Holistic processing of shape cues in face identification: Evidence from face inversion, composite faces, and acquired prosopagnosia

Abstract

Face recognition is based on two main sources of information: Three-dimensional (3-D) shape and two-dimensional surface reflectance (colour and texture). The respective contribution of these two sources of information in face identity matching task is usually equal, suggesting that there is no functional dissociation. However, there is recent evidence from electrophysiology and neuroimaging that contribution of shape and surface reflectance can be dissociated in time and neural localization. To understand the nature of a potential functional dissociation between shape and surface information during face individualization, we used a 3-D morphable model (Blanz & Vetter 1999) to generate pairs of face stimuli that differed selectively in shape, reflectance, or both. In three experiments, we provided evidence that the processing of shape and surface reflectance can be functionally dissociated. First, participants performed a delayed face matching task, in which discrimination between the sample and distractor faces with the same orientation (either upright or inverted) was possible based on shape information alone, reflectance information alone, or both. Inversion decreased performance for all conditions, but the effect was significantly larger when discrimination was based on shape information alone. Second, we found that participants' composite face effect, a marker of holistic processing, was caused primarily by the presence of interfering shape cues, with little interference from surface reflectance cues. Finally, contrary to normal observers, a well-known patient with acquired prosopagnosia suffering from holistic face perception impairment performed significantly better when discriminating faces based on reflectance than on shape cues. Altogether, these observations support the view that the diagnosticity of shape information for individualizing faces depends relatively more on holistic face processing than that of surface reflectance cues.

Keywords:

• Acquired prosopagnosia
• Composite faces
• Face inversion
• Shape and surface reflectance

Acknowledgements

This study was supported by a research grant (ARC 07/12-007, Communauté Française de Belgique–Actions de Recherche Concertées), by the Belgian National Research Fund (FNRS), and by a Human Frontier Science Program postdoctoral fellowship to FJ.

Notes

¹The term “configural” was originally used to refer to the interactive or interdependent processing of facial features (e.g., Hole, 1994; Sergent, 1984; Young et al., 1987). This meaning can be considered largely as a synonym of the term “holistic face processing” introduced later (Tanaka & Farah, 1993). However, other authors (e.g., Leder & Bruce, 2000; Leder & Carbon, 2006; Maurer et al., 2002) have attributed a specific meaning to the term “configural” as reflecting the relative distances between features (e.g., interocular distance, mouth–nose distance). In this framework, “configural processing” then would refer exclusively to the sensitivity of the system to these relative distances. We have argued elsewhere against making such a distinction, which we believe to be unnecessary (Rossion, 2008, 2009; see also McKone & Yovel, 2009). Here the terms “configural” and “holistic” will be used throughout the paper as synonyms, reflecting the processing of all facial features, including relative distances, as an integrated whole.

²For all participants, no response time was found more than three standard deviations below the their mean response time.

³This procedure introduces a response bias specifically for same (aligned) trials: Participants respond more often by pressing the “different” response keys. Although this bias has been taken by some as being problematic or supporting evidence that the composite face effect is primarily a decisional rather than a perceptual effect (Richler, Gauthier, Wenger, & Palmeri, 2008), the large majority of authors disagree with this view and consider that this response bias is created by the set-up of a visual illusion in a “same/different” response paradigm. It is in fact exactly what researchers aim at measuring (see, e.g., McKone & Robbins, 2007; Rossion & Boremanse, 2008).

⁴If anything, the fact that the bottom halves of the stimuli were adjusted in width for these two conditions increased the similarity of the two different bottom halves in these two conditions. Therefore, if this (indispensable) manipulation has an effect, it would be to slightly reduce the composite face effect in these two conditions, an effect that runs counter to our hypothesis.

⁵Note that there is no “absolute” way to quantify the physical variance in shape and texture. Pixel-by-pixel image comparisons would certainly be nonoptimal, because a simple global shift of the face would cause a huge image difference, even though it is the same face (and looks like the same face, too).