Familiarity enhances invariance of face representations in human ventral visual cortex: fMRI evidence

Abstract

Face recognition across different viewing conditions is strongly improved by familiarity. In the present study, we tested the hypothesis that the neural basis of this effect is a less view-dependent representation of familiar faces in ventral visual cortex by assessing priming-related fMRI repetition effects. 15 healthy volunteers made male/female judgements on familiar (famous) and unfamiliar (novel) faces preceded by the same image, a different image of the same face, or another (unprimed) face. Reaction times revealed priming by same and different images independent of familiarity and more pronounced for same than different images. In the imaging data, a main effect of prime condition was found in bilateral fusiform and orbitofrontal regions. A right anterior fusiform region expressed stronger response decreases to repetition of familiar than unfamiliar faces. Bilateral mid-fusiform areas showed stronger response decreases to repetition of same than different images. A regions-of-interest analysis focussing specifically on face responsive regions suggested differences in the degree of image dependency across fusiform cortex. Collapsing across familiarity, there was greater image dependency of repetition effects in right than left anterior fusiform, replicating previous imaging findings obtained with common objects. For familiar faces alone, there was greater generalisation of repetition effects over different images in anterior than middle fusiform. This suggests a role of anterior fusiform cortex in coding image-independent representations of familiar faces.

Introduction

We are remarkably good at recognising faces across relatively low-level changes in the visual input, such as changes in lighting conditions. We also have considerable abilities to recognise faces across more complex changes, such as changes in facial expression, hairstyle, or viewing angle, when we are familiar with those faces (e.g., the faces of family or celebrities). However, we are less able to recognise faces across equivalent transformations when we are unfamiliar with them (see Hancock et al., 2000, for a review). For example, we are surprisingly inaccurate at identifying a previously seen but unknown “target” face among multiple previously unseen “distractor” faces, in conditions resembling police “line-ups” (Bruce et al., 1999, Wells, 1993). This suggests that we need to possess an “abstract” representation of a person's face, presumably following gradual learning over multiple exposures to that face, before we can identify it across different viewing conditions.

One approach towards investigating our ability to recognise faces across viewing conditions utilises the phenomenon of priming. Priming reflects a change in the behavioural response to a stimulus owing to prior exposure to the same (or a related) stimulus. The degree to which such priming effects are found despite changes in various aspects of the stimulus is used to infer how that stimulus is mentally represented. For example, if a priming effect for faces is obtained in spite of changes in lighting conditions, it might be inferred that face representations are independent of cues such as lighting that are used to extract surface properties. Conversely, if a priming effect disappears when the viewing angle is changed, it might be inferred that faces are encoded by multiple view-dependent representations. A similar logic has been used with electrophysiological (e.g., EEG) and haemodynamic (e.g., fMRI) measures of repetition effects in order to establish the temporal and spatial correlates of neural representation in the cortex (e.g., Grill-Spector et al., 1999, Henson, 2003, Naccache and Dehaene, 2001).

Behavioural effects of priming for pictures of both faces and objects often exhibit a degree of perceptual specificity, in that these effects are reduced when the perceptual format of the stimuli is changed between first and second presentation (Ellis et al., 1987, Srinivas, 1993), suggesting a degree of view-dependence of the representations mediating priming. Nonetheless, a certain degree of priming across perceptual changes usually remains, and some have suggested that abstract and view-specific representations may be processed in neuroanatomically separate systems, with the right hemisphere supporting view-specific priming and the left hemisphere supporting more abstractive priming (Marsolek, 1995, Marsolek, 1999, Marsolek et al., 1992).

EEG studies measuring event-related potentials (ERPs) have found a negative deflection peaking at about 250 ms post-stimulus, maximal over temporal–occipital sites (using an average reference), for famous faces preceded by the same image relative to famous faces preceded by a different face (Schweinberger et al., 1995). Schweinberger et al. (2002) still found this effect, termed the “N250r”, when a famous face was preceded by a different photograph of that face, though the size of the effect was diminished (relative to repetition of the same photograph). This might suggest that the neural processes giving rise to the N250r operate over both view-dependent and more abstract representations for familiar faces.

Studies using fMRI have generally found a decreased haemodynamic response associated with repetition of stimuli, so-called “repetition suppression” or “MR adaptation” (Henson, 2003). For example, Grill-Spector et al. (1999) compared mean fMRI signal during blocks of different faces with that during blocks in which the image of the same face varied in either position, size, illumination, or rotation in depth. The anterior, but not posterior, portion of the lateral occipital complex (occipito-temporal cortex that is more responsive to visual objects than scrambled objects or textures, Malach et al., 1995) showed adaptation across changes in position or size. No reliable adaptation was found, however, for faces presented with varying illumination or rotation, suggesting that representations of faces in this region are not viewpoint-invariant. Using immediate repetition of faces in a randomised, event-related fMRI design, Eger et al. (2004) found adaptation to faces in the mid-fusiform gyrus across changes in the spatial filtering (high-pass or low-pass) of the image, suggesting that face representations at this stage of the visual-processing pathway also generalise across fine vs. coarse spatial frequency content. The question remains, however, whether such generalisation across viewing conditions is an automatic property of the visual system, implying that it would be found for novel stimuli, or whether it is only established after familiarisation with faces. Multiple presentations of the originally unfamiliar faces within and prior to the experiments of Eger et al. (2004) and Grill-Spector et al. (1999) might have led to the development of more abstract representations as the stimuli became familiarised.

Henson et al., 2000, Henson et al., 2002, Henson et al., 2003 directly compared fMRI repetition effects for familiar (famous) and unfamiliar (novel) faces. They found that a right mid-fusiform region exhibited repetition suppression for repeated photographs of familiar faces, but not for repeated photographs of unfamiliar faces. This suggests that representations in mid-fusiform regions are specific to the identity of known faces and do not simply represent the physical characteristics of a given image. However, this could not directly be tested as only one image of each face was used. These experiments further differed from those of Grill-Spector et al. (1999) and Eger et al. (2004) in using much longer intervals between first and second presentations of a face (with multiple intervening faces). This difference in “repetition lag” may be associated with differences in the mechanisms underlying the haemodynamic repetition effects (Henson, 2003; hence the different uses above of “adaptation” versus “repetition suppression”).

Using pictures of everyday (familiar) objects, rather than faces, in a long-lag repetition paradigm, Vuilleumier et al. (2002) confirmed size independence of object representations in ventral visual cortex. In a more anterior left fusiform region, they found (unlike Grill-Spector et al., 1999) some degree of generalisation over different views of the objects, whereas view-specificity was more pronounced on the right (see Koutstaal et al., 2001, Simons et al., 2003 for similar evidence of lateralisation using different exemplars with the same name; though see James et al. (2002) for different findings). Whether such findings generalise to faces is unclear, given arguments that the neural representation of faces differs from that of other classes of objects (Biederman and Kalocsai, 1997, but also see Tarr and Cheng, 2003 for an alternative view). Furthermore, given evidence from prosopagnosia (Landis et al., 1986) and split visual field experiments (see Rhodes, 1985) for a right-hemisphere dominance in processing faces, one might in fact expect the opposite lateralisation for faces relative to objects, namely greater generalisation across face views in right than left fusiform.

The aim of the present study was to assess the influence of familiarity on the degree of view-dependence of face priming as measured by haemodynamic repetition effects. We used an immediate priming paradigm, in which participants made sex decisions to familiar (famous) and unfamiliar (novel) faces (each repeated only once to avoid confounding effects of familiarisation during the course of the experiment). The stimuli were a superset of those used by Schweinberger et al. (2002). There was no explicit control of the visual changes (such as viewing angle or illumination) across different photographs of the same person; rather, the different photographs reflected various changes more typical of everyday exposures to a person's face. The faces were presented in a continuous sequence, in which a face was either preceded by the same photograph (same image trials), a different photograph of the same face (different image trials), or a different face of the same sex (unprimed trials). The order of trials was randomised together with additional single face (filler) trials that served to break-up the pattern of pairs of faces with the same sex. The experiment thus conformed to a 2 × 3 factorial design with two levels of familiarity (familiar and unfamiliar faces) and three levels of priming (same image, different image, and unprimed; Fig. 1).

Given the evidence reviewed above, we had three main predictions. Firstly, based on behavioural evidence for better recognition across views for familiar than unfamiliar faces (Hancock et al., 2000), we expected to find less image dependency of behavioural priming effects for familiar than unfamiliar faces, with a similar pattern in the priming-related haemodynamic response reductions in fusiform cortex (Henson et al., 2003). Secondly, based on Grill-Spector et al. (1999) and Vuilleumier et al. (2002), we expected less image dependency of repetition effects in anterior than posterior fusiform regions. Finally, if faces are processed similarly to objects, we expected less image dependency of repetition effects in left than right fusiform, based on Marsolek's (1999) and Vuilleumier et al. (2002) experiments using objects (alternatively, if faces are processed differently from objects, we might expect the opposite lateralisation instead, based on the right-lateralised processing of faces suggested by Landis et al., 1986, Rhodes, 1985).