Sketch Artist And Identi-Kit Procedures For Recalling Faces

PROCEDURES FOR RECALLING FACEStechnician to generate a facial image. The rating phasewas an experiment in which the images generated in thefirst phase were evaluated for goodness of fit by aseparate group of subjects. Procedures for these twophases will be described separately.Construction PhaseThis phase consisted of the image generation partof the overall study.Subjects. The subjects can be divided into twogroups, those who served as targets and those whoserved as witnesses. A total of 71 target subjects wereused, all white males. The targets were drawn from thestudent body at the University of Houston and from theHouston community at large. The only restrictionplaced on the selection of targets, beside being whitemales, was that they be unknown to the witnesssubjects, the sketch artists, and the Identi-kittechnicians. There were 142 witnesses with norestrictions placed on their selection. Most witnesseswere students at the University of Houston. A breakdown of these subjects by sex and race shows 41 males(37 white, 1 black, 2 Chicano, and 1 Oriental) and 101females (81 white, 9 black, 8 Chicano, and 3 Oriental).All subjects were volunteers and were paid 2 per hourfor participating.Design. The design included three experimentalvariables. The first was the image generation technique,consisting of the sketch artist or the Identi-kit. Thesecond variable, to be referred to as artist/technician,consisted of three artists and three Identi-kit technicians. This variable svas nested within technique;that is, the three artists and the three technicians weresix different people.The third variable was created by an auxiliary procedure in the experiment. After images had beencompleted from the witnesses' descriptions, the artistand technician each constructed a second image whileviewing the target directly. Thus, for each target andtechnique condition there were two images, one fromdescription and one from view. This variable is referredto as target presentation.Artists and technicians. Three people served assketch artists and three others as Identi-kit technicians.The three artists, SN (female). BM (male), and AM(male), were recent graduates of the University ofHouston with a Bachelor of Fine Arts degree. Allthree had a good deal of training and experience inportrait work. Prior to beginning the experiment, eachartist constructed several practice images from description. The three technicians, MM (male), RF (male),and JH (female), were enrolled as graduate students inpsychology at the University of Houston. MMattended a 2'/2-day Identi-kit training course sponsoredby the Identi-kit Company. RF and JH were trained inthe procedures by MM. All three technicians practicedextensively prior to the experiment.Procedure. The experimental task consisted of twoparts, target exposure and image generation. During thefirst part the target was exposed to two witnesses. Theimage generation followed and consisted of one witnessworking with a sketch artist and the other witnessworking with an Identi-kit technician.309The procedural aspects of each experimental sessioninvolved the following six people: the experimenter, asketch artist, an Identi-kit technician, a target, andtwo witnesses. Since it was necessary to carefullycontrol the timing and manner in which differentindividuals encountered each other, and because avariety of data was obtained from the various individuals, a relatively complex and carefully controlledprocedure was carried out. Details of the procedureare available elsewhere (Laughery, Duval, & Fowler,Note 1). The following description provides an outline.Two witnesses reported to a room where they filledout a subject data form. This form asked for informationabout the witness, including certain physical characteristics. Instructions were then presented to the witnesses including a description of the target exposureand image generation parts of the study. The targetmeanwhile reported to an adjacent room where theexperimenter, after finishing with the witnesses,instructed the target on the nature of the study.Following the instructions the witnesses wereescorted to the room where the target was located. Witheveryone seated at a table, the experimenter moderatedan 8-minute conversation, which is referred to as theexposure period. To the extent possible, the discussionfocused on the target's interests, activities, and so on.Although the setting may seem to have been somewhatstrained or artificial, in actual practice it generallyproceeded smoothly with reasonably good conversation.Following the exposure period one witness wasescorted to a room to work with a sketch artist togenerate an image while the other witness went to workwith an Identi-kit technician. At the beginning of theimage generation phase, each witness filled out ageneral description form about the target that was usedby the artist/technician as a starting point. Then thewitness and artist/technician interacted to construct theimage. The verbal interaction was tape recorded.Following the exposure period, the target completedthe subject data form and then posed for a bust-lengthfront photograph. After completing the images, witnesses filled out three additional forms. The first wasa subject comments sheet that solicited commentsregarding the manner in which they carried out the task.The other forms were the Belts Vividness of ImageryScale and Gordon Test of Visual Imagery Control(Richardson, 1969); both are paper-and-pencil tests ofimagery or verbal memory. The final step consisted ofthe artist and technician producing a sketch andcomposite while viewing the target directly.Each artist and technician completed images for 24targets, except SN and MM who did 23.Rating PhaseAn important set of issues in this study concerns themanner in which one evaluates facial images. Whatdoes one measure? How does one decide whether aparticular image is a good, fair, or poor representationof a real face, and how is this goodness of fit quantified?Ellis et al. (1975, 1978) used various rating proceduresin which judges simultaneously viewed the target faceand the image and rated them for goodness of fit. A

310KENNETH R. LAUGHERY AND RICHARD H. FOWLERsimilar' approach was used here. A rating procedurewas employed in which an independent group of subjects rated each image-photograph pair for goodness offit on a 6-point scale.Subjects. Sixty-four students enrolled in an introductory psychology course at the U n i v e r s i t y ofHouston served as subjects. Extra credit was given inthe course for participating.Task. The task consisted of showing the subjects asequence of pairs of slides. Each pair consisted of atarget photograph and one of the four images for thattarget. The pair was projected onto a screen in frontof the subject for 10 sec. The projected images wereapproximately life-size. The subject looked at theimages, made a decision regarding the goodness of fitof the image to the photograph, and then indicated therating on a response sheet. The ratings were made on a6-point scale, where the two ends of the scale weredefined as "most similar" and "least similar."Design. The similarity ratings were collected fromtwo different groups of subjects, each group rating adifferent subset of the 71 faces. The reasons for thisprocedure were twofold, both logistical. First, theimage generation experiment was spread over a longtime period, and it was desirable to complete someanalyses before the entire data collection phase ended.Second, the number of ratings to be provided by eachsubject was four times the number of target faces.Therefore, if all 71 were introduced in one session,subjects would be required to complete 284 ratings.Such a procedure potentially introduces factors such asfatigue. An analysis of the task led to the conclusionthat about 200 ratings is a reasonable maximum. As aresult, ratings were obtained on 51 targets (204 images)in the first part of the experiment (51 instead of 50 wassimply a convenience due to the availability of stimulusmaterials). The other 20 targets (80 images) were ratedin the second part.Both parts of the experiment consisted of a ratherelaborate ordering and counterbalancing of thesequence of images across different subjects. Thepurpose of introducing this procedure was twofold.The first related to the fact that for each differenttarget face, four images were to be rated. These imageswere the sketch from description, sketch from view,composite from description, and composite from view.It was important that the image slides for a particulartarget face not appear too close together because eachrating should be independent of how well the otherimages matched that target. The second reason forcounterbalancing was to eliminate practice effects.Twenty-four subjects were run in the first part of theexperiment, and 40 were run in the second.Materials. The materials consisted of 355 slides.These included a photograph and one of each of thefour different images for each of the 71 targets.Procedure. Subjects were brought into the laboratory where they sat in a classroom-type desk. Theviewing screen was located approximately 3 m in frontof the subjects, and two Kodak Carousel projectorswere above and behind them. Instructions were readinformally, and the subjects were given a set ofresponse sheets. A series of 10 sample pairs were thenpresented to familiarize subjects with the task. Thissample included pairs representing a range of goodness-of-fit values as determined in pilot work. The entire setof pairs (204 or 80) was then presented at a 10-sec rate.In all pairs the photograph appeared on the left and theimage on the right.ResultsSeveral analyses were carried out on thedata. The ratings were quantified 1-6, inwhich 1 was a very good ('' most similar") fitand 6 was a very poor ("least similar") fit.Figure 1 shows an example of a good andpoor sketch and composite. Using thisrating as a dependent measure, an analysisof variance examined the effects of fourvariables: replication (the two parts of therating experiment), technique (sketch artistor Identi-kit), artist/technician (which wasnested within technique), and presentation(description or view). The mean rating foreach of the conditions is shown in Table 1.The main effects of three variables weresignificant: technique, F(l, 46) 134.24,p .001; presentation, F(\, 46) 174.23,p .001; and artist/technician, F(4, 184) 19.54,p .001. Images were better if doneas sketches or from view. The artist/technician effect indicates simply that therewere differences between artists and technicians. The means for the three artists were3.1, 2.9, and 3.4 for SN, BM, and AM,respectively. The Identi-kit technician meanswere 4.1, 4.0, and 3.9 for RF, MM,and JH. It appears that artist differencesmay be significantly greater than techniciandifferences; however, this interaction cannot be statistically examined in the presentstudy, since the artist/technician variableis nested within technique.The Technique x Presentation interaction was significant, F(l, 46) 68.59, p .001. The data indicate a larger differencebetween view and description in thesketches (2.7 vs. 3.6) than in the composites(3.9 vs. 4.1). A significant Presentation xArtist/Technician interaction, F(4, 184) 13.85, p ,001, simply reflects largerdifferences between view and descriptionfor some artist/technicians than others.Although the replication variable did notproduce a significant main effect, it didinteract with technique, F(l, 46) 15.19,p .001, and presentation, F(l, 46) 28.43, p .001. The effects of technique

PROCEDURES FOR RECALLING FACES311Target 1PhotographTarget 1High SimilaritySketchTarget 1Low SimilaritySketchTarget 2PhotographTarget 2High SimilarityCompositeTarget 2Low SimilarityCompositeFigure 1. Examples of good and poor sketches and composites.and presentation were in the same directionbut greater in the second replication.Although it was not part of the formalanalysis of the rating experiment, aninformal aspect of the outcome is worthmentioning. It was common for subjectswho had completed the rating phase tocomment on how poor the overall qualitywas. This general impression of low-qualityrepresentations is consistent with the con-clusions of Ellis et al. (1975, 1978) on thePhotofit technique.As noted earlier, a variety of data wasobtained in addition to the images. Includedin the data were scores on the Betts andGordon imagery tests for witnesses. Also,for subjects who were students at theUniversity of Houston, Scholastic AptitudeTest (SAT) Verbal scores were obtained.Six correlations were computed: TheTable 1Means and Standard Deviations of Ratings on 1-6 Similarity ScaleSketchIdcnti-kitBMSNAMMMRFJHRating studyPresentationMSDMSDMSDMSDMSDMSDReplication 1(51 cation 2(20 .81.01.04.44.11.21.34.63.91.01.24.23.81.01.1Note. Lower scores represent better images. The letters SN, BM, AM, RF, MM, and JH refer to the initialsof the artists and technicians.

312KENNETH R. LAUGHERY AND RICHARD H. FOWLERTable 2Means and Standard Deviations of Time-Line MeasuresNo. offeaturestopsDifferentfeaturecodesTime perfeaturestop (sec)Total 99.010.8ratings for sketches from description andcomposites from description were eachcorrelated with the two imagery scores andthe SAT Verbal scores. Two of theoutcomes were significant: The correlationbetween the sketches and the Gordonimagery score (r -.213,p ,05)andthecorrelation between the composites and theSAT Verbal scores (r -.487, p .01).These results are in the expected direction;that is, better images related to greaterimagery and verbal ability. (The lower therating, the better the image, hence the negative values.) None of the other correlationsapproached significance.Correlations were computed between theratings of the two types of images and thetotal time used to generate the images. Thelatter measure was defined from thebeginning of the subject's interaction withthe artist/technician until the subject statedthat the image was as good as he/she couldconstruct (or words to that effect). Bothcorrelations had a value of .06, which wasnot significant.The sex and race characteristics of thewitness subjects served as a basis foranalysis. The ratings for images done bymale and female witnesses were comparedas were the images generated by thedifferent races. None of these results weresignificant.In addition to the goodness-of-fit andcorrelational analyses, several aspects ofthe results were examined by what is hereinreferred to as time-line analysis. During theactual process of generating the images,tape recordings were made of many of theverbal interactions between the artists/technicians and witnesses. The tapes of 62interactions were available for detailedanalysis. Twenty-three features were definedon the basis of the contents of the tapesand the experience of the artists/technicians. The 23 features were eyes, nose,mouth and lips, ears, forehead, cheeks andcheek bones, jaw and jawline, chin, hair,hairline, eyebrows, sideburns, moustache,beard, face shape, proportions, glasses,eye color, complexion, wrinkles and facelines, general expression, scars and moles,and neck. These features represent a finegrained breakdown of the face. Such finedetail is appropriate in developing a firststage classification scheme, since it is arelatively simple matter to combine featureslater.Following the definition of the 23 differentfeature codes, the boundaries between workon each successive feature were identifiedon the tapes. A feature stop is defined as thecontinuous work on a given feature. Itshould be noted that the number of featurestops will exceed the number of featurecodes, since witnesses typically work on agiven feature code more than once. The laststep in analyzing the tapes was to note thetime lapse for each successive feature stop.To summarize, the output of this analysiswas the sequence in which the features wereworked on and the length of time spent oneach.Means and standard deviations for thedifferent measures by technique are shownin Table 2. The technique differences areclear. In creating sketches, witnesses useda greater number of feature codes, mademore feature stops, spent less time perfeature stop, and used more total time.A second analysis of the time-line datafocused on the different features. Theproportion of feature stops to total time wascomputed. These measures reflect therelative amounts of time and effort devoted

313PROCEDURES FOR RECALLING FACESTable 3Most Attended Features in Time-Line Feature AnalysisProportion of feature stops to total stopsProportion of feature time to total portionto totalFeatureProportionto totalFeatureProportionto totalFeatureProportionto total 'HairEyesFace owsChin.151.119.113.105.097EyesHairNoseMouth & uth & lips.193.186.149.108.088to the various features. Table 3 shows thefive features that received the most attentionfor each technique. Clearly, there is a greatdeal of consistency across techniques inhow much time and effort is devoted to thevarious features.A factor of potential importance ingenerating facial images is the experienceof the artists/technicians. A brief description of the training and experience of eachartist/technician was presented earlier. Toexamine possible learning effects, ananalysis of the quality of images as afunction of experience was carried out.Each of the artist's/technician's imageswere grouped into blocks of five, and ananalysis of variance was carried out on theratings across blocks. There was no indication of any improvement or decline for anyartist or technician.It seems likely that faces vary in terms ofthe ease or difficulty with which accuraterepresentations can be created for them. Ifthe sketches and composites prepared fromview represent the "best possible" images,and if the images from view and descriptionare positively correlated, then it could beconcluded that one limiting factor in imagequality is related to characteristics of thetarget face. For sketches the correlationbetween images from view and descriptionwas significant (/ .334,p .01). Similarly,a significant correlation was found forcomposites (V .363, p .01).There are several factors that could accountfor the superiority of sketches. First, thereis a limited set of alternative faces one cancreate with the Identi-kit, whereas a sketchartist can produce an essentially infinite set.Hence, with the Identi-kit there may betimes (and according to technicians, thereare) when "the right nose is not there." Asecond reason for sketch superiority may bethe additional detail such as shading, agelines, and so forth, that typically is morepredominant in sketches than in composites. This added detail is reflected in thetime-line data where a greater number offeature codes are used for sketches than forcomposites. A related fact here is that arecent version of the Identi-kit (updatedsince this work was completed) includes awider range of features and greater shading—changes that should lead to better representations.A third possibility may be related to thetotal time difference between techniques.More time is spent generating sketches thancomposites. More time is not directly thepoint, however, since the time differencecould be accounted for simply by the factthat an artist requires more time to producea feature than the Identi-kit, with whichfeatures are simply selected. The key pointis that because of the greater productiontime requirements of the sketch, the witnessspends more time thinking about the target,which may lead to a more accurate memoryand description. There is a serious hitch inthisexplanation, however, since the correDiscussionlational results showed that within techThe results indicate that sketch artists niques, total time was not related to goodproduce better images than the Identi-kit. ness of fit. A fourth possible explanation

314KENNETH R. LAUGHERY AND RICHARD H. FOWLERemerges from the time-line data. In generating sketches, witnesses use more codes,make more feature stops, and spend lesstime per feature stop. These differencesseem to reflect more "moving around" ingenerating sketches than in generatingcomposites. The moving around may resultin better relationships (e.g., distances)between features than a process orientedtoward completing work on one featureprior to moving to another. Of course, thenature of the Identi-kit makes this latter,feature-oriented procedure more likely.The fact that there was little differencebetween images from description and viewwith the Identi-kit has an interestingimplication. It may be that a major limitingfactor in the quality of composites is theIdenti-kit itself, not the ability of technicians. This idea is further supported by thefact that there was little or no differencebetween technicians, whereas there wereartist differences.The results of the most-attended-tofeatures analysis (Table 3) seem to reflect atendency to give more time and attention tofeatures in the upper half of the face thanin the lower half. This outcome is consistentwith several previous findings (Ellis et al.,1975; Goldstein & Mackenberg, 1966;Laughery, Alexander, & Lane, 1971). Itmay be, as Goldstein and Mackenbergsuggest, that the upper half of the face conveys more information helpful to memory.The lack of any learning effect with artist/technician experience may be due toasymptotic performance being realized inthe preexperimental training. On the otherhand, it may be that using slightly over20 images was not su

provided by the sketch artist and the Identi-kit. The sketch artist procedure involves an artist sketching the target person while getting information from a witness through conversational interaction. The Identi-kit is a set of transparent celluloid sheets, each containing a line drawing of a facial feature. There are a large number of sheets .