Advanced Review Neural mechanisms of face perception, their emergence over development, and their breakdown Marlene Behrmann,1* K. Suzanne Scherf2 and Galia Avidan3 Face perception is probably the most developed visual perceptual skill in humans, most likely as a result of its unique evolutionary and social significance. Much recent research has converged to identify a host of relevant psychological mechanisms that support face recognition. In parallel, there has been substantial progress in uncovering the neural mechanisms that mediate rapid and accurate face perception, with specific emphasis on a broadly distributed neural circuit, comprised of multiple nodes whose joint activity supports face perception. This article focuses specifically on the neural underpinnings of face recognition, and reviews recent structural and functional imaging studies that elucidate the neural basis of this ability. In addition, the article covers some of the recent investiga- tions that characterize the emergence of the neural basis of face recognition over the course of development, and explores the relationship between these changes and increasing behavioural competence. This paper also describes studies that characterize the nature of the breakdown of face recognition in individuals who are impaired in face recognition, either as a result of brain damage acquired at some point or as a result of the failure to master face recognition over the course of development. Finally, information regarding similarities between the neural circuits for face perception in humans and in nonhuman primates is briefly cov- ered, as is the contribution of subcortical regions to face perception. © 2016 Wiley Periodicals, Inc. How to cite this article: WIREs Cogn Sci 2016, 7:247–263. doi: 10.1002/wcs.1388 INTRODUCTION challenges for the visual system. In terms of image properties, compared to other classes of visual inputs ace perception is probably the most developed (e.g., vehicles or even just different makes of cars) visual perceptual skill in humans, most likely as a faces are more similar to one another perceptually F fi result of its unique evolutionary and social signi - and are all essentially composed of the same local ele- cance. Perhaps surprisingly, in light of the value of ments (two eyes, a nose, cheeks, and a mouth) in the face recognition for survival, the discrimination and identical spatial layout (e.g., eyes above the nose). In individuation of faces present extraordinary addition, at any moment in time, faces carry a large amount of information about the individual includ- ing their age, gender, emotional state, and gaze direc- * Correspondence to: [email protected] tion, thereby increasing the complexity of processing 1 Department of Psychology and Center for the Neural Basis of the input. Notwithstanding these challenges, human Cognition, Carnegie Mellon University, Pittsburgh, PA, USA observers can identify individual faces accurately and 2Department of Psychology, Pennsylvania State University, Univer- sity Park, PA, USA rapidly even across radically different viewing condi- 3 tions (e.g., lighting, vantage point) and across struc- Department of Psychology, Ben Gurion University of the Negev, Beer Sheva, Israel tural geometric changes as the person ages or conveys different emotional expressions dynamically. Conflict of interest: The authors have declared no conflicts of inter- est for this article. While there is some variability in face recognition Volume7,July/August2016 ©2016WileyPeriodicals,Inc. 247 Advanced Review wires.wiley.com/cogsci abilities within the normal population (e.g., Refs and in marmosets10 [see also Refs 11–15]. We note, 1–5), most people can represent the identity of a very however, that the findings of single unit investiga- large number of faces, and can access the relevant tions and fMRI in monkeys are not always information such as the name and biographical consistent,16 and that there is a need to elucidate fur- knowledge associated with a particular face. Of inter- ther the relationship between these two domains. est too, is that these skills are derived in a relatively In humans, our understanding of the neural unsupervised fashion over the course of development substrate of face recognition has received the greatest (in contrast, for example, with word recognition that boost from the numerous fMRI studies investigating requires many hours of directed training for the this issue in adult humans. These studies, collectively, majority of individuals usually in the school setting). point to a number of regions that show a selective Here, we provide a review of recent studies that response to faces (compared to other stimuli) in explore the neural mechanisms supporting robust multiple regions, including the fusiform face area – and accurate face recognition. In addition, we review (FFA) gyrus,17 19 the lateral occipital face (LO) findings from studies that explore the emergence of region, the superior temporal sulcus (STS), and the – face perception over the course of development, and occipital face area (OFA).6,20 26 In addition to these report results from investigations of individuals who ‘core regions’ of face processing (adopting the termi- are impaired in face recognition, either as a result of nology of Ref 6), there are a number of other regions acquired brain damage or of a failure to master this outside the occipito-temporal cortex that constitute skill. In sum, based on the evidence, we argue that an ‘extended’ face recognition system and play a crit- the mature face recognition system comprises a dis- ical role in other aspects of face perception. These tributed network of multiple nodes whose joint activ- include for example, the anterior temporal lobe, ity supports reliable and robust face individuation which plays a key role in processing semantic and – (see also Ref 6). This network evolves and is fine- biographical information27 29 as well as the identity tuned over the course of development as evident by representations of faces,30 perhaps even independent both the emergence of the nodes of the network and of modality (for a recent review, see Refs 31–33. In their increased structural and functional connectivity. addition, the precuneus/posterior cingulate cortex Disrupting the network through damage to the and the anterior paracingulate cortex likely play a node/s themselves or through compromised connec- role in representing some knowledge of faces, consist- tivity between them results in impairments in face ent with the stronger activation for familiar versus recognition. We review the findings supporting the unknown faces in these regions obtained via various engagement of multiple cortical regions and discuss paradigms (e.g., generally famous faces,29 personally how novel methods, analytic techniques, cross- familiar faces,26 and visually familiar faces30). Others species comparisons, studies of prosopagnosia, etc. have implicated the precuneus/posterior cingulate have expanded our understanding of face perception region in the acquisition of face familiarity,31 and by documenting how these multiple cortical regions perhaps in the representation of familiarity more interact and relate to behavioral markers of face generally,34 and this is also consistent with studies expertise. showing selective activation for familiar voices in this region.32 Largely as a result of the imaging studies in FACE PERCEPTION humans and in nonhuman primates, there is now a growing consensus that face perception is accom- Neural Underpinnings: Normal plished by the activity of a well-connected face pro- Populations cessing network.18,35 However, considerable debate still continues to revolve around the particular role Cortical Contributions of the different face-selective regions, with some The neural underpinning of successful face represen- researchers even suggesting that the regions do not tation has been of much interest in visual neurosci- have assigned roles that are separate and distinct and ence likely because of the complexity of the process that all regions participate in all types of face percep- and the observers’ great facility with faces. Face tion. Advances in more sophisticated data analysis selective cells have long been identified in monkey approaches, including network analyses and multi- inferior temporal (IT) cortex (see Ref 7 for a review) voxel pattern analysis (MVPA) permit an examina- and, more recently, face selectivity has been con- tion of the properties of the face network as a firmed in the same regions in studies using functional whole36 in normal participants as well as in indivi- magnetic resonance imaging (fMRI) in monkeys8,9 duals with impaired face processing, and the 248 ©2016WileyPeriodicals,Inc. Volume7,July/August2016 WIREs Cognitive Science Neural mechanisms of face perception specification of the computational contribution of the is primarily on the distributed network and localization different face selective regions within the network at is better using MRI than using these other approaches a much finer grain of resolution. For example, using (although it is also better in MEG than in EEG), we MVPA, several studies have now implicated the ante- have mostly focussed on MRI studies. Many review rior temporal cortex as being critical for image invar- papers exist describing the findings from these iant representation of face identity (e.g., Refs 27, approaches, however, and the interested reader is 28, and 37) and, interestingly, these face representa- referred to electrophysiological