Human face processing consists of a number of distinct but interrelated tasks. The detection of faces within the visual environment, for example, enables the subsequent identity matching of unfamiliar faces, or the recognition of already-known identities. Each of these tasks has been studied in detail (see, e.g., Bindemann & Lewis, 2013; Burton, White, & McNeill, 2010; Johnston & Edmonds, 2009), but little is still known about whether these are conducted by shared or dissociable cognitive mechanisms. In turn, these tasks are characterised by substantial individual differences in performance (see, e.g., Bindemann, Avetisyan, & Rakow, 2012; Bindemann, Brown, Koyas, & Russ, 2012; Robertson, Noyes, Dowsett, Jenkins, & Burton, 2016; Russell, Duchaine, & Nakayama, 2009), but it is unresolved as to whether individuals who are good at face detection are similarly proficient at face matching or face memory. Therefore, the aim of this study is to assess relationships between individual performance in these three tasks.
Studies investigating face detection show that this process is fast and highly accurate under self-paced conditions (e.g., Burton & Bindemann, 2009; Crouzet & Thorpe, 2011; Lewis & Edmonds, 2005). However, detection performance is reduced when changes to the natural width-to-height ratios of faces are made (Pongakkasira & Bindemann, 2015). In contrast, face recognition appears to be remarkably robust to such manipulations (Bindemann, Burton, Leuthold, & Schweinberger, 2008; Hole, George, Eaves, & Rasek, 2002; see also Burton, Schweinberger, Jenkins, & Kaufmann, 2015). Such findings imply that, whilst face detection and face recognition involve the same stimulus category, these are dissociable processes. However, associations between such tasks have also been identified. Face recognition deficits in prosopagnosia, for example, have been linked to orienting failures to faces (see, e.g., Dalrymple, Corrow, Yonas, & Duchaine, 2012; Tsao & Livingstone, 2008), raising the alternative possibility that these tasks might engage similar mechanisms.
This prospect aligns with efforts to establish whether individual differences in performance across different, yet related, face processing tasks can be accounted for by a specific mechanism (see, e.g., Verhallen et al., 2017; Wilhelm et al., 2010). Using a battery of tests (see Herzmann, Danthiir, Schacht, Sommer, & Wilhelm, 2008), Wilhelm et al. (2010) found strong associations between face memory and face perception, and demonstrated that faces are processed independently of objects, implying a face-specific cognitive component. More recently, Verhallen et al. (2017) also found evidence to suggest that performance across four face-processing tasks could be accounted for by a common factor, which they referred to as f. This research showed that the ability to match unfamiliar faces is strongly associated with unfamiliar face recognition but correlates weakly with face detection. However, the test of face detection that was employed (the Mooney Face Test; see Mooney, 1956; see also, Verhallen et al., 2014) measures participants’ ability to visually organise black and white shapes into face-like arrangements, rather than assessing the detection of actual human faces. In addition, other research has implied that performance in the Mooney Test dissociates from visual search performance (see Foreman, 1991). This search component is a key element of human face detection, which requires the location of a target within visual scenes (see, e.g., Bindemann & Lewis, 2013; Burton & Bindemann, 2009). From these findings, therefore, it is difficult to establish whether face detection ability is associated with face matching and face memory.
One other recent study also investigated possible relationships between face detection, face matching, and face memory (Robertson, Jenkins, & Burton, 2017). This study identified a correlation between face matching and face memory. However, accuracy in these tasks was not associated with participants’ detection of face-like objects, such as pareidolia faces (Experiment 1) and cloud faces (Experiment 2). A third experiment also found no association between the ability to detect human faces in natural scenes and face matching accuracy, but did not include a measure of face memory.
These findings make intuitive sense, when considering that face matching and face memory both concern the identification of face stimuli. The former task requires observers to decide whether one face photograph matches that of another similar but potentially different identity. By contrast, face memory tasks entail a similar identity judgement, but which is based on the extent to which a face image that is stored in memory corresponds to a visual representation of a face that is presented. As a consequence, these tasks should, in theory, overlap to some degree. Indeed, this relationship has been observed repeatedly in previous work (see, e.g., Bobak, Hancock, & Bate, 2016; Burton et al., 2010; Fysh & Bindemann, 2018; Megreya & Burton, 2006; Robertson et al., 2017; Verhallen et al., 2017). By contrast, it seems less intuitive to assert that face detection should be associated with face matching and face memory. This is due to the fact that the detection of a face within a visual display entails a between-category distinction to separate faces from non-face objects. On the other hand, face identification entails within-category distinctions, to determine whether two similar face images match or mismatch, or whether a face encountered within the visual field matches a facial representation stored in memory.
However, two obstacles arise from the research of Robertson et al. (2017) that limit the extent to which firm conclusions can be drawn about whether face detection is dissociated from the recognition and matching of unfamiliar faces. First, it remains uncertain as to whether face-like objects operate as a reliable proxy for human faces. These objects, which include stimuli such as clouds, may share some characteristics with faces but also exhibit many differences and are, de facto, objects in their own right that are not faces (Churches, Baron-Cohen, & Ring, 2009; Moulson, Balas, Nelson, & Sinha, 2011; Takahashi & Watanabe, 2013). Second, Robertson et al. (2017) only utilised accuracy measures to assess face detection performance. This diverges from earlier studies, which utilised response times when investigating detection performance, given that accuracy is often close to ceiling (see, e.g., Bindemann & Burton, 2009; Bindemann & Lewis, 2013; Burton & Bindemann, 2009). For example, manipulating the orientation of faces to be detected from frontal to profile orientation reduces accuracy slightly from 93% to 89%, but elicits a comparatively large increase in visual search time from 593 ms to 704 ms (Bindemann & Lewis, 2013). In addition, comparisons between people with prosopagnosia and control subjects when detecting faces in visual displays reveal only marginal differences in accuracy, but considerable differences in search time (Garrido, Duchaine, & Nakayama, 2008). Considered together, these studies reflect that proficiency in face detection may be best characterised by response speed, as opposed to response accuracy, when investigating possible associations between this ability and performance in face matching and face memory tasks.
In light of these observations, the aim of the current study was to further examine relationships between individual performance in the detection, matching and memory of faces. Three tasks were employed for this purpose. The first of these comprised a task in which observers searched complex natural scenes for faces (see Burton & Bindemann, 2009; Pongakkasira & Bindemann, 2015). The second and third tasks comprised challenging tests of face matching and face memory; the Kent Face Matching Test (KFMT; Fysh & Bindemann, 2018) and the long version of the Cambridge Face Memory Test (CFMT+; Russell et al., 2009).
These tests differ from the Glasgow Face Matching Test (GFMT; Burton et al., 2010) and the standard version of the Cambridge Face Memory Test (CFMT; Duchaine & Nakayama, 2006), which were employed by Robertson et al. (2017) and may lack the sensitivity to fully explore the range of individual differences in face matching and face memory. The CFMT, for example, is typically employed as a tool for assessing prosopagnosia (Bobak, Parris, Gregory, Bennetts, & Bate, 2017; Duchaine & Nakayama, 2006; Ulrich et al., 2017), but does not distinguish between individuals at the higher end of the face recognition continuum (Russell et al., 2009). In addition, stimuli in the GFMT comprise two well-lit faces bearing the same pose and expression. Critically, identity-match trials depict the same person photographed minutes apart, thereby presenting the task as a best-case scenario (Burton et al., 2010). By contrast, stimuli in the KFMT comprise one controlled target photograph and a non-controlled image, in which expression, pose and lighting, are unconstrained. In addition, identity matches consist of target photographs that were taken many months apart, resulting in considerable within-person variability. As a consequence, the KFMT provides a more difficult test of face matching than the GFMT (see, Fysh & Bindemann, 2018). Therefore, by replacing the GFMT and CFMT with the KFMT and CFMT+, and by using response time as an additional measure of individual performance in face detection, this study sought to further explore whether correlations exist between face detection, matching and memory performance.
In this experiment, observers completed a face detection task, which involved searching for faces within complex natural scenes (see Bindemann & Burton, 2009; Burton & Bindemann, 2009; Pongakkasira & Bindemann, 2015). These scenes were displayed only briefly to maximise individual differences in accuracy. The detection task was followed by the KFMT (Fysh & Bindemann, 2018) and the CFMT+ (Russell et al., 2009). To investigate relationships between these tasks fully, accuracy and the speed with which faces were detected within scenes was explored, and these measures were correlated with face matching and face memory. If the absence of associations between face detection, matching and memory in the study of Robertson et al. (2017) were driven by a lack of sensitivity in the matching (GFMT) and memory (CFMT) tests that were employed, then correlations between performance in these tasks might emerge under these alternative conditions.