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.     

Different signals of personality and health from the two sides of the face

Previous studies demonstrate that people with different personality traits have different-looking faces. We investigated whether personality and health information are differently signalled by the two hemifaces. Using composite images created from women with high and low scores on health and personality dimensions, we investigated discrimination accuracy with original and mirrored hemifaces. By comparing discrimination accuracy for particular types of hemiface, we address issues regarding both the location of information signals and how these signals are conveyed. From the hemiface stimuli, participants could accurately identify three of the Big Five traits, along with health. We found differences in which hemiface could be more accurately identified, depending on the expressed trait. Emotional stability and health were more accurately discriminated from the right hemiface, while extraversion showed higher accuracy from the left hemiface. We found evidence for differences between hemifaces related to both directional asymmetries and to other information content. Finally, our results also address ongoing debate about which side of the face is more attractive, as we found attractiveness differences between hemifaces depended upon the personality trait most clearly expressed.     

Remembering and knowing personality traits: figure/ground asymmetries in person-related retrieval experience

In two experiments, the effect of category salience on retrieval experience was investigated. In Experiment 1, participants rated typicality or concreteness of personality traits that differed in stereotype reference (i.e., consistent, inconsistent, and neutral in relation to the age stereotype). More remember judgments were given for consistent and inconsistent traits in contrast to neutral traits, thereby indicating a figure/ground asymmetry. In Experiment 2, neutral traits were excluded and a classical figure/ground phenomenon was demonstrated for the retrieval experience of traits (i.e., reversibility of an ambiguous figure after typicality and untypicality ratings). Altogether, the results suggest that metacognitive trait representations depend on principles of figure/ground asymmetries rather than on functional principles of social information processing.     

All smiles perceived equally: Facial expressions trump target characteristics in impression formation

Race, gender, and emotionally expressive facial behavior have been associated with trait inferences in past research. However, it is unclear how interactions among these factors influence trait perceptions. In the current research, we test the roles of targets’ race, gender, and facial expression along with participants’ culture in predicting personality ratings. Caucasian and Asian-American participants rated the big-5 personality traits of either smiling or inexpressive photographs of Caucasian and Asian male and female faces. Ratings of extraversion, agreeableness, and conscientiousness differed significantly across inexpressive targets as a function of race and gender categorization and individual characteristics. Smiling was associated with reduced variation in perceptions of targets’ extraversion and agreeableness relative to ratings made of inexpressive targets. In addition, participant culture generally did not significantly impact trait ratings. Results suggest that emotionally expressive facial behavior reduces the use of information based on race or gender in forming impressions of interpersonally relevant traits.     

Not looking yourself: The cost of self‐selecting photographs for identity verification

Photo‐identification is based on the premise that photographs are representative of facial appearance. However, previous studies show that ratings of likeness vary across different photographs of the same face, suggesting that some images capture identity better than others. Two experiments were designed to examine the relationship between likeness judgments and face matching accuracy. In Experiment 1, we compared unfamiliar face matching accuracy for self‐selected and other‐selected high‐likeness images. Surprisingly, images selected by previously unfamiliar viewers – after very limited exposure to a target face – were more accurately matched than self‐selected images chosen by the target identity themselves. Results also revealed extremely low inter‐rater agreement in ratings of likeness across participants, suggesting that perceptions of image resemblance are inherently unstable. In Experiment 2, we test whether the cost of self‐selection can be explained by this general disagreement in likeness judgments between individual raters. We find that averaging across rankings by multiple raters produces image selections that provide superior identification accuracy. However, benefit of other‐selection persisted for single raters, suggesting that inaccurate representations of self interfere with our ability to judge which images faithfully represent our current appearance.     

Peer Sociometric Status and Personality Development from Middle Childhood to Preadolescence

Sociometric status, the regard that other group members confer to an individual, is one of the most ubiquitous and behaviourally relevant attributes assigned to the person by the social environment. Despite this, its contribution to personality development has received little attention. The present three‐wave longitudinal study, spanning the age range 7–13 years (n = 1222), sought to fill this gap by examining the transactional pathways between peer sociometric status (measured by peer nominations) and Five‐Factor personality traits (measured by self‐ratings and parent and teacher ratings). Sociometric status prospectively predicted the development of extraversion. By contrast, agreeableness and neuroticism prospectively predicted the development of sociometric status. Furthermore, individual‐level stability in extraversion was associated with individual‐level stability in sociometric status. The results were robust across different sources of personality ratings. We argue that peer sociometric status in the school classroom is the type of environmental effect that has potential to explain personality development. Because of its stability, broadness, and possible impact across a variety of personality processes, sociometric status can both repetitiously and simultaneously influence the network of multiple inter‐correlated micro‐level personality processes, potentially leading to a new network equilibrium that manifests in changes at the level of the broad personality trait. © 2019 European Association of Personality Psychology     

Self‐efficacy in the Workplace: Linking personality to domain‐specific efficacy beliefs

Based on a social cognitive approach to personality, the role of domain‐specific efficacy beliefs were examined in two studies. Study 1 developed a measure of expectations about success in 10 different domains of common work activities with domain‐specific expectancy ratings obtained for 157 individuals along with self‐ and peer ratings of personality. Results showed that predictable relationships emerged, with the traits of the five‐factor model being important determinants of construct‐relevant efficacy beliefs. Study 2 focused on expectations about activities related to conscientiousness as a link between personality and performance across jobs. Results (N = 97) confirmed the proposed relationships and found that efficacy beliefs about trait‐relevant tasks functioned as a mediator linking conscientious and supervisor ratings of performance.     

Social Networking Websites,Personality Ratings,and the Organizational Context: More Than Meets the Eye?1

We examined the psychometric properties of the Big Five personality traits assessed through social networking profiles in 2 studies consisting of 274 and 244 social networking website (SNW) users. First, SNW ratings demonstrated sufficient interrater reliability and internal consistency. Second, ratings via SNWs demonstrated convergent validity with self‐ratings of the Big Five traits. Third, SNW ratings correlated with job performance, hirability, and academic performance criteria; and the magnitude of these correlations was generally larger than for self‐ratings. Finally, SNW ratings accounted for significant variance in the criterion measures beyond self‐ratings of personality and cognitive ability. We suggest that SNWs may provide useful information for potential use in organizational research and practice, taking into consideration various legal and ethical issues.     

Cues to personality and health in the facial appearance of chimpanzees (Pan troglodytes)

Humans (Homo sapiens) and chimpanzees (Pan troglodytes) can extract socially-relevant information from the static, non-expressive faces of conspecifics. In humans, the face is a valid signal of both personality and health. Recent evidence shows that, like humans, chimpanzee faces also contain personality information, and that humans can accurately judge aspects of chimpanzee personality relating to extraversion from the face alone (Kramer, King, and Ward, 2011). These findings suggest the hypothesis that humans and chimpanzees share a system of personality and facial morphology for signaling socially-relevant traits from the face. We sought to test this hypothesis using a new group of chimpanzees. In two studies, we found that chimpanzee faces contained health information, as well as information of characteristics relating to extraversion, emotional stability, and agreeableness, using average judgments from pairs of individual photographs. In a third study, information relating to extraversion and health was also present in composite images of individual chimpanzees. We therefore replicate and extend previous findings using a new group of chimpanzees and demonstrate two methods for minimizing the variability associated with individual photographs. Our findings support the hypothesis that chimpanzees and humans share a personality signaling system.     

特质焦虑在面部表情前注意加工阶段的影响：来自ERP的证据

为探讨高特质焦虑者在前注意阶段对情绪刺激的加工模式以明确其情绪偏向性特点, 本研究采用偏差-标准反转Oddball范式探讨了特质焦虑对面部表情前注意加工的影响。结果发现: 对于低特质焦虑组, 悲伤面孔所诱发的早期EMMN显著大于快乐面孔, 而对于高特质焦虑组, 快乐和悲伤面孔所诱发的早期EMMN差异不显著。并且, 高特质焦虑组的快乐面孔EMMN波幅显著大于低特质焦虑组。结果表明, 人格特质是影响面部表情前注意加工的重要因素。不同于普通被试, 高特质焦虑者在前注意阶段对快乐和悲伤面孔存在相类似的加工模式, 可能难以有效区分快乐和悲伤情绪面孔。     

Accuracy in discrimination of self-reported cooperators using static facial information

People readily ascribe personality traits to others and believe that faces hold important guides to character. Here we examined the relationship between static facial appearance and self-reported cooperation/defection using the prisoner’s dilemma (N = 193). Study 1 combined face images of those self-reporting they would be most and least likely to cooperate. The composites of cooperators were seen as more cooperative than non-cooperators. Study 2 demonstrated accuracy with ratings of individual faces. Masculinity of face shape was negatively related to self-reported cooperation for men, but not women. Further, ratings of smile intensity were positively, but not significantly, related to self-reported cooperation. Overall, individuals appear able judge the potential of others to cooperate from static facial appearance alone at rates greater than chance.     

Individuation training with other‐race faces reduces preschoolers’ implicit racial bias: a link between perceptual and social representation of faces in children

The present study examined whether perceptual individuation training with other‐race faces could reduce preschool children's implicit racial bias. We used an ‘angry = outgroup’ paradigm to measure Chinese children's implicit racial bias against African individuals before and after training. In Experiment 1, children between 4 and 6 years were presented with angry or happy racially ambiguous faces that were morphed between Chinese and African faces. Initially, Chinese children demonstrated implicit racial bias: they categorized happy racially ambiguous faces as own‐race (Chinese) and angry racially ambiguous faces as other‐race (African). Then, the children participated in a training session where they learned to individuate African faces. Children's implicit racial bias was significantly reduced after training relative to that before training. Experiment 2 used the same procedure as Experiment 1, except that Chinese children were trained with own‐race Chinese faces. These children did not display a significant reduction in implicit racial bias. Our results demonstrate that early implicit racial bias can be reduced by presenting children with other‐race face individuation training, and support a linkage between perceptual and social representations of face information in children.     

Social influence modulates the neural computation of value

Social influence--individuals' tendency to conform to the beliefs and attitudes of others--has interested psychologists for decades. However, it has traditionally been difficult to distinguish true modification of attitudes from mere public compliance with social norms; this study addressed this challenge using functional neuroimaging. Participants rated the attractiveness of faces and subsequently learned how their peers ostensibly rated each face. Participants were then scanned using functional MRI while they rated each face a second time. The second ratings were influenced by social norms: Participants changed their ratings to conform to those of their peers. This social influence was accompanied by modulated engagement of two brain regions associated with coding subjective value--the nucleus accumbens and orbitofrontal cortex--a finding suggesting that exposure to social norms affected participants' neural representations of value assigned to stimuli. These findings document the utility of neuroimaging to demonstrate the private acceptance of social norms.     

Identity but not expression memory for unfamiliar faces is affected by ageing

We examined age‐related differences in memory for identity and emotional expression of unfamiliar faces. Younger and older adults were presented with happy and angry faces and were later asked to recognise the same faces displaying a neutral expression. When a face was recognised, they also had to remember what the initial expression of the face had been. In addition, states of awareness associated with both identity and expression memory were assessed with the remember/know/guess paradigm. Older adults showed less recollective experience than younger adults for identity but not for emotional expressions of the faces. This evidence indicates that age‐related differences in memory may depend on the nature of the to‐be‐remembered information, with emotional/social information being remembered as well in older as in younger adults.     

Face processing in different brain areas,and critical band masking

Neurophysiological evidence is described showing that some neurons in the macaque inferior temporal visual cortex have responses that are invariant with respect to the position, size, view, and spatial frequency of faces and objects, and that these neurons show rapid processing and rapid learning. Critical band spatial frequency masking is shown to be a property of these face‐selective neurons and of the human visual perception of faces. Which face or object is present is encoded using a distributed representation in which each neuron conveys independent information in its firing rate, with little information evident in the relative time of firing of different neurons. This ensemble encoding has the advantages of maximizing the information in the representation useful for discrimination between stimuli using a simple weighted sum of the neuronal firing by the receiving neurons, generalization, and graceful degradation. These invariant representations are ideally suited to provide the inputs to brain regions such as the orbitofrontal cortex and amygdala that learn the reinforcement associations of an individual's face, for then the learning, and the appropriate social and emotional responses generalize to other views of the same face. A theory is described of how such invariant representations may be produced by self‐organizing learning in a hierarchically organized set of visual cortical areas with convergent connectivity. The theory utilizes either temporal or spatial continuity with an associative synaptic modification rule. Another population of neurons in the cortex in the superior temporal sulcus encodes other aspects of faces such as face expression, eye‐gaze, face view, and whether the head is moving. These neurons thus provide important additional inputs to parts of the brain such as the orbitofrontal cortex and amygdala that are involved in social communication and emotional behaviour. Outputs of these systems reach the amygdala, in which face‐selective neurons are found, and also the orbitofrontal cortex, in which some neurons are tuned to face identity and others to face expression. In humans, activation of the orbitofrontal cortex is found when a change of face expression acts as a social signal that behaviour should change; and damage to the human orbitofrontal and pregenual cingulate cortex can impair face and voice expression identification, and also the reversal of emotional behaviour that normally occurs when reinforcers are reversed.     

Fear selectively modulates visual mental imagery and visual perception

Preferences for faces are thought to be the result of either general adaptations for mate selection, and thus influenced by sexual dimorphism, or mechanisms of general information processing and thus nonspecific to faces. If mate choice determines face preference then it should follow that the sex of a face should affect attractiveness judgements. To test this idea we used image morphing to generate three versions of face images: original, opposite sex, and antiface. First we established that the sex of the face was identifiable in our images. We then collected attractiveness ratings for the three face types. We found that attractiveness ratings to the original faces were correlated with, and did not differ significantly between, ratings to the opposite-sex faces. However, ratings for either the original or opposite face types were not correlated with and were significantly lower than ratings to the antifaces. Our findings failed to support the idea that attractiveness is related to sexual dimorphism in faces alone but suggest instead that other more generic factors influence preferences for all faces.     

Ensemble coding of faces occurs in children and develops dissociably from coding of individual faces

Ensemble coding allows adults to access useful information about average properties of groups, sometimes even in the absence of detailed representations of individual group members. This form of coding may emerge early in development with initial reports of ensemble coding for simple properties (size, numerosity) in young children and even infants. Here we demonstrate that ensemble coding of faces, which provides information about average properties of social groups, is already present in 6–8‐year‐old children. This access to average information increases with age from 6 to 18 years and its development is dissociable from age‐related improvements in the coding of individual face identities. This dissociation provides the first direct evidence that distinct processes underlie ensemble and individual coding of face identity, evidence that has been lacking from adult studies. More generally, our results add to the emerging evidence for impressively mature sensitivity to statistical properties of the visual environment in children. They indicate that children have access to gist information about social groups that may facilitate adaptive social behaviour.     

Using composite images to assess accuracy in personality attribution to faces

Several studies have demonstrated some accuracy in personality attribution using only visual appearance. Using composite images of those scoring high and low on a particular trait, the current study shows that judges perform better than chance in guessing others' personality, particularly for the traits conscientiousness and extraversion. This study also shows that attractiveness, masculinity and age may all provide cues to assess personality accurately and that accuracy is affected by the sex of both of those judging and being judged. Individuals do perform better than chance at guessing another's personality from only facial information, providing some support for the popular belief that it is possible to assess accurately personality from faces.     

When facial attractiveness is only skin deep

Whilst the relationship between aspects of facial shape and attractiveness has been extensively studied, few studies have investigated which characteristics of the surface of faces positively influence attractiveness judgments. As many researchers have proposed a link between attractiveness and traits that appear healthy, apparent health of facial skin might be a property of the surface of faces that positively influences attractiveness judgments. In experiment 1 we tested for a positive correlation between ratings of the apparent health of small skin patches (extracted from the left and right cheeks of digital face images) and ratings of the attractiveness of male faces. By using computer-graphics faces, in experiment 2 we aimed to establish if apparent health of skin influences male facial attractiveness independently of shape information. Results suggest that apparent health of facial skin is correlated both with ratings of male facial attractiveness (experiment 1) and with being a visual cue for judgments of the attractiveness of male faces (experiment 2). These findings underline the importance of controlling for the influence of visible skin condition in studies of facial attractiveness and are consistent with the proposal that attractive physical traits are those that positively influence others' perceptions of an individual's health.