Face identification proficiency test designed using item response theory.

Measures of face-identification proficiency are essential to ensure accurate and consistent performance by professional forensic face examiners and others who perform face-identification tasks in applied scenarios. Current proficiency tests rely on static sets of stimulus items and so cannot be administered validly to the same individual multiple times. To create a proficiency test, a large number of items of "known" difficulty must be assembled. Multiple tests of equal difficulty can be constructed then using subsets of items. We introduce the Triad Identity Matching (TIM) test and evaluate it using item response theory (IRT). Participants view face-image "triads" (N = 225) (two images of one identity, one image of a different identity) and select the different identity. In Experiment 3, university students (N = 197) showed wide-ranging accuracy on the TIM test, and IRT modeling demonstrated that the TIM items span various difficulty levels. In Experiment 3, we used IRT-based item metrics to partition the test into subsets of specific difficulties. Simulations showed that subsets of the TIM items yielded reliable estimates of subject ability. In Experiments 3a and b, we found that the student-derived IRT model reliably evaluated the ability of non-student participants and that ability generalized across different test sessions. In Experiment 3c, we show that TIM test performance correlates with other common face-recognition tests. In summary, the TIM test provides a starting point for developing a framework that is flexible and calibrated to measure proficiency across various ability levels (e.g., professionals or populations with face-processing deficits).

Confidence judgments are associated with face identification accuracy: Findings from a confidence forced-choice task.

Confidence is assumed to be an indicator of identification accuracy in legal practices (e.g., forensic face examination). However, it is not clear whether people can evaluate the correctness of their face-identification decisions reliably using confidence reports. In the current experiment, confidence in the correctness of the perceptual decision was measured with a confidence forced-choice methodology: Upon completion of two perceptual face-identity matching trials, the participants were asked to compare the two decisions and to select the trial on which they felt more confident. On each face-identity matching trial, participants viewed three face images (two same-identity images, one different-identity image) and were instructed to select the image of the different identity. In order to measure the extent to which difficulty level informs confidence decisions, we selected face-image triads using item-difficulty estimates extracted from psychometric modeling applied in a prior study. The difference in difficulty between the paired face-image triads predicted the proportion of high-confidence judgments allocated to the easier trial of the pair. Consistent with the impact of difficulty monitoring on confidence judgments, performance was significantly more accurate on trials associated with higher confidence. Overall, the results suggested that people reliably evaluate the correctness of their perceptual face-identity matching decisions and use trial difficulty to evaluate confidence.

Cortical processing of dynamic bodies in the superior occipito-temporal regions of the infants' brain: Difference from dynamic faces and inversion effect.

Previous functional neuroimaging studies imply a crucial role of the superior temporal regions (e.g., superior temporal sulcus: STS) for processing of dynamic faces and bodies. However, little is known about the cortical processing of moving faces and bodies in infancy. The current study used functional near-infrared spectroscopy (fNIRS) to directly compare cortical hemodynamic responses to dynamic faces (videos of approaching people with blurred bodies) and dynamic bodies (videos of approaching people with blurred faces) in infants' brain. We also examined the body-inversion effect in 5- to 8-month-old infants using hemodynamic responses as a measure. We found significant brain activity for the dynamic faces and bodies in the superior area of bilateral temporal cortices in both 5- to 6-month-old and 7- to 8-month-old infants. The hemodynamic responses to dynamic faces occurred across a broader area of cortex in 7- to 8-month-olds than in 5- to 6-month-olds, but we did not find a developmental change for dynamic bodies. There was no significant activation when the stimuli were presented upside down, indicating that these activation patterns did not result from the low-level visual properties of dynamic faces and bodies. Additionally, we found that the superior temporal regions showed a body inversion effect in infants aged over 5 months: the upright dynamic body stimuli induced stronger activation compared to the inverted stimuli. The most important contribution of the present study is that we identified cortical areas responsive to dynamic bodies and faces in two groups of infants (5-6-months and 7-8-months of age) and we found different developmental trends for the processing of bodies and faces.

Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms.

Achieving the upper limits of face identification accuracy in forensic applications can minimize errors that have profound social and personal consequences. Although forensic examiners identify faces in these applications, systematic tests of their accuracy are rare. How can we achieve the most accurate face identification: using people and/or machines working alone or in collaboration? In a comprehensive comparison of face identification by humans and computers, we found that forensic facial examiners, facial reviewers, and superrecognizers were more accurate than fingerprint examiners and students on a challenging face identification test. Individual performance on the test varied widely. On the same test, four deep convolutional neural networks (DCNNs), developed between 2015 and 2017, identified faces within the range of human accuracy. Accuracy of the algorithms increased steadily over time, with the most recent DCNN scoring above the median of the forensic facial examiners. Using crowd-sourcing methods, we fused the judgments of multiple forensic facial examiners by averaging their rating-based identity judgments. Accuracy was substantially better for fused judgments than for individuals working alone. Fusion also served to stabilize performance, boosting the scores of lower-performing individuals and decreasing variability. Single forensic facial examiners fused with the best algorithm were more accurate than the combination of two examiners. Therefore, collaboration among humans and between humans and machines offers tangible benefits to face identification accuracy in important applications. These results offer an evidence-based roadmap for achieving the most accurate face identification possible.

Facial expression is retained in deep networks trained for face identification.

Facial expressions distort visual cues for identification in two-dimensional images. Face processing systems in the brain must decouple image-based information from multiple sources to operate in the social world. Deep convolutional neural networks (DCNN) trained for face identification retain identity-irrelevant, image-based information (e.g., viewpoint). We asked whether a DCNN trained for identity also retains expression information that generalizes over viewpoint change. DCNN representations were generated for a controlled dataset containing images of 70 actors posing 7 facial expressions (happy, sad, angry, surprised, fearful, disgusted, neutral), from 5 viewpoints (frontal, 90° and 45° left and right profiles). Two-dimensional visualizations of the DCNN representations revealed hierarchical groupings by identity, followed by viewpoint, and then by facial expression. Linear discriminant analysis of full-dimensional representations predicted expressions accurately, mean 76.8% correct for happiness, followed by surprise, disgust, anger, neutral, sad, and fearful at 42.0%; chance \(\approx\)14.3%. Expression classification was stable across viewpoints. Representational similarity heatmaps indicated that image similarities within identities varied more by viewpoint than by expression. We conclude that an identity-trained, deep network retains shape-deformable information about expression and viewpoint, along with identity, in a unified form-consistent with a recent hypothesis for ventral visual stream processing.

Closing the gap between single-unit and neural population codes: Insights from deep learning in face recognition.

Single-unit responses and population codes differ in the "read-out" information they provide about high-level visual representations. Diverging local and global read-outs can be difficult to reconcile with in vivo methods. To bridge this gap, we studied the relationship between single-unit and ensemble codes for identity, gender, and viewpoint, using a deep convolutional neural network (DCNN) trained for face recognition. Analogous to the primate visual system, DCNNs develop representations that generalize over image variation, while retaining subject (e.g., gender) and image (e.g., viewpoint) information. At the unit level, we measured the number of single units needed to predict attributes (identity, gender, viewpoint) and the predictive value of individual units for each attribute. Identification was remarkably accurate using random samples of only 3% of the network's output units, and all units had substantial identity-predicting power. Cross-unit responses were minimally correlated, indicating that single units code non-redundant identity cues. Gender and viewpoint classification required large-scale pooling of units-individual units had weak predictive power. At the ensemble level, principal component analysis of face representations showed that identity, gender, and viewpoint separated into high-dimensional subspaces, ordered by explained variance. Unit-based directions in the representational space were compared with the directions associated with the attributes. Identity, gender, and viewpoint contributed to all individual unit responses, undercutting a neural tuning analogy. Instead, single-unit responses carry superimposed, distributed codes for face identity, gender, and viewpoint. This undermines confidence in the interpretation of neural representations from unit response profiles for both DCNNs and, by analogy, high-level vision.

First Impressions of Personality Traits From Body Shapes.

People infer the personalities of others from their facial appearance. Whether they do so from body shapes is less studied. We explored personality inferences made from body shapes. Participants rated personality traits for male and female bodies generated with a three-dimensional body model. Multivariate spaces created from these ratings indicated that people evaluate bodies on valence and agency in ways that directly contrast positive and negative traits from the Big Five domains. Body-trait stereotypes based on the trait ratings revealed a myriad of diverse body shapes that typify individual traits. Personality-trait profiles were predicted reliably from a subset of the body-shape features used to specify the three-dimensional bodies. Body features related to extraversion and conscientiousness were predicted with the highest consensus, followed by openness traits. This study provides the first comprehensive look at the range, diversity, and reliability of personality inferences that people make from body shapes.

Recognizing approaching walkers: Neural decoding of person familiarity in cortical areas responsive to faces, bodies, and biological motion.

In natural viewing environments, we recognize other people as they move through the world. Behavioral studies indicate that the face, body, and gait all contribute to recognition. We examined the neural basis of person recognition using a decoding approach aimed at discriminating the patterns of neural activity elicited in response to seeing visually familiar versus unfamiliar people in motion. Participants learned 30 identities by viewing multiple videos of the people in action. Recognition was tested inside a functional magnetic resonance imaging (fMRI) scanner using 8-s videos of 60 people (30 learned and 30 novel) approaching from a distance (~13m). Full brain images were taken while participants watched the approach. These images captured neural activity at four time points (TRs) corresponding to progressively closer views of the walker. We used pattern classification techniques to examine familiarity decoding in lateralized ROIs and the combination of left and right (bilateral) regions. Results showed accurate decoding of familiarity at the farthest distance in the bilateral posterior superior temporal sulcus (bpSTS). At a closer distance, familiarity was decoded in the bilateral extrastriate body area (bEBA) and left fusiform body area (lFBA). The most robust decoding was found in the time window during which the average behavioral recognition decision was made - and when the face came into clearer view. Multiple regions, including the right occipital face area (rOFA), bOFA, bFBA, bpSTS, and broadly distributed face- and body-selective voxels in the ventral temporal cortex decoded walker familiarity in this time window. At the closest distance, the lFBA decoded familiarity. These results reveal a broad system of ventral and dorsal visual areas that support person recognition from face, body, and gait. Although the face has been the focus of most person recognition studies, these findings remind us of the evolutionary advantage of being able to differentiate the people we know from strangers at a safe distance.

Creating Body Shapes From Verbal Descriptions by Linking Similarity Spaces.

Brief verbal descriptions of people's bodies (e.g., "curvy," "long-legged") can elicit vivid mental images. The ease with which these mental images are created belies the complexity of three-dimensional body shapes. We explored the relationship between body shapes and body descriptions and showed that a small number of words can be used to generate categorically accurate representations of three-dimensional bodies. The dimensions of body-shape variation that emerged in a language-based similarity space were related to major dimensions of variation computed directly from three-dimensional laser scans of 2,094 bodies. This relationship allowed us to generate three-dimensional models of people in the shape space using only their coordinates on analogous dimensions in the language-based description space. Human descriptions of photographed bodies and their corresponding models matched closely. The natural mapping between the spaces illustrates the role of language as a concise code for body shape that captures perceptually salient global and local body features.

Spatiotemporal changes in neural response patterns to faces varying in visual familiarity.

Increasing experience with a previously unfamiliar face improves human ability to recognize it in challenging and novel viewing conditions. Differential neural responses to familiar versus unfamiliar faces in multiple regions of the ventral-temporal and parietal cortex have been reported in previous work, but with limited attention to how behavioral and neural measures change with increasing familiarity. We examined changes in the spatial and temporal characteristics of neural response patterns elicited by faces that vary in their degree of visual familiarity. First, we developed a behavioral paradigm to familiarize participants to low-, medium-, and high-levels of familiarity with faces. Recognition of novel, naturalistic images of the learned individuals improved with increasing familiarity with faces. Next, a new set of participants learned faces using the behavioral paradigm, outside the fMRI scanner, and subsequently viewed blocks of whole-body images of the learned and novel people, inside the scanner. We found that the face-selective FFA and OFA, and a combination of the ventral-temporal areas (e.g., fusiform gyrus) and parietal areas (e.g., precuneus) contained patterns useful for classifying highly familiar versus unfamiliar faces. Classification along the temporal-sequence of the face blocks revealed an early separation of neural patterns elicited in response to highly familiar versus unfamiliar faces in the FFA and OFA, but not in other regions of interest. This indicates the potential for a rapid assessment of the "known versus unknown" status of faces in core face-selective regions of the brain. The present study provides a first look at the perceptual and neural correlates underlying experience gains with faces as they become familiar.

Perceptual expertise in forensic facial image comparison.

Forensic facial identification examiners are required to match the identity of faces in images that vary substantially, owing to changes in viewing conditions and in a person's appearance. These identifications affect the course and outcome of criminal investigations and convictions. Despite calls for research on sources of human error in forensic examination, existing scientific knowledge of face matching accuracy is based, almost exclusively, on people without formal training. Here, we administered three challenging face matching tests to a group of forensic examiners with many years' experience of comparing face images for law enforcement and government agencies. Examiners outperformed untrained participants and computer algorithms, thereby providing the first evidence that these examiners are experts at this task. Notably, computationally fusing responses of multiple experts produced near-perfect performance. Results also revealed qualitative differences between expert and non-expert performance. First, examiners' superiority was greatest at longer exposure durations, suggestive of more entailed comparison in forensic examiners. Second, experts were less impaired by image inversion than non-expert students, contrasting with face memory studies that show larger face inversion effects in high performers. We conclude that expertise in matching identity across unfamiliar face images is supported by processes that differ qualitatively from those supporting memory for individual faces.

The neural representation of faces and bodies in motion and at rest.

The neural organization of person processing relies on brain regions functionally selective for faces or bodies, with a subset of these regions preferring moving stimuli. Although the response properties of the individual areas are well established, less is known about the neural response to a whole person in a natural environment. Targeting an area of cortex that spans multiple functionally-selective face and body regions, we examined the relationship among neural activity patterns elicited in response to faces, bodies, and people in static and moving displays. When both stimuli were static or moving, pattern classification analyses indicated highly discriminable responses to faces, bodies, and whole people. Neural discrimination transferred in both directions between representations created from moving or static stimuli. It transferred also to stimuli experienced across static and dynamic presentations (one static and the other dynamic). In both transfer cases, however, discrimination accuracy decreased relative to the case where the representations were both created and tested with static or moving forms. Next, we examined the relative contribution of activity pattern and response magnitude to discrimination by comparing classifiers that operated with magnitude-normalized scans with classifiers that retained pattern and magnitude information. When both stimuli were moving or static, response magnitude contributed to classification, but the spatially distributed activity pattern accounted for most of the discrimination. Across static and moving presentations, activity pattern accounted completely for the discriminability of neural responses to faces, bodies, and people, with no contribution from response magnitude. Combined, the results indicate redundant and flexible access to person-based shape codes from moving and static presentations. The transfer of shape information across presentation types that preferentially access dorsal and ventral visual processing streams indicates that a common shape code may ground functional divisions in the processing of face and body information.

Identity recognition from faces and bodies in schizophrenia spectrum disorders.

Deficits in facial identity recognition and its association with poor social functioning are well documented in schizophrenia, but none of these studies have assessed the role of the body in these processes. Recent research in healthy populations shows that the body is also an important source of information in identity recognition, and the current study aimed to thoroughly examine identity recognition from both faces and bodies in schizophrenia. Sixty-five individuals with schizophrenia and forty-nine healthy controls completed three conditions of an identity matching task in which they attempted to match unidentified persons in unedited photos of faces and bodies, edited photos showing faces only, or edited photos showing bodies only. Results revealed global deficits in identity recognition in individuals with schizophrenia (ηp2 = 0.068), but both groups showed better recognition from bodies alone as compared to faces alone (ηp2 = 0.573), suggesting that the ability to extract useful information from bodies when identifying persons may remain partially preserved in schizophrenia. Further research is necessary to understand the relationship between face/body processing, identity recognition, and functional outcomes in individuals with schizophrenia-spectrum disorders.

Unaware person recognition from the body when face identification fails.

How does one recognize a person when face identification fails? Here, we show that people rely on the body but are unaware of doing so. State-of-the-art face-recognition algorithms were used to select images of people with almost no useful identity information in the face. Recognition of the face alone in these cases was near chance level, but recognition of the person was accurate. Accuracy in identifying the person without the face was identical to that in identifying the whole person. Paradoxically, people reported relying heavily on facial features over noninternal face and body features in making their identity decisions. Eye movements indicated otherwise, with gaze duration and fixations shifting adaptively toward the body and away from the face when the body was a better indicator of identity than the face. This shift occurred with no cost to accuracy or response time. Human identity processing may be partially inaccessible to conscious awareness.

First impressions from faces in the real world: Commentary on Sutherland and Young (2022).

The study of first impressions from faces now emphasizes the need to understand trait inferences made to naturalistic face images (British Journal of Psychology, 113, 2022, 1056). Face recognition algorithms based on deep convolutional neural networks simultaneously represent invariant, changeable and environmental variables in face images. Therefore, we suggest them as a comprehensive 'face space' model of first impressions of naturalistic faces. We also suggest that to understand trait inferences in the real world, a logical next step is to consider trait inferences made to whole people (faces and bodies). On the role of cultural contributions to trait perception, we think it is important for the field to begin to consider the way in which trait inferences motivate (or not) behaviour in independent and interdependent cultures.

First impressions: Integrating faces and bodies in personality trait perception.

Faces and bodies spontaneously elicit personality trait judgments (e.g., trustworthy, dominant, lazy). We examined how trait information from the face and body combine to form first impressions of the whole person and whether trait judgments from the face and body are affected by seeing the whole person. Consistent with the trait-dependence hypothesis, Experiment 1 showed that the relative contribution of the face and body to whole-person perception varied with the trait judged. Agreeableness traits (e.g., warm, aggressive, sympathetic, trustworthy) were inferred primarily from the face, conscientiousness traits (e.g., dependable, careless) from the body, and extraversion traits (e.g., dominant, quiet, confident) from the whole person. A control experiment showed that both clothing and body shape contributed to whole-person judgments. In Experiment 2, we found that a face (body) rated in the whole person elicited a different rating than when it was rated in isolation. Specifically, when trait ratings differed for an isolated face and body of the same identity, the whole-person context biased in-context ratings of the faces and bodies towards the ratings of the context. These results showed that face and body trait perception interact more than previously assumed. We combine current and established findings to propose a novel framework to account for face-body integration in trait perception. This framework incorporates basic elements such as perceptual determinants, nonperceptual determinants, trait formation, and integration, as well as predictive factors such as the rater, the person rated, and the situation.

Collaboration to improve cross-race face identification: Wisdom of the multi-racial crowd?

Face identification is particularly prone to error when individuals identify people of a race other than their own - a phenomenon known as the other-race effect (ORE). Here, we show that collaborative "wisdom-of-crowds" decision-making substantially improves face identification accuracy for own- and other-race faces over individuals working alone. In two online experiments, East Asian and White individuals recognized own- and other-race faces as individuals and as part of a collaborative dyad. Collaboration never proved more beneficial in a social setting than when individual identification decisions were combined computationally. The reliable benefit of non-social collaboration may stem from its ability to avoid the potential negative outcomes of group diversity such as conflict. Consistent with this benefit, the racial diversity of collaborators did not influence either general or race-specific face identification accuracy. Our findings suggest that collaboration between two individuals is a promising strategy for improving cross-race face identification that may translate effectively into forensic and eyewitness settings.

Neural correlates of own- and other-race face perception: spatial and temporal response differences.

Humans show an "other-race effect" for face recognition, with more accurate recognition of own- versus other-race faces. We compared the neural representations of own- and other-race faces using functional magnetic resonance imaging (fMRI) data in combination with a multi-voxel pattern classifier. Neural activity was recorded while Asians and Caucasians viewed Asian and Caucasian faces. A pattern classifier, applied to voxels across a broad range of ventral temporal areas, discriminated the brain activity maps elicited in response to Asian versus Caucasian faces in the brains of both Asians and Caucasians. Classification was most accurate in the first few time points of the block and required the use of own-race faces in the localizer scan to select voxels for classifier input. Next, we examined differences in the time-course of neural responses to own- and other-race faces and found evidence for a temporal "other-race effect." Own-race faces elicited a larger neural response initially that attenuated rapidly. The response to other-race faces was weaker at first, but increased over time, ultimately surpassing the magnitude of the own-race response in the fusiform "face" area (FFA). A similar temporal response pattern held across a broad range of ventral temporal areas. The pattern-classification results indicate the early availability of categorical information about own- versus other-race face status in the spatial pattern of neural activity. The slower, more sustained, brain response to other-race faces may indicate the need to recruit additional neural resources to process other-race faces for identification.

Face Recognition by Humans and Machines: Three Fundamental Advances from Deep Learning.

Deep learning models currently achieve human levels of performance on real-world face recognition tasks. We review scientific progress in understanding human face processing using computational approaches based on deep learning. This review is organized around three fundamental advances. First, deep networks trained for face identification generate a representation that retains structured information about the face (e.g., identity, demographics, appearance, social traits, expression) and the input image (e.g., viewpoint, illumination). This forces us to rethink the universe of possible solutions to the problem of inverse optics in vision. Second, deep learning models indicate that high-level visual representations of faces cannot be understood in terms of interpretable features. This has implications for understanding neural tuning and population coding in the high-level visual cortex. Third, learning in deep networks is a multistep process that forces theoretical consideration of diverse categories of learning that can overlap, accumulate over time, and interact. Diverse learning types are needed to model the development of human face processing skills, cross-race effects, and familiarity with individual faces.

Accuracy comparison across face recognition algorithms: Where are we on measuring race bias?

Previous generations of face recognition algorithms differ in accuracy for images of different races (race bias). Here, we present the possible underlying factors (data-driven and scenario modeling) and methodological considerations for assessing race bias in algorithms. We discuss data-driven factors (e.g., image quality, image population statistics, and algorithm architecture), and scenario modeling factors that consider the role of the "user" of the algorithm (e.g., threshold decisions and demographic constraints). To illustrate how these issues apply, we present data from four face recognition algorithms (a previous-generation algorithm and three deep convolutional neural networks, DCNNs) for East Asian and Caucasian faces. First, dataset difficulty affected both overall recognition accuracy and race bias, such that race bias increased with item difficulty. Second, for all four algorithms, the degree of bias varied depending on the identification decision threshold. To achieve equal false accept rates (FARs), East Asian faces required higher identification thresholds than Caucasian faces, for all algorithms. Third, demographic constraints on the formulation of the distributions used in the test, impacted estimates of algorithm accuracy. We conclude that race bias needs to be measured for individual applications and we provide a checklist for measuring this bias in face recognition algorithms.