Perceiving age and gender in unfamiliar faces: an fMRI study on face categorization

Efficient processing of unfamiliar faces typically involves their categorization (e.g., into old vs. young or male vs. female). However, age and gender categorization may pose different perceptual demands. In the present study, we employed functional magnetic resonance imaging (fMRI) to compare the activity evoked during age vs. gender categorization of unfamiliar faces. In different blocks, participants performed age and gender classifications for old or young unfamiliar faces (50% female respectively). Both tasks elicited activations in the bilateral fusiform gyri (fusiform face area, FFA) and bilateral inferior occipital gyri (occipital face area, OFA). Importantly, the same stimuli elicited enhanced activation during gender as compared to age categorization. This enhancement was significant in the right FFA and the left OFA, and may be related to increased configural processing. Our findings replicate and extend recent work, and shows that the activation of core components of the face processing network is strongly dependent on task demands.     

Face Recognition Depends on Specialized Mechanisms Tuned to View-Invariant Facial Features: Insights from Deep Neural Networks Optimized for Face or Object Recognition

Face recognition is a computationally challenging classification task. Deep convolutional neural networks (DCNNs) are brain-inspired algorithms that have recently reached human-level performance in face and object recognition. However, it is not clear to what extent DCNNs generate a human-like representation of face identity. We have recently revealed a subset of facial features that are used by humans for face recognition. This enables us now to ask whether DCNNs rely on the same facial information and whether this human-like representation depends on a system that is optimized for face identification. In the current study, we examined the representation of DCNNs of faces that differ in features that are critical or non-critical for human face recognition. Our findings show that DCNNs optimized for face identification are tuned to the same facial features used by humans for face recognition. Sensitivity to these features was highly correlated with performance of the DCNN on a benchmark face recognition task. Moreover, sensitivity to these features and a view-invariant face representation emerged at higher layers of a DCNN optimized for face recognition but not for object recognition. This finding parallels the division to a face and an object system in high-level visual cortex. Taken together, these findings validate human perceptual models of face recognition, enable us to use DCNNs to test predictions about human face and object recognition as well as contribute to the interpretability of DCNNs.     

The role of higher level adaptive coding mechanisms in the development of face recognition

Developmental improvements in face identity recognition ability are widely documented, but the source of children’s immaturity in face recognition remains unclear. Differences in the way in which children and adults visually represent faces might underlie immaturities in face recognition. Recent evidence of a face identity aftereffect (FIAE), in which adaptation (exposure) to a particular identity causes a previously neutral face to take on the computationally opposite identity, suggests that adults code faces in an opponent fashion relative to an average face. One previous study showed comparable FIAEs in 8-year-olds and adults but did not demonstrate that adaptation was selective for high-level representations in both groups. Using a developmentally appropriate FIAE task, we investigated whether children show adult-like adaptation for facial identity when adapting and test images differ in size. Both age groups showed an equivalent FIAE, suggesting that qualitative changes in the use of higher level adaptive coding mechanisms do not drive the developmental improvements in face recognition ability, at least from 8 years of age.     

Knowing what to look for: Voice affects face race judgements

The study examined the effect that auditory information (speaker language/accent: Japanese or French) had on the processing of visual information (the speaker's race: Asian or Caucasian) in two forced-choice tasks: Classification and perceptual judgement on animated talking characters. Two (male and female) sets of facial morphs were constructed such that a 3-D head of Caucasian appearance was gradually morphed (in 11 steps) into one of Asian appearance. Each facial morph was animated in association with spoken French/Japanese or English with a French/Japanese accent. To examine the auditory effect, each animation was played with or without sound. Experiment 1 used an Asian or Caucasian classification task. Results showed that faces heard in conjunction with Japanese or a Japanese accent were more likely to be classified as Asian compared to those presented without sound. Experiment 2 used a same or different judgement task. Results showed that accuracy was improved by hearing a Japanese accent compared to without sound. These results were discussed in terms of the voice information acting as a cue to assist in organizing and attending to face features.     

Sex differences in face processing: Are women less lateralized and faster than men?

The aim of this study was to determine the influence of sex on hemispheric asymmetry and cooperation in a face recognition task. We used a masked priming paradigm in which the prime stimulus was centrally presented; it could be a bisymmetric face or a hemi-face in which facial information was presented in the left or the right visual field and projected to the right or the left hemisphere. The target stimulus was always a bisymmetric face presented centrally. Faces were selected from Minear and Park’s (2004) database. Fifty-two right-handed students (26 men, 26 women) participated in this experiment, in which accuracy (percentage of correct responses) and reaction times (RTs in ms) were measured. Although accuracy data showed that the percentage of correct recognition – when prime and target matched – was equivalent in men and women, men’s RTs were longer than women’s in all conditions. Accuracy and RTs showed that men are more strongly lateralized than women, with right hemispheric dominance. These results suggest that men are as good at face recognition as women, but there are functional differences in the two sexes. The findings are discussed in terms of functional cerebral networks distributed over both hemispheres and of interhemispheric transmission.     

The contribution of shape and surface information in the other-race face effect

Faces from another race are generally more difficult to recognize than faces from one's own race. However, faces provide multiple cues for recognition and it remains unknown what are the relative contribution of these cues to this “other-race effect”. In the current study, we used three-dimensional laser-scanned head models which allowed us to independently manipulate two prominent cues for face recognition: the facial shape morphology and the facial surface properties (texture and colour). In Experiment 1, Asian and Caucasian participants implicitly learned a set of Asian and Caucasian faces that had both shape and surface cues to facial identity. Their recognition of these encoded faces was then tested in an old/new recognition task. For these face stimuli, we found a robust other-race effect: Both groups were more accurate at recognizing own-race than other-race faces. Having established the other-race effect, in Experiment 2 we provided only shape cues for recognition and in Experiment 3 we provided only surface cues for recognition. Caucasian participants continued to show the other-race effect when only shape information was available, whereas Asian participants showed no effect. When only surface information was available, there was a weak pattern for the other-race effect in Asians. Performance was poor in this latter experiment, so this pattern needs to be interpreted with caution. Overall, these findings suggest that Asian and Caucasian participants rely differently on shape and surface cues to recognize own-race faces, and that they continue to use the same cues for other-race faces, which may be suboptimal for these faces.     

Tolerance of geometric distortions in infant's face recognition

The aim of the current study is to reveal the effect of global linear transformations (shearing, horizontal stretching, and vertical stretching) on the recognition of familiar faces (e.g., a mother's face) in 6- to 7-month-old infants. In this experiment, we applied the global linear transformations to both the infants’ own mother's face and to a stranger's face, and we tested infants’ preference between these faces. We found that only 7-month-old infants maintained preference for their own mother's face during the presentation of vertical stretching, while the preference for the mother's face disappeared during the presentation of shearing or horizontal stretching. These findings suggest that 7-month-old infants might not recognize faces based on calculating the absolute distance between facial features, and that the vertical dimension of facial features might be more related to infants’ face recognition rather than the horizontal dimension.     

My face through the looking-glass: the effect of mirror reversal on reflection size estimation

People tend to grossly overestimate the size of their mirror-reflected face. Although this overestimation bias is robust, not much is known about its relationships to self-face perception. In two experiments, we investigated the overestimation bias as a function of the presentation of the own face (left–right reversed – as in a mirror – or nonreversed – as in a photograph), the identity of the seen face, and prior exposure to a real mirror. For this we developed a computerized task requiring size estimations of displayed faces. We replicated the observation that people overestimate the size of their mirror-reflected face and showed that the overestimation can be reduced following a brief mirror exposure. We also found that left–right reversal modulates the overestimation bias, depending on the perceived face’s identity. These data underline the enhanced familiarity of left–right reversed self-faces and the importance of size perception for understanding mirror reflection processing.     

Females excel at basic face perception

Females are generally better than males at recognizing facial emotions. However, it is not entirely clear whether and in what way females may also excel at non-affective face recognition. Here, we tested males and females on two perceptual face recognition tasks that involved only neutral expressions: detection and identity discrimination. On face detection (Experiment 1), females were significantly more accurate than males in detecting upright faces. This gender difference was reduced during inverted face detection, and not present during tree detection, suggesting that the magnitude of the gender difference for performance co-varies with the extent to which face processing mechanisms are involved. On facial identity discrimination (Experiment 2), females again outperformed males, particularly when face images were masked by visual noise, or the delay between comparison face images was extended from 0.5 to 3 s. These results reveal a female advantage in processing face-specific information and underscore the role of perceptual factors in socially relevant gender differences.     

Seeing faces in the noise: Stochastic activity in perceptual regions of the brain may influence the perception of ambiguous stimuli

Research on binocular rivalry and motion direction discrimination suggests that stochastic activity early in visual processing influences the perception of ambiguous stimuli. Here, we extend this to higher level tasks of word and face processing. In Experiment 1, we used blocked gender and word discrimination tasks, and in Experiment 2, we used a face versus word discrimination task. Stimuli were embedded in noise, and some trials contained only noise. In Experiment 1, we found a larger response in the N170, an ERP component associated with faces, to the noise-alone stimulus when observers were performing the gender discrimination task. The noise-alone trials in Experiment 2 were binned according to the observer’s behavioral response, and there was a greater response in the N170 when they reported seeing a face. After considering various top-down and priming-related explanations, we raise the possibility that seeing a face in noise may result from greater stochastic activity in neural faceprocessing regions.     

Classification objects, ideal observers & generative models

A successful vision system must solve the problem of deriving geometrical information about three-dimensional objects from two-dimensional photometric input. The human visual system solves this problem with remarkable efficiency, and one challenge in vision research is to understand how neural representations of objects are formed and what visual information is used to form these representations. Ideal observer analysis has demonstrated the advantages of studying vision from the perspective of explicit generative models and a specified visual task, which divides the causes of image variations into the separate categories of signal and noise. Classification image techniques estimate the visual information used in a task from the properties of “noise” images that interact most strongly with the task. Both ideal observer analysis and classification image techniques rely on the assumption of a generative model. We show here how the ability of the classification image approach to understand how an observer uses visual information can be improved by matching the type and dimensionality of the model to that of the neural representation or internal template being studied. Because image variation in real world object tasks can arise from both geometrical shape and photometric (illumination or material) changes, a realistic image generation process should model geometry as well as intensity. A simple example is used to demonstrate what we refer to as a “classification object” approach to studying three-dimensional object representations.     

ERP evidence for task modulations on face perceptual processing at different spatial scales

Does the perceptual processing of faces flexibly adapt to the requirements of the categorization task at hand, or does it operate independently of this cognitive context? Behavioral studies have shown that the fine and coarse spatial scales of a face are differentially processed depending on the categorization task performed, thus suggesting that the latter can influence stimulus perception. Here, we investigated the time course of these task influences on perceptual processing by examining the visual N170 face‐sensitive Event‐Related Potential (ERP), while observers categorized faces for their gender and familiarity. Stimuli were full spectrum, or filtered versions that preserved either coarse or fine scale information of the faces. Behavioral results replicated previous findings of a differential processing of coarse and fine spatial scales across tasks. In addition, the N170 amplitude was larger in the Gender task as compared to the Familiarity task for LSF faces exclusively, thus showing that task demands differentially modulated the spatial scale processing on faces. These results suggest that the diagnosticity of scale‐specific cues in categorization tasks can modulate face processing.     

Electrophysiological correlates of face distortion after-effects

When observers are exposed to a distorted face the perceived configuration of a subsequently presented face is altered, a phenomenon called face distortion after-effect (FDAE). We compared the face-related components of the event-related potential (ERP) after adaptation to noise images—veridical and distorted faces. We found large bilateral adaptation effects on the P100 and N170 components that are related to face detection. Moreover, we found smaller adaptation effects on the N170, recorded over the right hemisphere, which can be related to the behavioural distortion after-effect and to face configurations. Our results suggest that the observed ERP adaptation effects are general for various steps of face processing and that the FDAEs similarly to gender after-effects are related to the early face-specific ERP components.     

Perceptual load effects on processing distractor faces indicate face-specific capacity limits

The claim that face perception is mediated by a specialized “face module” that proceeds automatically, independently of attention (e.g., Kanwisher, 2000) can be reconciled with load theory claims that visual perception has limited capacity (e.g., Lavie, 1995) by hypothesizing that face perception has face-specific capacity limits. We tested this hypothesis by comparing the effects of face and nonface perceptual load on distractor face processing. Participants searched a central array of either faces or letter strings for a pop star versus politician's face or name and made speeded classification responses. Perceptual load was varied through the relevant search set size. Response competition effects from a category-congruent or -incongruent peripheral distractor face were eliminated with more than two faces in the face search task, but were unaffected by perceptual load in the name search task. These results support the hypothesis that face perception has face-specific capacity limits and resolve apparent discrepancies in previous research.     

Superordinate shape classification using natural shape statistics

This paper investigates the classification of shapes into broad natural categories such as animal or leaf. We asked whether such coarse classifications can be achieved by a simple statistical classification of the shape skeleton. We surveyed databases of natural shapes, extracting shape skeletons and tabulating their parameters within each class, seeking shape statistics that effectively discriminated the classes. We conducted two experiments in which human subjects were asked to classify novel shapes into the same natural classes. We compared subjects’ classifications to those of a naive Bayesian classifier based on the natural shape statistics, and found good agreement. We conclude that human superordinate shape classifications can be well understood as involving a simple statistical classification of the shape skeleton that has been “tuned” to the natural statistics of shape.     

Gender differences in global-local perception? Evidence from orientation and shape judgments

Direct examinations of gender differences in global-local processing are sparse, and the results are inconsistent. We examined this issue with a visuospatial judgment task and with a shape judgment task. Women and men were presented with hierarchical stimuli that varied in closure (open or closed shape) or in line orientation (oblique or horizontal/vertical) at the global or local level. The task was to classify the stimuli on the basis of the variation at the global level (global classification) or at the local level (local classification). Women’s classification by closure (global or local) was more accurate than men’s for stimuli that varied in closure on both levels, suggesting a female advantage in discriminating shape properties. No gender differences were observed in global-local processing bias. Women and men exhibited a global advantage, and they did not differ in their speed of global or local classification, with only one exception. Women were slower than men in local classification by orientation when the to-be-classified lines were embedded in a global line with a different orientation. This finding suggests that women are more distracted than men by misleading global oriented context when performing local orientation judgments, perhaps because women and men differ in their ability to use cognitive schemes to compensate for the distracting effects of the global context. Our findings further suggest that whether or not gender differences arise depends not only on the nature of the visual task but also on the visual context.     

Face adaptation to gender: Does adaptation transfer across age categories?

We examined gender adaptation effects for the faces of children and adults and measured the transfer of these effects across age categories. Face gender adaptation is defined by a bias to classify the gender of a gender-neutral face to be opposite to that of an adapting face. An androgynous face, for example, appears male following adaptation to a female face. Participants adapted to male or female faces from the two age categories and classified the gender of morphed adult and child faces from a male–female morph trajectory. Gender adaptation effects were found for children's and adults’ faces and for the transfer between the age categories. The size of these effects was comparable when participants adapted to adult faces and identified the gender of either adult or child faces, and when participants adapted to child faces and identified the gender of child faces. A smaller adaptation effect was found when participants adapted to a child's face but identified the gender of an adult's face. The results indicate an interconnected and partially shared representation of the gender information for child and adult faces. The lack of symmetry in adaptation transfer between child and adult faces suggests that adaptation to adult faces is more effective than adaptation to child faces in activating a gender representation that generalizes across age categories.     

Attention to low- and high-spatial frequencies in categorizing facial identities, emotions and gender in children with autism

This study was aimed at investigating face categorization strategies in children with autistic spectrum disorders (ASD). Performance of 17 children with ASD was compared to that of 17 control children in a face-matching task, including hybrid faces (composed of two overlapping faces of different spatial bandwidths) and either low- or high-pass filtered faces. Participants were asked to match faces on the basis of identity, emotion or gender. Results revealed that children with ASD used the same strategies as controls when matching faces by gender. By contrast, in the identity and the emotion conditions, children with ASD showed a high-pass bias (i.e., preference for local information), contrary to controls. Consistent with previous studies on autism, these findings suggest that children with ASD do use atypical (local-oriented) strategies to process faces.     

Perceptual specialization and configural face processing in infancy

Adults’ face processing expertise includes sensitivity to second-order configural information (spatial relations among features such as distance between eyes). Prior research indicates that infants process this information in female faces. In the current experiments, 9-month-olds discriminated spacing changes in upright human male and monkey faces but not in inverted faces. However, they failed to process matching changes in upright house stimuli. A similar pattern of performance was exhibited by 5-month-olds. Thus, 5- and 9-month-olds exhibited specialization by processing configural information in upright primate faces but not in houses or inverted faces. This finding suggests that, even early in life, infants treat faces in a special manner by responding to changes in configural information more readily in faces than in non-face stimuli. However, previously reported differences in infants’ processing of human versus monkey faces at 9 months of age (but not at younger ages), which have been associated with perceptual narrowing, were not evident in the current study. Thus, perceptual narrowing is not absolute in the sense of loss of the ability to process information from other species’ faces at older ages.     

Face processing in cotton-top tamarins (Saguinus oedipus)

Current research on face processing in primates has focused on a few species, mostly macaques and chimpanzees; to date, only one New World monkey, the squirrel monkey, has been tested. We explored face processing, and the inversion effect in particular, in a New World primate species, the cotton-top tamarin (Saguinus oedipus). In phase 1 of our study, we trained subjects to discriminate between two faces and two scrambled faces; we then presented the tamarins with a series of novel probes in order to determine the features underlying classification. Results showed that the tamarins relied on the external contour of the face for discrimination more than the internal features and their configuration. Statistical analyses revealed no differences in accuracy or response times to upright versus inverted stimuli, and thus no inversion effect. In phase 2, we provided subjects with additional training on the face versus scrambled face discrimination task in order to focus their attention on the configuration of the internal features. Accuracy data revealed individual differences in how tamarins classified these stimuli, even though each subject was trained in the same way. In phase 3, we tested for generalization to a new set of face stimuli, as well as for the capacity to show an inversion effect. For one subject who attended to the configuration of internal features, we found significant evidence of generalization, but no evidence for an inversion effect. Electronic Publication