The Man Behind the Mask-Disguise and Face ...

HUMAN LEARNING, VOL.

The Man Behind the

3, 83_95 (1984)

Mask-Disguise and Face Recognition

GRAHAM DAVIES AND RHONA FLIN Department of Psychology, Uniuersity of Aberdeen, Old Aberdeen, Scotlanil, AB9 2UB

SUMMARY Three experiments are described which explore the effects upon subsequent recognition of disguising a face with a nylon stocking mask. Experiment 1 established that the wearing ol such masks had a substantial deleterious effect upon subsequent identification and that the effects ofa plain mask were just as great as for a patterened stocking. Instructions designed to improve the initial encoding of the face were ineffective in reducing the masking effect. Experiment 2 examined whether the effect arose from the distortion of the facial features or from the elimination of complexion cues and high frequency information. It was concluded that feature distortion was the most powerful factor. Experiment 3 considered whether identification of a masked face might be facilitated if the faces composing the recognition array were also masked. The latter condition produced significantly better recognition rates than the conventional unmasked conditions. The practical implications of these findings for police procedures for identifying masked criminals are discussed.

With the widespread

of security cameras in banks and public places, criminals are

use

more frequently resorting to the use of disguise to avoid subsequent identiflcation. One of the simplest and most common forms of disguise is the wearing of a nylon stocking pulled down over the head in such a way as to obscure the facial features. This paper

examines experimentally the impact of this form of disguise upon subsequent identification, attempts to pinpoint the source of the masking effect and considers various ways in which the impact of such a disguise may be ameliorated. The effects of disguise on voice recognition have received some attention (see Clifford, 1983 for a recent review) but few experiments have investigated the effects of disguise on person identification. Patterson and Baddeley (1977) investigated the effects of the presence or absence of wigs, beards and spectacles on the ability of subjects to identify a number of male faces in a sequence of alternatives. Approximately 89 per cent of the faces were recognized when no attempt was made at disguise. This figure was, however, reduced to 67 per cent when targets added or removed a beard and to 64 per cent for changes to the hair. The effect of spectacles on identification was minimal unless combined with a major change in lacial pose from study to test. A similar ordering for the effects of these three forms of disguise was reported by Laughery and Fowler (1917), though on this occasion spectacles, too, significantly reduced overall accuracy of recognition. Both of these studies employed adult subjects. Carey and Diamon d (1971) reported that the ability of young children to recognize the faces of strangers was disproportionately affected by forms of disguise such as wigs, hats, spectacles or scarves. Unfortunately, no. information on the relative effectiveness of these different forms of 027 7 -67 07 I 84/02008 3-1

O

1984 by John Wiley

3

$01. 30

& Sons, Ltd.

Receiued 5 September 1983

Retised 22 Notember 1983

84

Graham Dauies and Rhona Flin

disguise were provided, even in the extended account of their findings (Diamond and Carey, 1977). However, overall accuracy of identification in a two-alternative, forcedchoice recognition task improved from chance at six years ofage to nearly 90 per cent at age 16. In contrast to the selective changes in facial features achieved by accessories such as wigs and spectacles, the nylon mask appears to inflict a global degradation on the face as a whole. The soft tissue areas of the face are flattened, hair compressed, and complexion cues and fine detail are eliminated. Under these circumstances, one might expect to find the effects of a mask disguise to be as great, if not greater, than that achieved by such accessories. Experiment 1 explores the magnitude of the mask effect and examines whether it may be ameliorated through simple instructions designed to improve the quality of the initial encoding of the face.

EXPERIMENT I Considerable attention was given to how best to present the masked face. Previous studies had all employed still photographs, an approach which may be criticized for lack of ecological realism. However, ethical considerations precluded masked actors confronting unsuspecting subjects. In the end a video presentation was employed, together with instructions to subjects to role-play a witness scenario. Such an approach seemed to offer a reasonable compromise between realism and control. A second consideration concerned the choice of stocking for the mask. Nylon stockings are typically patterned or plain mesh. There is some evidence that the superimposition of a pattern upon a face disrupts the normal recognition process (Ellis, Davies and Shepherd, 1978). It was decided to include patterned and plain mesh masks

as a factor in the design to ascertain whether patterning had any effects upon identification over and above those due to the mask as such. Finally, it seemed important to establish how, if at all, any masking effect might be reduced. Patterson and Baddeley (1971) reported that instructions to make attributional judgements on faces disguised by accessories led to a signiflcant improvement in recognition for a set of 6 targets, though the effect was not maintained when the target set size was increased. Zavala and Paley (1972), on the other hand, have emphasized the value of searching faces for distinctive features as an aid to later identification (see also Winograd, 1981). A condition was therefore included in the design which involved a brief training session, based on these two techniques, to see if this would minimize any masking effect. Method Subjects and design A total of 138 male and female students, aged between l8 to 23 years volunteered for this study. They were arbitrarily assigned to one of six groups. These corresponded to the cells of a 3 x 2factorialdesign which combined three levels of stimulus presentation

with two of instructions. The conditions of presentation were (1) control, where the targets appeared in normal appearance, (11) plain mask, where they wore a plain stocking pulled over their heads, and (il1) patterned mask, where a patterned stocking

Disguise and Face

Recognition

85

replaced the plain one. In addition, half the subjects in each of the presentation conditions received standard instructions to remember whereas the remainder received special instructions aimed at improving their face memory skills. Materials Three video films were specially prepared for this study. All three showed the same four male actors sitting at a table playing cards. The actors were student volunteers aged between 19 and 23 years and were selected on the basis of their being unknown to the subject population and devoid of exceptional physical peculiarities of appearance. All were clean shaven and wore no spectacles during filming.

The format of each of the three fllms was identical. The film opened with an establishing shot lasting 30 s which showed the actors sitting at the card table. The camera then showed each actor's face in close-up in turn for 30 s. During the close-up shots the actors were instructed to change their pose so that they could be seen in fullface and profile and to show appropriate affect. Following the close-up shots, the camera tracked back to show the complete scene again, the whole fllm lasting some 2 min 45 s. In the control version, the actors appeared without disguise. \n the plain rmsk version, each actor wore a plain 40 denier tan stocking pulled down over his head in such a way as to completely cover his face. In the patterned mcsk version, the plain stocking masks were replaced by stockings of a patterned'fishnet'design (see Figure 1). The set of slides for the identification task consisted of 24 pairs of coloured slides, each pair consisting of a front and profile view of the face of different young men. All four targets appeared within the set at positions 4, ll, 14 and 20 ofthe 24 face sequence. A11 members

of the recognition array were photographed wearing identical

green

surgical gowns to obscure any clothing cues. Procedure

Subjects in each of the three encoding conditions followed the same procedure at presentation. They were conducted as a group to the experimental room and seated from 1 to 3 m from a large colour television monitor. Subjects in the standard instruction gr-dtiD were given the following instructions for role-playing. They were told to imagine that a bank robbery had taken place in their neighbourhood. As they are returning from their class, they happen to look into a window and see four men corresponding to the descriptions ofthe robbers issued by the police. Their task is to try to remember these men so that they would be able to recognize them later. These same instructions were also given to the Special Instructions condition but in addition they received information designed to assist in remembering the target faces accurately. The Special Instruction subjects were told to look at each ofthe target faces in turn and, first, to try to decide for which occupational group the person was most suited and, second, to scan the face and decide which feature was most distinctive and idiosyncratic. Subjects were warned that although this might not be their normal procedure for remembering faces they were to try to employ it on this occasion. Still photographs of three older males were projected in turn to give some practice in the use of the technique, the whole instructional sequence lasting some 7 to 10 min. Following completion of the practice trials, subjects viewed the videotape for the appropriate encoding conditions. After the tape had been shown, subjects in all conditions were told that they would now have to try and recognize the faces of the persons shown in the film. They would see

86

Graham Daoies and Rhona Flin

Figure 1. Examples of conditions employed in Experiment 1: (top) control; (bottom left) patterned mask; (bottom right) plain mask

front and profile views of a number of men and for each they must indicate on the protocol provided, whether he was one of the 'robbers' shown in the film. The pairs of slides were then projected life-size onto a white screen at the front of the room via two Kodak'Carousel'projectors whose shutters had been electronically linked to permit simultaneous projection. Each pair of slides was shown on the screen for 6 s with a 2 s inter-slide interval,'the full face view always appearing on the left and the profile on the

Disguise and Face

Recognition

87

right. After all the slides had been shown, the protocols were collected and the subjects debriefed and thanked for their assistance. Results and discussion Each protocol was scored for hits and false alarms. These were then converted into the measure of accuracy d', derived from signal detection theory (see Banks, 1970). Mean hit, false alarm and d' scores as a function of condition are shown in Table 1. Separate analyses were conducted on all three measures of performance but as the pattern was the same, only the results of d' will be reported. A3 x 2 analysis of variance showed a single significant effect, that for encoding condition, F (2,L32) 58.73, p < 0.001. Inspection of the differences between the means using the Newman-Keuls test indicated that the no-mask control faces were more accurately recognized than either of the two masking conditions (p < 0.001). The effects of the patterned mask

:

were no more disruptive than the plain and both had a substantial deleterious effect upon identification accuracy. Indeed, the performance of subjects in the plain mask condition indicated chance performance whereas that of the patterned mask subjects was only just above chance. In order to compare the results of the current experiment with those from earlier disguise studies which used accessories, the mean percentage hit rate was computed for each condition. Direct comparison is hampered by differences in the absolute levels of performance in the two experiments. [n the current study which employed fewer targets but a living and mobile facial image, 63 per cent of the targets were identified on average under normal non-masked conditions. The wearing of a patterned mask reduced this figure to 40 per cent and the plain mask to some 29 per cent. Clearly, whether measured absolutely or relatively, nylon masks are an effective means of reducing the probability

of identification.

lt is also evident that the instructions to make attributional and distinctive feature judgements were ineffective in reducing the magnitude of the effect. Brief training of the kind employed in this study was insufficient to eliminate the gross disruption to the normal processes of face recognition induced by masks. It might be that a more prolonged training programme along the current lines could be of greater value, but the history of face memory training courses, whether based on the distinctive feature Table 1. Mean hits, false alarms and d' scores as a function of encoding condition and instructions: Experiment

1

Standard instructions

Hits

2.70

1.22

t.96

False alarms

t.l7

d'

2.34

4.17 0.13

4.13 0.80

2.30 0.52 2.27

1.13

1.26

2.6s 0.40

3.78 0.46

Special instructions

Hits False alarms

d'

88


approach (Woodhead, Baddeley and Simmonds, 1979) or the use of attributional judgements (Malpass, 1981) provides little basis for optimism. Given that nylon masks appear an effective method of disguise, how do they interfere with the normal processes of face encodipg? Knowledge as to the basis of the effect might provide useful clues as to how best to counter their impact. This issue is explored in Experiment 2.

EXPERIMENT

2

Observing the face in a nylon mask suggests two hypotheses, not necessarily mutually exclusive, concerning the way in which face encoding might be degraded. The first hypothesis concerns the distortion inflicted on individual features within the face. The soft tissue areas, particularly the nose and ears, are flattened and distorted and the hair is pressed to the scalp. Although the nose and ears do not emerge as particularly salient cues for identification purposes in published studies, hair and face shape frequently figure among the most important (see Shepherd, Davies and Ellis, 1981). Loss of such primary sources of face information might be sufficient to reduce recognition rates to near chance levels. A second hypothesis centres around the filtering effects of the nylon mesh. Studies have suggested that face recognition may be particularly sensitive to certain frequencies of information encoded from the human face (e.g. Tieger and Ganz, 1979). As every aging cinema actress knows, nylon gauze between subject and camera is particularly effective in eliminating the high frequency information concerning age lines and details of complexion. The wearing of a stocking might be expected to have a similar effect and the colouring of the stocking material might further reduce complexion cues. If such high frequency information does play a role in face recognition (see Ellis (1981) for a discussion of this approach), then this factor too might contribute to the masking effect. In order to partial out the effects of filtering from those of feature distortion it was necessary to introduce conditions in which, first, filtering was present without distortion and, secondly, distortion occurred in the absence of f,ltering. The first was achieved by photographing the unmasked targets through nylon material whereas the second involved persuading targets to be filmed wearing tight-fitting but ventilated polythene bags over their heads. Consideration was again given as to whether there should be an attempt at mundane realism in the presentation of the targets. Given that the study sought to discriminate

between rival theoretical interpretations of masking rather than to establish its magnitude under field conditions, it was decided to opt for orthodox experimental procedures in this and the subsequent experiment.

Method Subjects and design A total of 48 male and female students, age d l8-22years volunteered as subjects for this study. They were randomly assigned to one offour groups of 12 subjects. These groups constituted the cells of a 2x2 factoial design which contrasted the presence and absence of filtering and distortion during the initial observation of a series of target

Disguise and Face Recognition

89

four conditions: (i) no filtering, no distortion (NN), (ii) no flltering, distortion (ND), (iii) filtering, no distortion (FN) and (iv) filtering and distortion (FD). Accuracy of identification was tested for all groups by a series of arrays composed of faces without masks. faces. There were thus

Materials and targets Four male students aged 18-19 years agreed to act as targets. All were clean shaven and none wore spectacles. Each was photographed, first for the NN condition: a head and shoulders shot in three quarter pose, looking slightly away from the camera with a neutral expression. A second photograph was taken for the FD condition, this was in an identical pose but with the target wearing a plain nylon mask made from a 40 denier tan stocking. A third photograph was taken for the ND condition in which the target wore a plain polythene bag drawn tightly over the facial features. For a further photograph, for the FN condition, a nylon stocking was stretched over a picture frame and the frame then interposed between the camera and the target's face. In this way four sets of stimulus material were constructed, each set corresponding to one of the conditions of initial observation. Each set consisted of photographs of the same four target persons. All photographs were standard 9 cm x 7 cm 'polaroid' colour prints. A separate recognition array was constructed for each ofthe four targets. Each array was made up of seven photographs of men, of whom one was a target and the remainder

were distractors of similar build and physical appearance. The pictures were all small 4 cm x 5 cm monochrome prints of the kind dispensed by instant photobooths and showed each person full face in close-up against a neutral background. Thus, there were no contextual, pose or clothing cues which could identify the target; identification by the face was essential. Procedure Subjects were approached individually and invited to participate in a short experiment. They were told they would be shown four photographs of young men whom they would later have to recognize. The subject then examined each of the faces in turn drawn from the appropriate condition; 30 s was allowed for observation of each face and a further two minutes elapsed before undertaking the recognition task. The subjects were then handed the first ofthe recognition arrays with the request that they select from the array the face which they thought was the target. Subjects were allowed unlimited time to make their selections. The procedure was then repeated for the remaining three arrays, the number of any photographs chosen being noted in each instance. The position of the target in the array and the order in which the arrays were presented varied randomly across subjects.

Results and discussion

The mean number of targets correctly identified by each subject as a function of encoding condition is shown in Table 2. On this occasion, mean scores in all four conditions including the plain mask were significantly greater than those expected by chance (0.57).

As the identiflcation scores fulfilled the criteria for homogeneity of variance, they were analysed by a 2 x 2 analysis of variance, using filtering and distortion as main effects. Distortion significantly reduced overall rates of identification, F (1,44) : 55.17,

90

Graham Dauies and Rhona Flin Table 2. Mean hits (and standard deviations) as a function of condition of presentation: Experiment 2

Distortion Present

FD

Filtering

r.33 (0.78) ND l.1s (0.86)

Present

Absent

Absent

FN 3.50 (0.e0)

NN 3.25 (0.87)

FD is the nylon mask condition; FN is the nylon screen; ND, the polythene bag, and NN, the control.

p < 0.001, but there was no effect of filtering, F

< 1, nor any significant interaction F (1,44) : 1.82, NS. Thus, the main effect of the stocking mask appears to be in the distortion offeatures ofthe face which render them less recognizable when seen later under normal conditions. An equivalent reduction in recognition accuracy was achieved by stretching a clear polythene bag tightly over the facial features. This latter result, combined with the negative findings of the FN condition, suggests that loss of high frequency and skin tone information makes a negligible contribution to the task effect. However, some caution is required before entirely discounting this factor. The current study used young Caucasian males exclusively for targets and distractors. Loss of skin tone and complexion cues might have played a more critical role in reducing identification if coloured or older persons had been employed as targets. Nevertheless, even allowing for these circumstances, the loss of feature information appears to account for the major element in the masking effect. Given that the normally proficient mechanisms of identification are disrupted by masking, how best can its effects be countered? The inability of subjects to identify the masked face could be due to one of two factors. First, subjects may simply fail to encode any useful facial information and thus be totally unable to identify the face concerned. A second, more optimistic, view would be that subjects do encode information from the face but it is of a type which is inappropriate for identifying the normal unmasked face. If the latter hypothesis were correct then presenting the recognition array in a masked rather than a normal state might conceivably yield higher rates of recognition. This

possibility is explored in Experiment

3.

EXPERIMENT 3 There is considerable evidence from verbal learning studies for an encoding specificity theory of memory: probability of retrieval improves to the degree to which the conditions of encoding are reinstated at the time of recall (Tulving and Thomson, 1973; Eich, 1980; Baddeley, 1982). Davies (1983)has demonstrated a similar phenomenon in relation to faceq. Line drawings and photographs of faces are equally well recognized

Disguise and Face

Recognition

9l

but a change from one mode ofrepresentation to the other from study to test produces fall in recognition accuracy. It appears that information encoded from one form of the stimulus is inappropriate to recognising it in the other. This same argument may be applied to masked faces: subjects may encode useful information from the masked face but simply be unable to transfer it to the same face in the unmasked state. Experiment 3 investigated this possibility by using a full factorial design in which subjects initially observed target faces either with or without masks and then attempted to identify the same faces with masks either present or absent. If useful information was being encoded from the masked face, then subjects observing masked faces at presentation should show higher recognition accuracy when tested on faces with masks still present, rather than unmasked as in the preceding experiments. If, on the other hand, no useful information was derived from the masked face, then conditions involving masking should lead to uniformly low and inferior recognition a sharp

performance relative to the no-mask controls.

Method Subjects and design A total of 32 male and 32 female subjects aged between 17 and 48 years acted as subjects for this study. The subjects were a mixture of undergraduate students and local citizens, who were allocated to one of four groups of 16 subjects. Allocation to groups was arbitrary except that care was taken to ensure equal representation of subjects of different sexes and social backgrounds. The four groups corresponded to the cells ofa 2 x 2 factorial design which contrasted masked and unmasked targets at presentation with masked and unmasked targets and distractors in the recognition array.

There were, thus, four conditions: (i) no masks at presentation or test (NN), (ii) masks at presentation and test (MM), (iii) masks at presentation but not at test (MN)and (iv) no masks at presentation, masks at test (NM). Materials and targets Four male students agreed to be targets for this study. They were aged between 17 and 19 years and all were clean shaven and wore no spectacles. Each was initially photographed in a plain gaberdine raincoat staring full-face at the camera. Two sets of photographs were taken, one in which the targets appeared unmasked and a second in which a plain nylon mask similar to that used in the previous experiments was worn. These photographs constituted the stimuli for the conditions involving masked and unmasked faces at presentation. Targets then removed their raincoats and were rephotographed dressed in casual clothing. Two sets ofphotographs were again taken, this time in three quarter pose where one set showed the targets masked and the second, without masks. These photographs served as stimuli for the recognition arrays involving masked and unmasked faces, respectively. All photographs were head and shoulders views and were standard 9 cm x 7 cm polaroid colour prints. It was necessary to recruit a further 12 male students to act as additional distractor faces for the recognition array. All were of similar age and build to the four targets and were photographed dressed in casual clothing and adopted the same three-quarter pose as the targets. Two photographs were taken of each volunteer, one in which he wore a nylon mask and a second in which he was unmasked. Changing the pose of the face and

92


clothing between the target and recognition arrays ensured that recognition of the targets could not be mediated by pose or clothing cues. Procedure Subjects were approached individually and asked whether they would be willing to take part in a small experiment. If they agreed, they were told that they would be shown the

faces

of four men wearing

macintoshes who they would later be called upon to

recognize. They were then shown the appropriate set offaces, each face being exposed for 30 s. Order of presentation of the faces and condition of presentation employed (masked/unmasked) was varied randomly across subjects. After a delay of one minute which was filled with casual conversation, subjects were called upon to try to recognize the four faces in the array of 16. For half the subjects the array of masked faces was employed whereas the remainder saw the array of unmasked faces. All four targets were present in both arrays, the precise position ofeach target being varied randomly across

subjects. Subjects were restricted to four selections from the array but were allowed

unrestricted time to make their choice. Results and

discussion

I

The number of targets correctly identified by each subject was tabulated as a function of condition, and the mean scores on this measure are shown in Table 3. Performance in all four conditions was significantly above the chance level of 1 item correct. The data again proved suitable for analysis of variance and were subjected to a2 x 2 x 2 analysis which used study condition, test condition and sex as main variables. The presence of a mask at study had a powerful impact on subsequent recognition, F (1,56) 34.15,p < 0.001, but there was no significant overall influence of the form of the test, F (1,56) < 1. However, both of these factors were qualified by a significant interaction, F (1,56) 29.88, p < 0.001. Examination of the simple effects of this interaction

:

:

indicated that, as expected, NN produced significantly superior performance to all other conditions. However, the all mask MM condition showed a score equivalent to that of NM, whereas both these latter conditions were significantly better than MN (all comparisons, p < 0.01). Sex of subject did not influence recognition rates, nor did it interact with either of the other main variables, all Fs < 1. It is clear that the results offer some partial support for encoding specificity. Subjects who study masked faces at presentation are better able to recognize them later still masked than if the masks are removed. This result is not a trivial one in that there was a change of pose and clothing between study and test so that the result could not be attributed to the presence ofincidental cues in the photographs. Subjects were able to Table 3. Mean hits (hnd standard deviations) as a function of masked faces at study and test: Experiment 3 Faces

Mask No mask at study Mask at study

at test

2.50 (0.82) 2.44 (0.63)

No mask 3.44 (0.51) 1.56

(0.63)


93

extract information from the masked faces which they successfully applied to recognizing them at test. However, as comparison with the no mask NN condition shows, the information extracted from the masked face is less effective than that for the normal face. Another indication of the lower rate of useful information extraction from the masked face comes from a comparison of performance in the NM and MN conditions. Subjects find it much easier to identify a masked face on the basis of information from a

normal face than they do the reverse. Thus, there is evidence to support both hypotheses advanced earlier: useful identity information c:rn be extracted from the masked face but it is much less than that encoded from a normal face. The reason for the inferior quality of the masked face could lie either in the fact that it is a more impoverished stimulus or that subjects are just less familiar with them as a class of stimuli, or both. A training study using masked faces would be necessary to resolve the

latter

issue.

CONCLUSIONS It appears from the results of Experiment

that the nylon mask provides a simple and effective means of disguising the face. Although direct comparison of results is diffrcult owing to procedural considerationsrthe near chance performance observed in the initial experiment suggests an impact on identification comparable to that of the addition of a wig or beard (cf. Patterson and Baddeley, 1977). Performance in the recognition task declined to levels approaching chance for both forms of nylon mask employed. Are there any reasons for believing that these findings, depressing from the perspective of law enforcement, are not representative of the criminal context? Certainly, the witness to a real crime might have access to other, non-facial cues which might assist in identifying the criminal. These include gait, physique and, perhaps, voice information, all of which can be employed for the purpose of identification (Clifford and Bull, 1978). However, against this must be set the very short time delays operating in the laboratory study and the much lower levels of stress, both of which might have assisted the experimental subjects (see Cliford (1978) for a discussion of the representativeness of laboratory simulations). Experiment 2 provides some useful information as to how masks disrupt the face encoding process. It appears that their primary impact is upon discrimination of feature information. Current theories of face recognition tend to discount exclusive reliance on facial features in favour of relational or semantic information (Bruce, in press). However, the mask does not disrupt the relationship between features in the face, but rather distorts the features themselves. To the extent that failures of recognition can be taken as diagnostic of the workings of the cognitive system as a whole, these current results would tend to suggest a role for feature information in face recognition, particularly where novel faces are involved. In this respect the report of Carey and Diamond (1977) that recognition of familiar faces was much less affected by their disguise manipulations is of interest. From a theoretical standpoint, it would be intriguing to see to what extent the faces of friends and acquaintances were 1

recognizable beneath a nylon mask. Given its potency, how can the effect of masking be best countered in practice? It is clear from Experiment 1 that simple instructions to bank or retail staffto adopt a salient

94

Graham Daoies and Rhona Flin

feature or attributional approach are likely to be ineffective. The results of Experiment 3, however, provide some grounds for believing that recognition rates for masked persons might be boosted if witnesses have the opportunity to observe them masked rather than unmasked. It is notable that the condition which most closely approximates to the orthodox identification sequence: observe suspect masked, attempt to identify

him with others all unmasked, produced the lowest recognition rates of all in Experiment 3. There seem good grounds, on the basis of this result, for arguing that where the type of mask is known, members of an identification parade might be paraded wearing masks as well as in the conventional unmasked state. Of course, caution is always necessary before extrapolating from laboratory studies to police procedures; the use in Experiment 3 of forced choice recognition procedure is clearly at variance with practice at identification parades where the option of making no selection is explicitly offered to witnesses. Nevertheless, the results of Experiment 3 are sufficiently encouraging as to support experiments in the use of masks on parades in appropriate cases. Such an innovation would be well within the discretion afforded by current Home Office Guidelines to the conducting officers at identification parades.

ACKNOWLEDGEMENTS The authors' research is supported by Contract PR/81 l3l53ll from the Scientific Research and Development Branch of the Home Office. Our thanks are due to our targets who submitted to the minor indignities inflicted upon them with humour and forebearance and to Linda Rodger, Stephen King, Pauline Hopson and others who assisted in data collection. The views expressed in this paper are those of the authors and do not necessarily reflect those of the Police Department, Home Office.

REFERENCES Baddeley, A. D. Domains of recollection. Psychological Reuiew, 1982,89,708-729.

Banks, W. P. Signal detection theory and human memory. Psychological Bulletin, 1970,74, 81-99. Bruce, V. Recognising faces. PhilosophicalTransactions of the Royal Society (London) (Series B),

(in press). Carey, D. and Diamond, R. From piecemeal to configurational representation of faces. Science. 1977, 195,312-314. Clifford, B. R. Memory for voices: The feasibility and quality of earwitness evidence. In S. LloydBostock and B. R. Clifford (EAs), Eualuating witness et:idence. Chichester; Wiley, 1983. Clifford, B. R. A critique of eyewitness research. In M. N. Gruneberg, P. E. Morris and R. W. Sykes (Eds), Practical aspects of memory. London: Academic Press, 1978. Clifford, B. R. and Bull, R.The psychology of person iilentification. London: Routledge and Kegan

Paul, 1978. Davies, G. M. The recognition of persons from drawings and photographs. Human Learning.

1983,2,237-249. Diamond, R. and Carey, S. Developmental changes in the representation of faces. Journal of Experimental Child P s ycholog y. 197 7, 23, l-22. Eich, J. E. The cue-dependent nature of state-dependent retrieval. Memory anil Cognition.l980, 8, 157-173.

Ellis, H. D. Theoretical aspects of face recognition. In G. M. Davies, H. D. Ellis and J. W. Shepherd (Eds), Perceit:ing and remembering faces. London: Academic Press, 1981.


95

Ellis, H. D., Davies, G. M. and Shepherd, J. W. Remembering pictures of 'real'and'unreal'faces. Some practical and theoretical considerations. Eritish Journal of Psychology. 1978, 69,

467474. Laughery, K. R. and Fowler, R. H. Facial recognition: Eflects of changing accessories. Proceedings of the Human Factors Society Zlst Annual Meeting, San Fransisco, 1977. Malpass, R. Training people in recognising faces. In G. M. Davies, H. D. Ellis and J. W. Shepherd (Eds), Perceiuing and remembering faces. London: Academic Press, 1981. Patterson, K. E. and Baddeley, A. D. When face recognition fails. Journal of Experimental Psychology: Human learning and Memory. 1977,3,406_!17. Shepherd, J. W., Davies, G. M. and Ellis, H. D. Studies of cue saliency. In G. M. Davies, H. D. Ellis and J. W. Shepherd (Ed,s), Perceiaing and remembering faces. London: Academic Press, 1981. Tieger, T. and Ganz, L. Recognition of flaces in the presence of two-dimensional sinusoidal masks. Perception and Psychophysics. 1979,26, 163-167. Tulving, E. and Thomson, D. M. Encoding specificity and retrieval processes in episodic memory. Psychological Reuiew. 1973, 80, 353-370. Winograd, E. Elaboration and distinctiveness in memory for faces. Journal of Experimental Psychology: Human Learning and Memory. 1981,7, l8l-190. Woodhead, M. M., Baddeley, A. D. and Simmonds, D. C. V. On training people to recognise faces. Ergonomic s. 1979, 22, 333-343. Zavala, A. and Paley, J. Personal appearance identiJication. Springfield, Illinois: Thomas, 1972.