, pose and ecological validity. Applied Cognitive Psychology, 1, 53-69.
Perrett, D. I., & Oram, M. W. (1991). Visual cells in the temporal cortex sensitive to face view and gaze direction. Experimental Brain Research, 86, 159-173.
Terry, R. L., (1993). Effects of facial transformations
on accuracy of recognition. The Journal of Social Psychology, 134, 483-492.
Angle of viewing (degrees) 0 12.5 25 37.5
M = 10.20
M = 10.64 M =
12.77 M = 12.25
SD = 3.58 SD = 5.20 SD = 4.68 SD = 4.79
Effect of weather targets or foils:
Targets Foils
M = 14.87 M = 8.01
SD = 3.20 SD = 3.23
Interaction:
Angle of viewing (degrees) 0 12.5 25 37.5
M = 12.20
M = 16.13 M =
18.27 M = 15.87
SD = 2.76 SD = 2.65 SD = 2.53 SD = 3.28
M = 8.20 M = 5.93 M = 9.27 M = 8.63
SD = 3.18 SD = 2.05 SD = 3.58 SD = 3.01
Angle of viewing (degrees) 0 12.5 25 37.5
M = 3.81 M = 3.88 M = 3.89 M = 3.59
SD = 1.37 SD = 1.45 SD = 1.42 SD = 1.49
Effect of recognition angle:
Angle of viewing (degrees) 0 12.5 25 37.5
M
= 3.13 M = 3.88
M = 4.00 M =
4.17
SD = 1.53 SD = 1.42 SD = 1.35 SD = 1.19
Interaction:
Angle of viewing (degrees) 0 12.5 25 37.5
M
= 4.20 M = 2.90
M = 3.47 M =
1.97
SD = 1.26 SD = 1.32 SD = 1.48 SD = 1.16
M = 3.30 M = 4.73 M = 3.73 M = 3.73
SD = 1.56 SD = .75
SD = 1.58 SD
= 1.28
M = 4.07
M = 4.10 M =
4.03 M = 3.80
SD = 1.23 SD = 1.45
SD = 1.54 SD
= 1.21
M = 3.67 M = 3.80 M = 4.33 M = 4.87
SD = 1.26 SD = 1.54 SD = .88 SD
= .35