|
Article Excerpt
The perception of a face varies according to personal experience. At one extreme, a face may have only been seen once, often referred to as an 'unfamiliar' face, and the resulting memory is fragile and strongly modulated by image-specific factors such as lighting (Bruce, 1982; Hill & Bruce, 1996), viewing angle (Bruck, Cavanagh, & Ceci, 1991; Davies & Milne, 1982; Hill & Bruce, 1996), facial expression (Bruce, 1982; Davies & Milne, 1982) and background (Memon & Bruce, 1983). At the other extreme, given many encounters, identification becomes more robust and is largely invariant of such factors (e.g. Bruce et al., 1999; Bruce, Henderson, Newman, & Burton, 2001; Burton, Wilson, Cowan, & Bruce, 1999; Young, Hay, McWeeny, Flude, & Ellis, 1985).
In general, familiar faces enjoy high recognition (e.g. Bruce, 1988; Bruce et al., 2001), even under difficult conditions such as poor quality video (Burton et al., 1999), highly pixelated images (e.g. Bruce & Young, 1998) or brief presentation (Lee & Perrett, 1997). The misidentification of familiar faces is very low (e.g. Young, Hay, & Ellis, 1985). In contrast, under the influence of these image-specific factors, unfamiliar faces tend to be identified less well than familiar faces, but misidentified much more often, even under very favourable conditions such as high quality media (e.g. Henderson, Bruce, & Burton, 2001). Indeed, misidentification, through unfamiliar face processing, is believed to be the principal cause of wrongful conviction (Rattner, 1988).
The relative importance or salience of different areas of the face also appears to change with familiarity. Ellis, Shepherd, and Davies (1979) found that familiar faces were recognized more accurately from their internal features--the region including the eyes, brows, nose and mouth--than from their external facial features--head shape, hair and ears--but for unfamiliar faces, the external features dominated. Similarly, Young et al. (1985) found faster reaction times and fewer errors made for the internal features of familiar faces using a face-matching paradigm. In general, an advantage for the inner face of familiar items emerges with adulthood (Campbell et al., 1999) and may be observed following relatively few encounters (O'Donnell & Bruce, 2001). For unfamiliar faces, the external parts of a face tend to have a more salient role in face perception than their internal counterparts (Bruce et al., 1999; de Haan & Hay, 1986; Ellis et al., 1979; Hancock et al., 2000; Young et al., 1985).
One applied area of psychology that involves both familiar and unfamiliar face processing is the production and recognition of facial composites. These are pictures of suspects to a crime produced by witnesses (or victims) with the help of computer software or a sketch artist. In general, witnesses see a suspect for a short time--maybe only for a few seconds--and so composite construction engages unfamiliar face perception, although the goal is to promote identification by people familiar with the target when a composite is published. While composites remain an important tool for the apprehension of suspects, research has suggested that they are of poor quality and generally only named around 20% of the time, at best, when witnesses construct them soon after seeing a target face (Bruce, Ness, Hancock, Newman, & Rarity, 2002; Brace, Pike, & Kemp, 2000; Davies, van der Willik, & Morrison, 2000; Frowd et al., in press; Frowd, Carson, Ness, Richardson, et al., 2005; Frowd, Hancock, & Carson, 2004). Unfortunately, composite quality appears to be even worse when construction occurs 2 days after inspecting a target face, as tends to occur in real life (Frowd et al., in press; Frowd, Carson, Ness, McQuiston, et al., 2005; Koehn & Fisher, 1997).
As unfamiliar faces are remembered and processed generally better by their external features, and witnesses typically engage in this type of processing, it was predicted that the external features of facial composites should be better constructed than their internal ones. However, given the relative importance of the internal features of a face for identifying a person (e.g. Bruce et al., 1999; Ellis et al., 1979; Young et al., 1985), it is possible that composites are not well named due to their poor representation of internal features--the parts of a face most critical for identifying familiar people.
The focus of the current paper is twofold. In the first two of three experiments, we investigate the relative recognizability of the internal and external features of composites produced using procedures that mirror some aspects of real life: for example, from participant-witnesses who were unfamiliar with a target face and waited a couple of days prior to construction of composites from memory. Our prediction is that the quality of the internal composite features should be worse than the quality of the external composite features, as indicated by two matching-type tasks. This part also asks whether there is any useful information in general conveyed by the internal features. In Experiment 3, we went on to investigate the reasons why internal features are poorly depicted in the composites, investigating whether the problem concerns the unfamiliarity of the target faces or some inherent difficulty in constructing internal vs. external features.
INTRODUCTION TO EXPERIMENTS 1 AND 2
The materials used for the first two experiments were derived from a set of composites and target photographs used by Frowd, Carson, Ness, McQuiston, et al. (2005). The reader is referred to this paper for full details of the procedures used to construct these composites. In brief, participant-witnesses inspected a photograph of an unknown male, described his face and constructed a composite 2 days later working with an operator using one of five different composite systems. The methods included three popular computerized systems--E-FIT (UK), PRO-fit (UK) and FACES (US)--that required witnesses to assemble a face by selecting facial features from a predefined set of parts (hair, face shape, eyes, noses, mouth, etc); a UK police sketch artist, who drew out the composite face by hand; and an early version of EvoFIT, a UK 'holistic' system in development (Frowd et al., 2004), where a composite was 'evolved' through the repeated selection and 'breeding' of whole faces.
Frowd, Carson, Ness, McQuiston, et al.'s (2005) study used 10 target faces. These were of celebrities generally unknown to people aged over 30, who were the witnesses in the study, but familiar to participants later in the experiment, who were in their early twenties and carried out the main evaluation by naming. Further details of the targets may be found in Materials, Experiment 1 (below). The study also provided a check to verify that the targets were not recognized by participant-witnesses and thus the composites were constructed of unfamiliar faces.
After it had been established that the target was unfamiliar, participant-witnesses first inspected a photograph of a celebrity face for 1 minute in the knowledge that a composite would later be required (i.e. intentional learning was employed). Two days later, each participant-witness constructed a single composite using procedures that followed real witnesses as far as possible. These procedures included: the use of experimenters who were experienced in assisting witnesses construct a composite (referred to here as composite operators for computerized systems); a cognitive interview, to assist recall of the target face; open-ended construction sessions, to promote good performance; and software painting tools, whose use may improve the appearance of a face (e.g. Christie, Davies, Shepherd, & Ellis, 1981; Davies, Milne, & Shepherd, 1983; Frowd, Carson, Ness, Richardson, et al., 2005; Geiselman, Fisher, MacKinnon, & Holland, 1986; Gibling & Bennett, 1994). Composites took about an hour to construct using the E-FIT, PRO-fit and FACES systems, and about 2 hours for Sketch and EvoFIT.
Composites of each of the 10 celebrity targets were constructed using each of the five systems to produce 50 composites in...
|
