Faces in Perception [1166] Perception, 1987, volume 16, pages 747-759
Configurational information in face perception
Andrew W Young, Deborah Hellawell, Dennis C Hay Department of Psychology, University of Lancaster, Bailrigg, Lancaster LA1 4YF. UK Received 23 July 1987, in revised form 19 November 1987
Abstract. A new facial composites technique is demonstrated, in which photographs of the top and bottom halves of different familiar faces fuse to form unfamiliar faces when aligned with each other. The perception of a novel configuration in such composite stimuli is sufficiently convincing to interfere with identification of the constituent parts (experiment 1}, but this effect disappears when stimuli are inverted (experiment 2) . Difficulty in identifying the parts of upright composites is found even for stimuli made from parts of unfamiliar faces that have only ever been encountered as face fragments (experiment 3). An equivalent effect is found for compo sites made from internal and external facial features of well-known people (experiment 4 ). These findings demonstrate the importance of configurational information in face perception, and that configurations are only properly perceived in upright faces. 1
Introduction
In 1878 Francis Galton presented a paper to the Anthropological Institute in which he described his investigations of composite portraits. By superimposing images of people's faces on top of each other he considered that he could arrive at a photograph which would show their typical characteristics whilst reducing or eliminating unusual variation. Galton (1879, 1883} thought that composite portraiture could have many potential uses. It might, for instance, be used to arrive at a better idea of the appearance of historical figures, by combining the works of different artists in such a way that their separate mistakes and stylistic idiosyncrasies would disappear (to this end, he showed a composite of six different versions of the profile of Alexander the Great). It might also be used to illustrate the typical appearance of people in health (demonstrated with a composite of twelve officers and eleven privates from the Royal Engineers} or with different types of disease ( fifty-six consumptive cases, and fifteen tubercular patients). With the assistance of Sir Edmund Du Cane, HM Director of Prisons, Galton was investigating the physiognomical types associated with different forms of criminality. Other applications would include animal breeding, and forecasting the appearance of the offspring of different marriages. Although few scientists would now accept Galton's assumption that facial features can provide veridical access to all of these types of information (Shepherd 1988}, there was an interesting and astute observation behind his approach. Galton had noticed that faces form such a homogeneous class of visual stimuli that we must become attuned to quite minor variations in order to differentiate them. He expressed it thus: "The differences in human features must be reckoned great, inasmuch as they enable us to distinguish a single known face among those of thousands of strangers, though they are mostly too minute for measurement. At the same time, they are exceedingly numerous. The general expression of a face is the sum of a multitude of small details, which are viewed in such rapid succession that we seem to perceive them all at a single glance. If any one of them disagrees with the recollected traits of a known face, the eye is quick at observingit, and it dwells upon the difference. One small discordance overweighs a multitude of similarities and suggests a general unlikeness ... ." (Galton 1883, page 3)
Faces in Perception [1167] 748
A W Young, D Hellawell, D C Hay
In this extract Galton touched on two of the ideas that have become central to current thinking about how we recognise familiar faces. First, the point that faces demand identification of individual members of a relatively homogeneous stimulus class (Ellis 1975, 1981; Damasio et al 1982). Second, the idea that we take in a seen face as a 'whole' or, as Galton expressed it, "at a single glance" (Calis et al 1984). This importance of the facial gestalt has perhaps been most neatly expressed by Carey and Diamond (1977), who drew a distinction between individual facial features themselves and the configuration formed by these features. The importance of configurational information to face perception is widely accepted. It can be seen, for instance, in the paintings of faces created since the mid-sixteenth century by allegorical painters such as Giuseppe Arcimboldo and Utagawa Kuniyoshi (Hulten 1987; Okabe 1987). In such paintings the individual facial features may be replaced by bodies, animals, flowers, fruits, or even books, according to the purpose of the allegory, yet a facial configuration is readily perceived (Kaufmann 1987). Other lines of evidence pointing to the importance of facial configuration include the difficulty of constructing a Photofit likeness, and the effectiveness of different types of disguise (Baddeley 1979; Davies 1982). Haig (1984) has demonstrated the way in which small changes in the configuration of a constant set of features can alter a person's appearance, and Sergent (1984) has established the joint influence of feature and configurational information in a face matching task. In this paper we report and investigate a phenomenon that further illustrates the importance of configurational information to face perception. This involves the use of composites in which the features of different people are arbitrarily combined (but not, as in Galton's technique, superimposed). Consider the faces shown in figure 1. Each is made from the top and bottom ·halves of the faces of different people, yet these readily fuse to form an effective whole. More interestingly, the top and bottom halves we have used come from highly familiar faces, yet this is not immediately apparent; instead the novel configuration creates in each case an initial impression of an unfamiliar face. The faces used in figure 1 are those of Margaret Thatcher and Shirley Williams; effective fusion occurs despite the difference in political persuasions. Highly familiar faces may thus be combined to create an unfamiliar facial configuration, and the experience of seeing a new configuration is sufficiently powerful that it can interfere with the identification of the individual features from which such composites are formed. We will explore different properties of this new tYPe of facial composite for their contribution to understanding the importance of configurational information in face
Figure I.
Composites made by joining the top and bottom halves of faces of different people.
Faces in Perception [1168] 749
Facial configuration
perception. In experiment 1 we begin by establishing that the basic phenomenon holds across a group of observers. 2
Experiment
l
Our aim in this experiment was to determine whether the perception of a complete face in top and bottom half-face composites interferes with identification of the constituent parts. This was achieved by comparing the reaction times for identifying parts of composite faces to the reaction times for identifying noncomposite stimuli in which the top and bottom halves of familiar faces were aligned in a way that did not form a face shape. 2.1 Method 2.1.1 Subjects. Six male and six female subjects were paid to participate in the experiment. All were staff or students of Lancaster University, and had normal or corrected-to-normal vision. 2.1.2 Stimuli and procedure. Black and white press photographs of five people were used as stimuli; the people were Reginald Bosanquet, Max Bygraves, Terry Scott, Lord Snowdon, and Terry Wogan. These were chosen because they are familiar in Britain, do not have unconventional hairstyles, and because a photograph with full-face pose was available. One full-face photograph of each person was selected. Because of restrictions inherent in the use of press photographs, expresssion was not standardised; in three of the photographs the person was shown with an expression in which the mouth was closed, and in two photographs the mouth was slightly open. These pictures were rephotographed and printed with the faces a standard width. Each face was then cut away from its background, and divided into top and bottom segments by cutting along a horizontal line below the eyes. The resulting five top face segments and five bottom-face segments were used to construct stimuli that were presented to the subjects in the form of back-projected slides. Two types of stimuli were made, which we will refer to as composites and noncomposites. Examples are given in figure 2: Composites. A top-face segment was joined to the bottom segment of a different face. All of the twenty composites resulting from joining the four inappropriate bottom segments to each of the five top segments were used.
(a)
(b)
Top half of Max Bygrave's face and the bottom half of Lord Snowdon's face in (a) composite and (b) noncomposite arrangements.
Figure 2.
Faces in Perception [1169] A W Young, D Hellawell, D C Hay
750
Noncomposites. A top-face segment was positioned above the bottom segment of a different face in such a manner that the two segments did not form a new composite face; this was achieved by positioning the nose from the bottom segment close to one edge of the top segment. All of the twenty possible top and bottom pairings used to make the composite s timuli were used again to make noncomposites. Because the noncomposite stimuli occupy a wider horizontal visual angle than do the composites ( see figure 2), slides were taken so that the centre of each noncomposite stimulus was in the centre of the slide; the left and right positions of top and bottom segments of the noncomposites were counterbalanced across stimuli. Composites occupied counterbalanced positions slightly to the left or to the right of the centre of each slide, with the left and right positions being those in which parts of the non composites could occur. Figure 3 shows the original faces used in the experiment, and examples of composites and noncomposites made from each. Original
Composite
Noncomposite
The five familiar faces used in experiment 1 ( 'original' ), examples of composite stimuli created from these, and the equivalent noncomposites.
Figure 3.
Faces in Perception [1170] Facial configuration
751
The experiment began with a warm-up session, in which subjects were shown the original photographs from which the composites and noncomposites had been constructed. They were asked to name each person, and reminded of any names they had difficulty remembering. In the experiment proper, each of the twenty composites and twenty noncomposites appeared in random order as back-projected slides, with the composites occupying a horizontal visual angle of approximately 5 deg. Two blocks of trials were run. In one block subjects were asked to name the top segment from each stimulus as quickly as possible, and in the other block to name the bottom segment. The order of presentation of these blocks of trials was counterbalanced across subjects. Each block was preceded by appropriate practice trials in which either top or bottom segments of faces were presented in isolation, and subjects were asked to name each person. Naming responses were made by giving the person's surname (Bosanquet, Bygraves, Scott, Snowdon, or Wogan), and vocal reaction times (timed from stimulus onset) were recorded via a voice key. Thus this is a within-subjects design in which naming latencies were recorded for the top segments of twenty composite and twenty noncomposite stimuli, and for the bottom segments of twenty composite and twenty noncomposite stimuli. The responses (ie names) required were the same for all cells of the design, and the same face segments were presented in the form of composite and noncomposite stimuli. 2.2 Results Mean reaction times for correct identification of the top and bottom halves of composite and noncomposite stimuli are shown in table 1. Error rates were low (less than 4% for all cells of the design) and will not be considered further. A two-factor analysis of variance (ANOVA) of the reaction times for correct identi fication was carried out, to determine the effects of task (top-half or bottom-half naming; repeated measure) and stimulus type (composite or noncomposite; repeated measure). This revealed a main effect of stimulus type (Fl. 11 = 27.63, p < 0.001), with faster responses to noncomposites than to composites. A task X stimulus type inter action was also found (Fl. 11 = 6.62, p < 0.05). Tukey tests ( a = 0.05) showed that responses were faster to noncomposite than to composite stimuli for both top-half and bottom-half identification, but that responses to the noncomposites were slower when the bottom half had to be identified (Tukey critical difference = 66 ms). A comparable slowing of bottom-half identifications was not found for the composite stimuli. Mean vocal reaction times (in ms) for correct identification of the top and bottom halves of upright composite and noncomposite stimuli (experiment 1 ) .
Table 1.
Top half Bottom half
Composite
Noncomposite
1297 1282
1041 1123
2.3 Discussion Reaction times were substantially slower to composite than to noncomposite stimuli, indicating that the perception of a novel facial configuration in the composites interfered with the identification of the constituent parts. Although this effect was more marked when subjects were identifying top halves (task x stimulus type interaction), a difference between composite and noncomposite arrangements was found both for top half and for bottom-half face identification. This finding was made despite the fact that the task is apparently only one of discriminating five top-half faces or five bottom-half faces, and despite the fact that no steps were taken to prevent subjects concentrating on
Faces in Perception [1171] A W Young, D Hellawell, D C Hay
752
particular cues (such as hairstyle or mouth shape} to effect these discriminations. Thus it clearly reflects a compelling perceptual phenomenon. 3 Experiment
2
A curious property of faces is that they become difficult to recognise when they are inverted (Yin 1969; Valentine and Bruce 1986}. This does not simply reflect the fact that we usually see upright faces, because the effect of inversion is known to be greater for faces than for many other stimuli that we are also used to seeing in a particular orientation (Yin 1969; Scapinello and Yarmey 1970}. It has been hypothesised that inversion interferes with encoding of the facial configuration (Carey and Diamond 1977}. Our facial composites seemed to us to support this idea. If figure 1 or figure 2 is viewed upside down it is immediately apparent that the top and bottom halves of the composite faces no longer fuse so effectively. In order to provide formal confirmation of this observation we carried out a second experiment, in which the stimuli from experiment 1 were presented in upright or inverted orientation. 3.1 Method 3.1.1 Subjects. Six male and six female subjects were selected in the same manner as for experiment 1. 3.1.2 Stimuli and procedure. Stimuli were the same as those used in experiment 1. However, they were presented upright (as in experiment 1} for half the trials, and inverted for the other half. Trials with upright or with inverted stimuli were blocked together, and the order of these blocks was counterbalanced across subjects. Subjects were always asked to identify the segment of the face containing the eyes and hairline (ie, the 'top' segment in the upright faces). All other details of design and procedure were as for experiment 1. Thus this is a within-subjects design in which latencies for naming the face segment containing the eyes and hairline were recorded from twenty upright composite stimuli, twenty upright noncomposites, twenty inverted composites, and twenty inverted noncomposites. 3.2 Results Error rates were very low (less than 1% for all cells of the design}, and will not be considered further. Mean reaction times for correct responses to upright and inverted composites and noncomposites are shown in table 2. A two-factor ANOVA of the reaction times for correct identification was carried out, to determine the effects of orientation (upright or inverted; repeated measure) and stimulus type (composite or noncomposite; repeated measure}. This revealed a main effect of stimulus type (F1 11 = 54.90, p < 0.001}, with faster responses to non composites than to composites, and an orientation x stimulus type interaction (F1 11 = 46.01, p < 0.001}. Tukey tests ( a 0.05} were used to examine the ori�ntation x stimulus type interaction. These showed that reaction times were slower for composite than noncomposite stimuli otily in the upright condition; for inverted stimuli there was no difference between composites and noncomposites (Tukey critical =
Mean vocal reaction times (in ms) for correct identification of the part of the upright and inverted composite and noncomposite stimuli containing the hairline and eyes (experi ment 2).
Table 2.
Upright Inverted
Composite
Noncomposite
1167 1039
955 1030
Faces in Perception [1172] Facial configuration
753
difference = 64 ms). For the noncomposite stimuli, responses were faster when they were upright than when inverted, but for composites reaction times for inverted stimuli were actually faster than for upright stimuli! 3.3 Discussion The upright condition replicates the finding from experiment 1 that perception of the facial composites interferes with the identification of their constituent parts (ie, slower responses to upright composites than to upright noncomposites). This effect is completely eliminated by inversion, with no difference between reaction times to composites and noncomposites in the inverted condition. In fact, inversion actually made the task easier for the composite stimuli. This finding of an inverted-face superiority provides a dramatic illustration of the absence of interference from configurational information in the inverted composites. 4
Experiment 3
In experiments 1 and 2 we investigated the perception of composites made from parts of familiar faces. We did this largely because it is convenient to use familiar faces if reaction time for identification is to be measured. There are, however, several demonstrations in the literature of differences between the processing of familiar and unfamiliar faces (Bruce and Young 1986; Ellis 1986). Thus it seemed conceivable (though in our view unlikely) that our findings might not apply to face perception in general, but only to the perception of familiar faces. This might, for instance, happen if subjects cannot avoid trying to identify part of a familiar face by referring to the appearance of the whole face; composites might then interfere more with this process than noncomposites. Experiment 3 was thus designed to use faces of unfamiliar people. Of course, these have to become familiar to some extent for an identification to be made, but what we did was to teach subjects only the top halves in isolation. These were then combined with the bottom halves from a different set of faces to form composite and non composite stimuli. Thus we ensured that subjects did not know what each person should look like, as they had only been trained on the top halves. 4.1 Method 4.1.1 Subjects. Six male and six female subjects were selected in the same way as for experiment 1. 4.1.2 Stimuli and procedure. Black and white press photographs showing full-face poses of the faces of unfamiliar people were rephotographed to a standard size and divided into top and bottom segments by cutting along a horizontal line below the eyes. Five top segments that were considered by the authors to show a range of hairstyles comparable to the range used in the familiar faces of experiment 1 were then selected. Five bottom segments that matched the expressions used in the familiar faces of experiment 1 were also selected. These top and bottom segments all came from the faces of different people. Slides showing twenty composite and twenty noncomposite stimuli were then prepared from these top and bottom segments, in a similar manner to that used for experiment 1. For the experiment, subjects were asked to name the top segment of each stimulus. Five surnames were chosen from the Lancaster, Morecambe, and Heysham 1984 Telephone Directory, and one of these surnames was assigned at random to each of the top segments. The surnames were chosen to have the same mean frequency of occur rence in the telephone directory as the surnames used in experiment 1, to have the same initial letters, and to have the same number of letters.
-
Faces in Perception [1173] 754
A W Young, D Hellawell, D C Hay
The experiment began with a warm-up session, in which subjects learnt the surnames by going through the top segments presented in isolation at least six times. The twenty composite and twenty noncomposite stimuli were then presented one at a time in random order, for subjects to name the top segment of each as quickly and accurately as possible. All other features of design and procedure were as for experiment 1. In experiment 3, then, subjects named the top segments of composites made from parts of the faces of unknown people. It is important to note that they learnt arbitrarily assigned surnames to the top segments presented in isolation, and that they never saw the actual bottom segment of any of the faces whose top segments were to be named, since the bottom segments used to form the composite and noncomposite stimuli came from different people. 4.2 Results Mean reaction times for correct responses were 1175 ms for composites, and 1092 ms for noncomposites. Error rates were less than 5% in each case. A one-factor ANOVA, used to determine the effect of stimulus type (composite or noncomposite; repeated measure), showed that responses to composites were slower than to noncomposites (F1, 11 24.31, p < 0.01). =
4.3 Discussion The finding of slower responses to composites than to noncomposites in identifying parts of faces only learnt as fragments again confirms that it is the perception of a novel facial configuration in the composites that underlies the effect. The size of the differ ence between reaction times to composites and noncomposites is, however, somewhat smaller than that found in experiments 1 and 2. The importance of this is difficult to assess because small sets of stimuli were used in the experiments, and thus any effect of minor variations in the stimuli themselves would tend to be exaggerated. The main point is that a difference between reaction times to upright composites and non composites was found in all three experiments. 5
Experiment
4
In experiments 1 to 3, composites were made by joining together face fragments separated along an arbitrary line drawn just below the eyes. The purpose of experi ment 4 was to investigate the creation of composites from fragments that are psychologically more real. Recent studies have emphasised differences between the processing of internal (eyes, nose, mouth) and external (hair, face shape, chin) features of familiar and unfamiliar faces, and established the differential salience of the internal features of familiar faces (Ellis et al 1979; Endo et al 1984; Young et al 1985). We were thus interested in whether or not combinations of incorrect internal and external features would create novel perceptual configurations, and in whether there would be more interference from incorrect internal features on the recognition of external features or vice versa. 5.1 Method 5.1.1 Subjects. Six male and six female subjects were selected in the same manner as for experiment 1. 5.1.2 Stimuli and procedure. The faces of four pairs of famous people were used as stimuli (David Frost and Paul McCartney; Mick Jagger and John Kennedy; Barry Manilow and Sting; David Owen and Terry Wogan). These were selected so that the internal and external features could be easily interchanged to form composite faces. A within-subjects design was used, with two conditions. In one condition the subjects were asked to identify internal features (eyes, nose, and mouth) of the stimulus, and in
Faces in Perception [1174] Facial configuration
755
the other condition they were asked to identify external features (hair, face shape, and chin). Each condition was divided into two phases; a pretest phase and a test phase. In the pretest phase, subjects were given practice at identifying the eight famous people from their internal or external features only (according to the appropriate condition). Each of the eight people was presented four times in each pretest session, in a random order. In the test phase the subjects were again asked to identify the eight famous people from their internal or from their external features. Vocal reaction times (measured from stimulus onset) and accuracies were recorded. In the internal-feature naming condition, each target person's internal features occurred three times in each of the four following ways: (i) Whole face: internal features combined with the correct external features. (ii) Part face: internal features only. (iii) Noncomposite: internal features combined with the inverted external features of the other member of the pair. (iv) Composite: internal features combined with the external features of the other member of the pair. Four corresponding types of stimuli were used for the external-feature naming condition: (i) Whole face: external features combined with the correct internal features. (ii) Part face: external features only. (iii) Noncomposite: external features combined with the inverted internal features of the other member of the pair. (iv) Composite: external features combined with the internal features of the other member of the pair. Examples of stimuli of each type are shown in figure 4. Half the subjects did the internal-feature naming condition first, and half did the external-feature naming condition first. Thus this is a within-subjects design in which naming latencies were recorded for the internal features of twenty-four whole faces, twenty-four part faces, twenty-four non composites, and twenty-four composites, and for the external features of twenty-four whole faces, twenty-four part faces, twenty-four noncomposites, and twenty-four composites. 5.2 Results Mean reaction times for correct identification of internal and external features from whole faces, part faces, noncomposites, and composites, are shown in table 3. Error rates were less than 6% for all cells. A two-factor ANOVA of the reaction times for correct identification was carried out, to determine the effects of task (internal-feature or external-feature naming; repeated measure) and stimulus type (whole face, part face, noncomposite, or composite; repeated measure). The only significant effect was that of stimulus type (F3,33 93.32, p < 0.001). There was no effect of task (F < 1), and the task x stimulus type inter action did not approach significance (f;, 33 1.67, p > 0.1). Tukey tests (a 0.05) =
=
=
Mean vocal reaction times (in ms) for correct identification of the internal or external features from whole faces, part faces, noncomposites, and composites (experiment 4 ).
Table 3.
Internal features External features Overall
W hole face
Part face
Noncomposite Composite
951 959
984 996
1034 1045
1213 1168
955
990
1040
1190
Faces in Perception [1175] A W Young, D Hellawell, D C Hay
7513
were used to examine the effect of stimulus type. These showed that all means were significantly different from each other except those for whole faces and part faces ( Tukey critical difference 44 ms ). =
5.3 Discussion In experiments 1 and 3 we were able to create noncomposites by misaligning the top and bottom halves of the faces. This procedure was not feasible with internal and external feature combinations, so instead we produced noncomposites in which one part of the face was inverted, and also included a condition in which only part of the face was presented. Again, reaction times to composites were slower than to noncomposites, showing interference from the composite configuration onto identification of the individual Internal-feature naming task
External-feature naming task
Whole face
Part face
Noncomposite
Composite
Examples of stimuli used in experiment 4. The correct ·response to all stimuli in the first and third columns is "Mick Jagger", and to stimuli in the second and fourth columns " John Kennedy".
Figure 4.
Faces in Perception [1176] Facial configuration
757
features. In experiment 4 a condition was also included which involved composites formed by rejoining the correct face fragments (whole face); significant facilitation was found for this condition in comparison to the noncomposites, but the facilitation in comparison to the part faces did not reach statistical significance on the somewhat conservative Tukey test. This pattern was found both for internal-feature naming and for external-feature naming, with no evidence of greater interference for one condition than for the other. 6
General discussion
Our facial composites technique clearly taps a compelling phenomenon. With the method used in experiments 1 to 3 the only difference between composite and non composite stimuli was the presence of a facial configuration in the composites. We have thus shown that the perception of a novel configuration in facial composites interferes with recognition of their constituent parts for top-half and bottom-half face composites (experiments 1 to 3) and for internal and external facial feature composites (experi ment 4). Even when the stimuli had only ever been previously seen as top-half face fragments, interference was still found when composites were formed by joining these to bottom-half face fragments (experiment 3). Configurational information, then, is very important to face perception. To some extent this may be considered an example of a more general finding that under certain conditions global stimulus properties take precedence in visual perception (Navon 1977; Martin 1979; Pomerantz 1983) and Humphreys et al (1985) review several studies that examine how global and local form descriptions are combined with each other. Experiment 2, however, demonstrated that interference from the configurational properties of composite faces is only found for upright composites; for inverted stimuli there was no difference between composites and noncomposites. This finding bears out Carey and Diamond's (1977) claim that the facial configuration is not perceived in inverted faces, and we note that Sergent (1984) also found that feature differences were more important than configurational differences when people were asked to match inverted faces. This was known to Arcimboldo, who created 'reversible' paintings such as The Cook (ca 1750) or The Vegetable Gardener (ca 1590), in which a face-like appearance was only evident in one orientation of the picture (see Hulten 1987). Our own results show that it is actually easier to identify the constituent parts of facial composites if they are inverted than if they are upright; the world's first inverted-face superiority. The absence of interference from inverted composites shows, then, that a meaningful configuration is not necessarily perceived in any complex visual stimulus with an unbroken boundary. This raises the question as to the extent to which the use of configurational information is specific to face perception (or, rather, specific to the perception of upright faces). The widely reported susceptibility of faces to performance impairments on inversion has in the past been used to support claims of face specificity (Y in 1969, 1970; Carey and Diamond 1977; see Ellis and Young 1988 for a review). It is now clear, however, that comparable inversion effects can be found for other stimulus classes than faces provided that their members share similar appearances and that the perceiver is expert at differentiating them (Diamond and Carey 1986). This suggests that the importance of configurational information to face perception reflects our very great expertise at discriminating between the otherwise very similar faces of people we meet. Our findings, then, demonstrate directly the importance of configurational informa tion to face perception, and that this configurational information is only readily derived from an upright face. In emphasising the importance of configurational information we
Faces in Perception [1177] 758
A W
Young,
D Hellawell, D C Hay
do not, of course, seek to deny that individual facial features can also be important. It is not, for instance, impossible to see that the features in the composite faces shown in figure 1 are those of Margaret Thatcher and Shirley Williams. Our point is only that it is not at first easy to see this in the presence of a novel configuration. Configurational and featural information are, however, both likely to contribute to normal face recognition ( Sergent 1984; Bruce and Young 1986; Ellis 1986). Indeed, we all know that some people possess striking individual features that make them easy to identify ( eg Denis Healey's eyebrows ). An important task for future research will thus be to specify how configurational and featural cues interact in the recognition of familiar faces. We are grateful to the Press Association for help in finding suitable photo graphs for use as stimuli, and for permission to reproduce those used in our illustrations.
Acknowledgement.
References
Baddeley A D, 1979 "Applied cognitive and cognitive applied psychology: the case of face recognition" in Perspectives on Memory Research ed. L G Nilsson (Hillsdale, NJ: Lawrence Erlbaum Associates) pp 367-388 Bruce V, Young A W, 1986 "Understanding face recognition" British Journal of Psychology 77 305-327 Calis G, Sterenborg J, Maarse F, 1984 "Initial microgenetic steps in single-glance face recog nition" Acta Psychologica55 215-230 Carey S, Diamond R, 1977 "From piecemeal to configurational representation of faces" Science 195 312-314 Damasio A R, Damasio H, van Hoesen G W, 1982 "Prosopagnosia: anatomic basis and behavioral mechanisms" Neurology 32 3 31-341 Davies G M, 1982 " Composite systems for recalling faces-'helping the police with their enquiries?' " Reconstructing the Past: the Role of Psychologists in Criminal Trials ed. A Trankell (Stockholm: Norstedts) pp 299-313 Diamond R, Carey S, 1986 " W hy faces are and are not special: an effect of expertise" Journal of Experimental Psychology: Generalll5 107-117 Ellis H D, 1975 "Recognizing faces" British Journal of Psychology 66 409-426 Ellis H D, 1981 " Theoretical aspects of face recognition" in Perceiving and Remembering Faces eds G M Davies, H D Ellis, J W Shepherd (London: AcademicPress) pp 171-197 Ellis H D, 1986 "Processes underlying face recognition" in The Neuropsychology of Face Perception and Facial Expression ed. R Bruyer (Hillsdale, NJ: LawrenceErlbaum Associates) pp 1-27 Ellis H D, Shepherd J W, Davies G M, 1979 "Identification of familiar and unfamiliar faces from internal and external features: some implications for theories of face recognition" Perception 8 431-439 Ellis H D, Young A W, 1988 "Are faces special?" in Handbook of Research on Face Processing eds A W Young, H D Ellis (Amsterdam: North-Holland) Endo M, Takahashi K, Maruyama K, 1984 "Effects of observer's attitude on the familiarity of faces: using the difference in cue value between central and peripheral facial elements as an index of familiarity" Tohoku Psychologica Folia 43 23-34 Galton F, 1879 " Composite portraits, made by combining those of many different persons into a single resultant figure" Journal of theAnthropological Institute 8 132-144 Galton F, 1883 Inquiries into Human Faculty and Its Development (London: Macmillan) Haig N D, 1984 " The effect of feature displacement on face recognition" Perception 1 3 505-512 Hulten P (ed.), 1987 TheArcimboldo Effect (London: Thames and Hudson) Humphreys G W, Riddoch M J, Quinlan P T, 1985 "Interactive processes in perceptual organisation: evidence from visual agnosia" in Attention and Performance eds M I Posner, 0 S M Marin (Hillsdale, NJ: LawrenceErlbaumAssociates) pp 301-318 Kaufmann T Da C, 1987 " The allegories and their meaning" in The Arcimboldo Effect ed. P Hulten (London: Thames and Hudson) pp 89-108 Martin M, 1979 " Local and global processing: the role of sparsity" Memory and Cognition 7 476-484 Navon D, 1977 "Forest before trees: the precedence of global features in visual processing" Cognitive Psychology 9 353-383
Okabe A, 1987 "Kao at the dawn of modernity" in Thames and Hudson) pp 235-239
The Arcimboldo Effected. P
Hulten (London:
--.
Faces in Perception [1178] 759
Facial configuration
Pomerantz J R,
1983
"Global and local precedence: selective attention in form and motion
perception" Journal of ExperimentalPsychology: Generallll Scapinello K F, Yarmey AD,
1970
516-540
"The role of faniiliarity and orientation in immediate and
delayed recognition of pictorial stimuli" Psychonomic Science 24 329-330
Sergent J,
1984
"An investigation into component and configural processes underlying face
perception" British Journal ofPsychology 75 Shepherd J,
1988
221-242
"The face and social attribution" in Handbook of Research on Face Processing
eds A W Young, HD Ellis (Amsterdam: North-Holland) Valentine T, Bruce V,
1986
"The effect of race, inversion and encoding activity upon face
61259-273 K, 1969 "Looking at upside-down faces" Journal ofExperimentalPsychology 81 141-145 Y in R K, 1970 "Face recognition by brain-injured patients: a dissociable ability?" Neuro psychologia 8 395-402 Young A W, HayDC, McWeeny K H, F lude B M, Ellis A W, 1985 "Matching familiar and unfamiliar faces on .internal and external features" Perception 14 737-746 recognition" ActaPsychologica
Y in R
© 1987 a Pion publication printed in Great Britain
Configurational information in face perception
Andrew W Young, Deborah Hellawell, Dennis C Hay Department of Psychology, University of Lancaster, Bailrigg, Lancaster LA1 4YF. UK Received 23 July 1987, in revised form 19 November 1987
Abstract. A new facial composites technique is demonstrated, in which photographs of the top and bottom halves of different familiar faces fuse to form unfamiliar faces when aligned with each other. The perception of a novel configuration in such composite stimuli is sufficiently convincing to interfere with identification of the constituent parts (experiment 1}, but this effect disappears when stimuli are inverted (experiment 2) . Difficulty in identifying the parts of upright composites is found even for stimuli made from parts of unfamiliar faces that have only ever been encountered as face fragments (experiment 3). An equivalent effect is found for compo sites made from internal and external facial features of well-known people (experiment 4 ). These findings demonstrate the importance of configurational information in face perception, and that configurations are only properly perceived in upright faces. 1
Introduction
In 1878 Francis Galton presented a paper to the Anthropological Institute in which he described his investigations of composite portraits. By superimposing images of people's faces on top of each other he considered that he could arrive at a photograph which would show their typical characteristics whilst reducing or eliminating unusual variation. Galton (1879, 1883} thought that composite portraiture could have many potential uses. It might, for instance, be used to arrive at a better idea of the appearance of historical figures, by combining the works of different artists in such a way that their separate mistakes and stylistic idiosyncrasies would disappear (to this end, he showed a composite of six different versions of the profile of Alexander the Great). It might also be used to illustrate the typical appearance of people in health (demonstrated with a composite of twelve officers and eleven privates from the Royal Engineers} or with different types of disease ( fifty-six consumptive cases, and fifteen tubercular patients). With the assistance of Sir Edmund Du Cane, HM Director of Prisons, Galton was investigating the physiognomical types associated with different forms of criminality. Other applications would include animal breeding, and forecasting the appearance of the offspring of different marriages. Although few scientists would now accept Galton's assumption that facial features can provide veridical access to all of these types of information (Shepherd 1988}, there was an interesting and astute observation behind his approach. Galton had noticed that faces form such a homogeneous class of visual stimuli that we must become attuned to quite minor variations in order to differentiate them. He expressed it thus: "The differences in human features must be reckoned great, inasmuch as they enable us to distinguish a single known face among those of thousands of strangers, though they are mostly too minute for measurement. At the same time, they are exceedingly numerous. The general expression of a face is the sum of a multitude of small details, which are viewed in such rapid succession that we seem to perceive them all at a single glance. If any one of them disagrees with the recollected traits of a known face, the eye is quick at observingit, and it dwells upon the difference. One small discordance overweighs a multitude of similarities and suggests a general unlikeness ... ." (Galton 1883, page 3)
Faces in Perception [1167] 748
A W Young, D Hellawell, D C Hay
In this extract Galton touched on two of the ideas that have become central to current thinking about how we recognise familiar faces. First, the point that faces demand identification of individual members of a relatively homogeneous stimulus class (Ellis 1975, 1981; Damasio et al 1982). Second, the idea that we take in a seen face as a 'whole' or, as Galton expressed it, "at a single glance" (Calis et al 1984). This importance of the facial gestalt has perhaps been most neatly expressed by Carey and Diamond (1977), who drew a distinction between individual facial features themselves and the configuration formed by these features. The importance of configurational information to face perception is widely accepted. It can be seen, for instance, in the paintings of faces created since the mid-sixteenth century by allegorical painters such as Giuseppe Arcimboldo and Utagawa Kuniyoshi (Hulten 1987; Okabe 1987). In such paintings the individual facial features may be replaced by bodies, animals, flowers, fruits, or even books, according to the purpose of the allegory, yet a facial configuration is readily perceived (Kaufmann 1987). Other lines of evidence pointing to the importance of facial configuration include the difficulty of constructing a Photofit likeness, and the effectiveness of different types of disguise (Baddeley 1979; Davies 1982). Haig (1984) has demonstrated the way in which small changes in the configuration of a constant set of features can alter a person's appearance, and Sergent (1984) has established the joint influence of feature and configurational information in a face matching task. In this paper we report and investigate a phenomenon that further illustrates the importance of configurational information to face perception. This involves the use of composites in which the features of different people are arbitrarily combined (but not, as in Galton's technique, superimposed). Consider the faces shown in figure 1. Each is made from the top and bottom ·halves of the faces of different people, yet these readily fuse to form an effective whole. More interestingly, the top and bottom halves we have used come from highly familiar faces, yet this is not immediately apparent; instead the novel configuration creates in each case an initial impression of an unfamiliar face. The faces used in figure 1 are those of Margaret Thatcher and Shirley Williams; effective fusion occurs despite the difference in political persuasions. Highly familiar faces may thus be combined to create an unfamiliar facial configuration, and the experience of seeing a new configuration is sufficiently powerful that it can interfere with the identification of the individual features from which such composites are formed. We will explore different properties of this new tYPe of facial composite for their contribution to understanding the importance of configurational information in face
Figure I.
Composites made by joining the top and bottom halves of faces of different people.
Faces in Perception [1168] 749
Facial configuration
perception. In experiment 1 we begin by establishing that the basic phenomenon holds across a group of observers. 2
Experiment
l
Our aim in this experiment was to determine whether the perception of a complete face in top and bottom half-face composites interferes with identification of the constituent parts. This was achieved by comparing the reaction times for identifying parts of composite faces to the reaction times for identifying noncomposite stimuli in which the top and bottom halves of familiar faces were aligned in a way that did not form a face shape. 2.1 Method 2.1.1 Subjects. Six male and six female subjects were paid to participate in the experiment. All were staff or students of Lancaster University, and had normal or corrected-to-normal vision. 2.1.2 Stimuli and procedure. Black and white press photographs of five people were used as stimuli; the people were Reginald Bosanquet, Max Bygraves, Terry Scott, Lord Snowdon, and Terry Wogan. These were chosen because they are familiar in Britain, do not have unconventional hairstyles, and because a photograph with full-face pose was available. One full-face photograph of each person was selected. Because of restrictions inherent in the use of press photographs, expresssion was not standardised; in three of the photographs the person was shown with an expression in which the mouth was closed, and in two photographs the mouth was slightly open. These pictures were rephotographed and printed with the faces a standard width. Each face was then cut away from its background, and divided into top and bottom segments by cutting along a horizontal line below the eyes. The resulting five top face segments and five bottom-face segments were used to construct stimuli that were presented to the subjects in the form of back-projected slides. Two types of stimuli were made, which we will refer to as composites and noncomposites. Examples are given in figure 2: Composites. A top-face segment was joined to the bottom segment of a different face. All of the twenty composites resulting from joining the four inappropriate bottom segments to each of the five top segments were used.
(a)
(b)
Top half of Max Bygrave's face and the bottom half of Lord Snowdon's face in (a) composite and (b) noncomposite arrangements.
Figure 2.
Faces in Perception [1169] A W Young, D Hellawell, D C Hay
750
Noncomposites. A top-face segment was positioned above the bottom segment of a different face in such a manner that the two segments did not form a new composite face; this was achieved by positioning the nose from the bottom segment close to one edge of the top segment. All of the twenty possible top and bottom pairings used to make the composite s timuli were used again to make noncomposites. Because the noncomposite stimuli occupy a wider horizontal visual angle than do the composites ( see figure 2), slides were taken so that the centre of each noncomposite stimulus was in the centre of the slide; the left and right positions of top and bottom segments of the noncomposites were counterbalanced across stimuli. Composites occupied counterbalanced positions slightly to the left or to the right of the centre of each slide, with the left and right positions being those in which parts of the non composites could occur. Figure 3 shows the original faces used in the experiment, and examples of composites and noncomposites made from each. Original
Composite
Noncomposite
The five familiar faces used in experiment 1 ( 'original' ), examples of composite stimuli created from these, and the equivalent noncomposites.
Figure 3.
Faces in Perception [1170] Facial configuration
751
The experiment began with a warm-up session, in which subjects were shown the original photographs from which the composites and noncomposites had been constructed. They were asked to name each person, and reminded of any names they had difficulty remembering. In the experiment proper, each of the twenty composites and twenty noncomposites appeared in random order as back-projected slides, with the composites occupying a horizontal visual angle of approximately 5 deg. Two blocks of trials were run. In one block subjects were asked to name the top segment from each stimulus as quickly as possible, and in the other block to name the bottom segment. The order of presentation of these blocks of trials was counterbalanced across subjects. Each block was preceded by appropriate practice trials in which either top or bottom segments of faces were presented in isolation, and subjects were asked to name each person. Naming responses were made by giving the person's surname (Bosanquet, Bygraves, Scott, Snowdon, or Wogan), and vocal reaction times (timed from stimulus onset) were recorded via a voice key. Thus this is a within-subjects design in which naming latencies were recorded for the top segments of twenty composite and twenty noncomposite stimuli, and for the bottom segments of twenty composite and twenty noncomposite stimuli. The responses (ie names) required were the same for all cells of the design, and the same face segments were presented in the form of composite and noncomposite stimuli. 2.2 Results Mean reaction times for correct identification of the top and bottom halves of composite and noncomposite stimuli are shown in table 1. Error rates were low (less than 4% for all cells of the design) and will not be considered further. A two-factor analysis of variance (ANOVA) of the reaction times for correct identi fication was carried out, to determine the effects of task (top-half or bottom-half naming; repeated measure) and stimulus type (composite or noncomposite; repeated measure). This revealed a main effect of stimulus type (Fl. 11 = 27.63, p < 0.001), with faster responses to noncomposites than to composites. A task X stimulus type inter action was also found (Fl. 11 = 6.62, p < 0.05). Tukey tests ( a = 0.05) showed that responses were faster to noncomposite than to composite stimuli for both top-half and bottom-half identification, but that responses to the noncomposites were slower when the bottom half had to be identified (Tukey critical difference = 66 ms). A comparable slowing of bottom-half identifications was not found for the composite stimuli. Mean vocal reaction times (in ms) for correct identification of the top and bottom halves of upright composite and noncomposite stimuli (experiment 1 ) .
Table 1.
Top half Bottom half
Composite
Noncomposite
1297 1282
1041 1123
2.3 Discussion Reaction times were substantially slower to composite than to noncomposite stimuli, indicating that the perception of a novel facial configuration in the composites interfered with the identification of the constituent parts. Although this effect was more marked when subjects were identifying top halves (task x stimulus type interaction), a difference between composite and noncomposite arrangements was found both for top half and for bottom-half face identification. This finding was made despite the fact that the task is apparently only one of discriminating five top-half faces or five bottom-half faces, and despite the fact that no steps were taken to prevent subjects concentrating on
Faces in Perception [1171] A W Young, D Hellawell, D C Hay
752
particular cues (such as hairstyle or mouth shape} to effect these discriminations. Thus it clearly reflects a compelling perceptual phenomenon. 3 Experiment
2
A curious property of faces is that they become difficult to recognise when they are inverted (Yin 1969; Valentine and Bruce 1986}. This does not simply reflect the fact that we usually see upright faces, because the effect of inversion is known to be greater for faces than for many other stimuli that we are also used to seeing in a particular orientation (Yin 1969; Scapinello and Yarmey 1970}. It has been hypothesised that inversion interferes with encoding of the facial configuration (Carey and Diamond 1977}. Our facial composites seemed to us to support this idea. If figure 1 or figure 2 is viewed upside down it is immediately apparent that the top and bottom halves of the composite faces no longer fuse so effectively. In order to provide formal confirmation of this observation we carried out a second experiment, in which the stimuli from experiment 1 were presented in upright or inverted orientation. 3.1 Method 3.1.1 Subjects. Six male and six female subjects were selected in the same manner as for experiment 1. 3.1.2 Stimuli and procedure. Stimuli were the same as those used in experiment 1. However, they were presented upright (as in experiment 1} for half the trials, and inverted for the other half. Trials with upright or with inverted stimuli were blocked together, and the order of these blocks was counterbalanced across subjects. Subjects were always asked to identify the segment of the face containing the eyes and hairline (ie, the 'top' segment in the upright faces). All other details of design and procedure were as for experiment 1. Thus this is a within-subjects design in which latencies for naming the face segment containing the eyes and hairline were recorded from twenty upright composite stimuli, twenty upright noncomposites, twenty inverted composites, and twenty inverted noncomposites. 3.2 Results Error rates were very low (less than 1% for all cells of the design}, and will not be considered further. Mean reaction times for correct responses to upright and inverted composites and noncomposites are shown in table 2. A two-factor ANOVA of the reaction times for correct identification was carried out, to determine the effects of orientation (upright or inverted; repeated measure) and stimulus type (composite or noncomposite; repeated measure}. This revealed a main effect of stimulus type (F1 11 = 54.90, p < 0.001}, with faster responses to non composites than to composites, and an orientation x stimulus type interaction (F1 11 = 46.01, p < 0.001}. Tukey tests ( a 0.05} were used to examine the ori�ntation x stimulus type interaction. These showed that reaction times were slower for composite than noncomposite stimuli otily in the upright condition; for inverted stimuli there was no difference between composites and noncomposites (Tukey critical =
Mean vocal reaction times (in ms) for correct identification of the part of the upright and inverted composite and noncomposite stimuli containing the hairline and eyes (experi ment 2).
Table 2.
Upright Inverted
Composite
Noncomposite
1167 1039
955 1030
Faces in Perception [1172] Facial configuration
753
difference = 64 ms). For the noncomposite stimuli, responses were faster when they were upright than when inverted, but for composites reaction times for inverted stimuli were actually faster than for upright stimuli! 3.3 Discussion The upright condition replicates the finding from experiment 1 that perception of the facial composites interferes with the identification of their constituent parts (ie, slower responses to upright composites than to upright noncomposites). This effect is completely eliminated by inversion, with no difference between reaction times to composites and noncomposites in the inverted condition. In fact, inversion actually made the task easier for the composite stimuli. This finding of an inverted-face superiority provides a dramatic illustration of the absence of interference from configurational information in the inverted composites. 4
Experiment 3
In experiments 1 and 2 we investigated the perception of composites made from parts of familiar faces. We did this largely because it is convenient to use familiar faces if reaction time for identification is to be measured. There are, however, several demonstrations in the literature of differences between the processing of familiar and unfamiliar faces (Bruce and Young 1986; Ellis 1986). Thus it seemed conceivable (though in our view unlikely) that our findings might not apply to face perception in general, but only to the perception of familiar faces. This might, for instance, happen if subjects cannot avoid trying to identify part of a familiar face by referring to the appearance of the whole face; composites might then interfere more with this process than noncomposites. Experiment 3 was thus designed to use faces of unfamiliar people. Of course, these have to become familiar to some extent for an identification to be made, but what we did was to teach subjects only the top halves in isolation. These were then combined with the bottom halves from a different set of faces to form composite and non composite stimuli. Thus we ensured that subjects did not know what each person should look like, as they had only been trained on the top halves. 4.1 Method 4.1.1 Subjects. Six male and six female subjects were selected in the same way as for experiment 1. 4.1.2 Stimuli and procedure. Black and white press photographs showing full-face poses of the faces of unfamiliar people were rephotographed to a standard size and divided into top and bottom segments by cutting along a horizontal line below the eyes. Five top segments that were considered by the authors to show a range of hairstyles comparable to the range used in the familiar faces of experiment 1 were then selected. Five bottom segments that matched the expressions used in the familiar faces of experiment 1 were also selected. These top and bottom segments all came from the faces of different people. Slides showing twenty composite and twenty noncomposite stimuli were then prepared from these top and bottom segments, in a similar manner to that used for experiment 1. For the experiment, subjects were asked to name the top segment of each stimulus. Five surnames were chosen from the Lancaster, Morecambe, and Heysham 1984 Telephone Directory, and one of these surnames was assigned at random to each of the top segments. The surnames were chosen to have the same mean frequency of occur rence in the telephone directory as the surnames used in experiment 1, to have the same initial letters, and to have the same number of letters.
-
Faces in Perception [1173] 754
A W Young, D Hellawell, D C Hay
The experiment began with a warm-up session, in which subjects learnt the surnames by going through the top segments presented in isolation at least six times. The twenty composite and twenty noncomposite stimuli were then presented one at a time in random order, for subjects to name the top segment of each as quickly and accurately as possible. All other features of design and procedure were as for experiment 1. In experiment 3, then, subjects named the top segments of composites made from parts of the faces of unknown people. It is important to note that they learnt arbitrarily assigned surnames to the top segments presented in isolation, and that they never saw the actual bottom segment of any of the faces whose top segments were to be named, since the bottom segments used to form the composite and noncomposite stimuli came from different people. 4.2 Results Mean reaction times for correct responses were 1175 ms for composites, and 1092 ms for noncomposites. Error rates were less than 5% in each case. A one-factor ANOVA, used to determine the effect of stimulus type (composite or noncomposite; repeated measure), showed that responses to composites were slower than to noncomposites (F1, 11 24.31, p < 0.01). =
4.3 Discussion The finding of slower responses to composites than to noncomposites in identifying parts of faces only learnt as fragments again confirms that it is the perception of a novel facial configuration in the composites that underlies the effect. The size of the differ ence between reaction times to composites and noncomposites is, however, somewhat smaller than that found in experiments 1 and 2. The importance of this is difficult to assess because small sets of stimuli were used in the experiments, and thus any effect of minor variations in the stimuli themselves would tend to be exaggerated. The main point is that a difference between reaction times to upright composites and non composites was found in all three experiments. 5
Experiment
4
In experiments 1 to 3, composites were made by joining together face fragments separated along an arbitrary line drawn just below the eyes. The purpose of experi ment 4 was to investigate the creation of composites from fragments that are psychologically more real. Recent studies have emphasised differences between the processing of internal (eyes, nose, mouth) and external (hair, face shape, chin) features of familiar and unfamiliar faces, and established the differential salience of the internal features of familiar faces (Ellis et al 1979; Endo et al 1984; Young et al 1985). We were thus interested in whether or not combinations of incorrect internal and external features would create novel perceptual configurations, and in whether there would be more interference from incorrect internal features on the recognition of external features or vice versa. 5.1 Method 5.1.1 Subjects. Six male and six female subjects were selected in the same manner as for experiment 1. 5.1.2 Stimuli and procedure. The faces of four pairs of famous people were used as stimuli (David Frost and Paul McCartney; Mick Jagger and John Kennedy; Barry Manilow and Sting; David Owen and Terry Wogan). These were selected so that the internal and external features could be easily interchanged to form composite faces. A within-subjects design was used, with two conditions. In one condition the subjects were asked to identify internal features (eyes, nose, and mouth) of the stimulus, and in
Faces in Perception [1174] Facial configuration
755
the other condition they were asked to identify external features (hair, face shape, and chin). Each condition was divided into two phases; a pretest phase and a test phase. In the pretest phase, subjects were given practice at identifying the eight famous people from their internal or external features only (according to the appropriate condition). Each of the eight people was presented four times in each pretest session, in a random order. In the test phase the subjects were again asked to identify the eight famous people from their internal or from their external features. Vocal reaction times (measured from stimulus onset) and accuracies were recorded. In the internal-feature naming condition, each target person's internal features occurred three times in each of the four following ways: (i) Whole face: internal features combined with the correct external features. (ii) Part face: internal features only. (iii) Noncomposite: internal features combined with the inverted external features of the other member of the pair. (iv) Composite: internal features combined with the external features of the other member of the pair. Four corresponding types of stimuli were used for the external-feature naming condition: (i) Whole face: external features combined with the correct internal features. (ii) Part face: external features only. (iii) Noncomposite: external features combined with the inverted internal features of the other member of the pair. (iv) Composite: external features combined with the internal features of the other member of the pair. Examples of stimuli of each type are shown in figure 4. Half the subjects did the internal-feature naming condition first, and half did the external-feature naming condition first. Thus this is a within-subjects design in which naming latencies were recorded for the internal features of twenty-four whole faces, twenty-four part faces, twenty-four non composites, and twenty-four composites, and for the external features of twenty-four whole faces, twenty-four part faces, twenty-four noncomposites, and twenty-four composites. 5.2 Results Mean reaction times for correct identification of internal and external features from whole faces, part faces, noncomposites, and composites, are shown in table 3. Error rates were less than 6% for all cells. A two-factor ANOVA of the reaction times for correct identification was carried out, to determine the effects of task (internal-feature or external-feature naming; repeated measure) and stimulus type (whole face, part face, noncomposite, or composite; repeated measure). The only significant effect was that of stimulus type (F3,33 93.32, p < 0.001). There was no effect of task (F < 1), and the task x stimulus type inter action did not approach significance (f;, 33 1.67, p > 0.1). Tukey tests (a 0.05) =
=
=
Mean vocal reaction times (in ms) for correct identification of the internal or external features from whole faces, part faces, noncomposites, and composites (experiment 4 ).
Table 3.
Internal features External features Overall
W hole face
Part face
Noncomposite Composite
951 959
984 996
1034 1045
1213 1168
955
990
1040
1190
Faces in Perception [1175] A W Young, D Hellawell, D C Hay
7513
were used to examine the effect of stimulus type. These showed that all means were significantly different from each other except those for whole faces and part faces ( Tukey critical difference 44 ms ). =
5.3 Discussion In experiments 1 and 3 we were able to create noncomposites by misaligning the top and bottom halves of the faces. This procedure was not feasible with internal and external feature combinations, so instead we produced noncomposites in which one part of the face was inverted, and also included a condition in which only part of the face was presented. Again, reaction times to composites were slower than to noncomposites, showing interference from the composite configuration onto identification of the individual Internal-feature naming task
External-feature naming task
Whole face
Part face
Noncomposite
Composite
Examples of stimuli used in experiment 4. The correct ·response to all stimuli in the first and third columns is "Mick Jagger", and to stimuli in the second and fourth columns " John Kennedy".
Figure 4.
Faces in Perception [1176] Facial configuration
757
features. In experiment 4 a condition was also included which involved composites formed by rejoining the correct face fragments (whole face); significant facilitation was found for this condition in comparison to the noncomposites, but the facilitation in comparison to the part faces did not reach statistical significance on the somewhat conservative Tukey test. This pattern was found both for internal-feature naming and for external-feature naming, with no evidence of greater interference for one condition than for the other. 6
General discussion
Our facial composites technique clearly taps a compelling phenomenon. With the method used in experiments 1 to 3 the only difference between composite and non composite stimuli was the presence of a facial configuration in the composites. We have thus shown that the perception of a novel configuration in facial composites interferes with recognition of their constituent parts for top-half and bottom-half face composites (experiments 1 to 3) and for internal and external facial feature composites (experi ment 4). Even when the stimuli had only ever been previously seen as top-half face fragments, interference was still found when composites were formed by joining these to bottom-half face fragments (experiment 3). Configurational information, then, is very important to face perception. To some extent this may be considered an example of a more general finding that under certain conditions global stimulus properties take precedence in visual perception (Navon 1977; Martin 1979; Pomerantz 1983) and Humphreys et al (1985) review several studies that examine how global and local form descriptions are combined with each other. Experiment 2, however, demonstrated that interference from the configurational properties of composite faces is only found for upright composites; for inverted stimuli there was no difference between composites and noncomposites. This finding bears out Carey and Diamond's (1977) claim that the facial configuration is not perceived in inverted faces, and we note that Sergent (1984) also found that feature differences were more important than configurational differences when people were asked to match inverted faces. This was known to Arcimboldo, who created 'reversible' paintings such as The Cook (ca 1750) or The Vegetable Gardener (ca 1590), in which a face-like appearance was only evident in one orientation of the picture (see Hulten 1987). Our own results show that it is actually easier to identify the constituent parts of facial composites if they are inverted than if they are upright; the world's first inverted-face superiority. The absence of interference from inverted composites shows, then, that a meaningful configuration is not necessarily perceived in any complex visual stimulus with an unbroken boundary. This raises the question as to the extent to which the use of configurational information is specific to face perception (or, rather, specific to the perception of upright faces). The widely reported susceptibility of faces to performance impairments on inversion has in the past been used to support claims of face specificity (Y in 1969, 1970; Carey and Diamond 1977; see Ellis and Young 1988 for a review). It is now clear, however, that comparable inversion effects can be found for other stimulus classes than faces provided that their members share similar appearances and that the perceiver is expert at differentiating them (Diamond and Carey 1986). This suggests that the importance of configurational information to face perception reflects our very great expertise at discriminating between the otherwise very similar faces of people we meet. Our findings, then, demonstrate directly the importance of configurational informa tion to face perception, and that this configurational information is only readily derived from an upright face. In emphasising the importance of configurational information we
Faces in Perception [1177] 758
A W
Young,
D Hellawell, D C Hay
do not, of course, seek to deny that individual facial features can also be important. It is not, for instance, impossible to see that the features in the composite faces shown in figure 1 are those of Margaret Thatcher and Shirley Williams. Our point is only that it is not at first easy to see this in the presence of a novel configuration. Configurational and featural information are, however, both likely to contribute to normal face recognition ( Sergent 1984; Bruce and Young 1986; Ellis 1986). Indeed, we all know that some people possess striking individual features that make them easy to identify ( eg Denis Healey's eyebrows ). An important task for future research will thus be to specify how configurational and featural cues interact in the recognition of familiar faces. We are grateful to the Press Association for help in finding suitable photo graphs for use as stimuli, and for permission to reproduce those used in our illustrations.
Acknowledgement.
References
Baddeley A D, 1979 "Applied cognitive and cognitive applied psychology: the case of face recognition" in Perspectives on Memory Research ed. L G Nilsson (Hillsdale, NJ: Lawrence Erlbaum Associates) pp 367-388 Bruce V, Young A W, 1986 "Understanding face recognition" British Journal of Psychology 77 305-327 Calis G, Sterenborg J, Maarse F, 1984 "Initial microgenetic steps in single-glance face recog nition" Acta Psychologica55 215-230 Carey S, Diamond R, 1977 "From piecemeal to configurational representation of faces" Science 195 312-314 Damasio A R, Damasio H, van Hoesen G W, 1982 "Prosopagnosia: anatomic basis and behavioral mechanisms" Neurology 32 3 31-341 Davies G M, 1982 " Composite systems for recalling faces-'helping the police with their enquiries?' " Reconstructing the Past: the Role of Psychologists in Criminal Trials ed. A Trankell (Stockholm: Norstedts) pp 299-313 Diamond R, Carey S, 1986 " W hy faces are and are not special: an effect of expertise" Journal of Experimental Psychology: Generalll5 107-117 Ellis H D, 1975 "Recognizing faces" British Journal of Psychology 66 409-426 Ellis H D, 1981 " Theoretical aspects of face recognition" in Perceiving and Remembering Faces eds G M Davies, H D Ellis, J W Shepherd (London: AcademicPress) pp 171-197 Ellis H D, 1986 "Processes underlying face recognition" in The Neuropsychology of Face Perception and Facial Expression ed. R Bruyer (Hillsdale, NJ: LawrenceErlbaum Associates) pp 1-27 Ellis H D, Shepherd J W, Davies G M, 1979 "Identification of familiar and unfamiliar faces from internal and external features: some implications for theories of face recognition" Perception 8 431-439 Ellis H D, Young A W, 1988 "Are faces special?" in Handbook of Research on Face Processing eds A W Young, H D Ellis (Amsterdam: North-Holland) Endo M, Takahashi K, Maruyama K, 1984 "Effects of observer's attitude on the familiarity of faces: using the difference in cue value between central and peripheral facial elements as an index of familiarity" Tohoku Psychologica Folia 43 23-34 Galton F, 1879 " Composite portraits, made by combining those of many different persons into a single resultant figure" Journal of theAnthropological Institute 8 132-144 Galton F, 1883 Inquiries into Human Faculty and Its Development (London: Macmillan) Haig N D, 1984 " The effect of feature displacement on face recognition" Perception 1 3 505-512 Hulten P (ed.), 1987 TheArcimboldo Effect (London: Thames and Hudson) Humphreys G W, Riddoch M J, Quinlan P T, 1985 "Interactive processes in perceptual organisation: evidence from visual agnosia" in Attention and Performance eds M I Posner, 0 S M Marin (Hillsdale, NJ: LawrenceErlbaumAssociates) pp 301-318 Kaufmann T Da C, 1987 " The allegories and their meaning" in The Arcimboldo Effect ed. P Hulten (London: Thames and Hudson) pp 89-108 Martin M, 1979 " Local and global processing: the role of sparsity" Memory and Cognition 7 476-484 Navon D, 1977 "Forest before trees: the precedence of global features in visual processing" Cognitive Psychology 9 353-383
Okabe A, 1987 "Kao at the dawn of modernity" in Thames and Hudson) pp 235-239
The Arcimboldo Effected. P
Hulten (London:
--.
Faces in Perception [1178] 759
Facial configuration
Pomerantz J R,
1983
"Global and local precedence: selective attention in form and motion
perception" Journal of ExperimentalPsychology: Generallll Scapinello K F, Yarmey AD,
1970
516-540
"The role of faniiliarity and orientation in immediate and
delayed recognition of pictorial stimuli" Psychonomic Science 24 329-330
Sergent J,
1984
"An investigation into component and configural processes underlying face
perception" British Journal ofPsychology 75 Shepherd J,
1988
221-242
"The face and social attribution" in Handbook of Research on Face Processing
eds A W Young, HD Ellis (Amsterdam: North-Holland) Valentine T, Bruce V,
1986
"The effect of race, inversion and encoding activity upon face
61259-273 K, 1969 "Looking at upside-down faces" Journal ofExperimentalPsychology 81 141-145 Y in R K, 1970 "Face recognition by brain-injured patients: a dissociable ability?" Neuro psychologia 8 395-402 Young A W, HayDC, McWeeny K H, F lude B M, Ellis A W, 1985 "Matching familiar and unfamiliar faces on .internal and external features" Perception 14 737-746 recognition" ActaPsychologica
Y in R
© 1987 a Pion publication printed in Great Britain
