Robin S. S. Kramer

Identity from Variation: Representations of Faces Derived from Multiple Instances.

Research in face recognition has tended to focus on discriminating between individuals, or telling people apart. It has recently become clear that it is also necessary to understand how images of the same person can vary, or telling people together. Learning a new face, and tracking its representation as it changes from unfamiliar to familiar, involves an abstraction of the variability in different images of that persons face. Here, we present an application of principal components analysis computed across different photos of the same person. We demonstrate that people vary in systematic ways, and that this variability is idiosyncratic-the dimensions of variability in one face do not generalize well to another. Learning a new face therefore entails learning how that face varies. We present evidence for this proposal and suggest that it provides an explanation for various effects in face recognition. We conclude by making a number of testable predictions derived from this framework.

Viewers extract the mean from images of the same person: A route to face learning

Research on ensemble encoding has found that viewers extract summary information from sets of similar items. When shown a set of four faces of different people, viewers merge identity information from the exemplars into a representation of the set average. Here, we presented sets containing unconstrained images of the same identity. In response to a subsequent probe, viewers recognized the exemplars accurately. However, they also reported having seen a merged average of these images. Importantly, viewers reported seeing the matching average of the set (the average of the four presented images) more often than a nonmatching average (an average of four other images of the same identity). These results were consistent for both simultaneous and sequential presentation of the sets. Our findings support previous research suggesting that viewers form representations of both the exemplars and the set average. Given the unconstrained nature of the photographs, we also provide further evidence that the average representation is invariant to several high-level characteristics.

Signals of personality and health: The contributions of facial shape, skin texture, and viewing angle.

To what extent does information in a persons face predict their likely behavior? There is increasing evidence for association between relatively neutral, static facial appearance and personality traits. By using composite images rendered from three dimensional (3D) scans of women scoring high and low on health and personality dimensions, we aimed to examine the separate contributions of facial shape, skin texture and viewing angle to the detection of these traits, while controlling for crucial posture variables. After controlling for such cues, participants were able to identify Agreeableness, Neuroticism, and Physical Health. For personality traits, we found a reliable laterality bias, in that the right side of the face afforded higher accuracy than the left. The separate contributions of shape and texture cues varied with the traits being judged. Our findings are consistent with signaling theories suggesting multiple channels to convey multiple messages.

Facial cosmetics have little effect on attractiveness judgments compared with identity

The vast majority of women in modern societies use facial cosmetics, which modify facial cues to attractiveness. However, the size of this increase remains unclear—how much more attractive are individuals after an application of cosmetics? Here, we utilised a ‘new statistics approach, calculating the effect size of cosmetics on attractiveness using a within-subjects design, and compared this with the effect size due to identity—that is, the inherent differences in attractiveness between people. Women were photographed with and without cosmetics, and these images were rated for attractiveness by a second group of participants. The proportion of variance in attractiveness explained by identity was much greater than the variance within models due to cosmetics. This result was unchanged after statistically controlling for the perceived amount of cosmetics that each model used. Although cosmetics increase attractiveness, the effect is small, and the benefits of cosmetics may be inflated in everyday thinking.

Miscalibrations in judgements of attractiveness with cosmetics

Women use cosmetics to enhance their attractiveness. How successful they are in doing so remains unknown—how do men and women respond to cosmetics use in terms of attractiveness? There are a variety of miscalibrations where attractiveness is concerned—often, what one sex thinks the opposite sex finds attractive is incorrect. Here, we investigated observer perceptions about attractiveness and cosmetics, as well as their understanding of what others would find attractive. We used computer graphic techniques to allow observers to vary the amount of cosmetics applied to a series of female faces. We asked observers to optimize attractiveness for themselves, for what they thought women in general would prefer, and what they thought men in general would prefer. We found that men and women agree on the amount of cosmetics they find attractive, but overestimate the preferences of women and, when considering the preferences of men, overestimate even more. We also find that models’ self-applied cosmetics are far in excess of individual preferences. These findings suggest that attractiveness perceptions with cosmetics are a form of pluralistic ignorance, whereby women tailor their cosmetics use to an inaccurate perception of others’ preferences. These findings also highlight further miscalibrations of attractiveness ideals.

Face averages enhance user recognition for smartphone security.

Our recognition of familiar faces is excellent, and generalises across viewing conditions. However, unfamiliar face recognition is much poorer. For this reason, automatic face recognition systems might benefit from incorporating the advantages of familiarity. Here we put this to the test using the face verification system available on a popular smartphone (the Samsung Galaxy). In two experiments we tested the recognition performance of the smartphone when it was encoded with an individual’s ‘face-average’ – a representation derived from theories of human face perception. This technique significantly improved performance for both unconstrained celebrity images (Experiment 1) and for real faces (Experiment 2): users could unlock their phones more reliably when the device stored an average of the user’s face than when they stored a single image. This advantage was consistent across a wide variety of everyday viewing conditions. Furthermore, the benefit did not reduce the rejection of imposter faces. This benefit is brought about solely by consideration of suitable representations for automatic face recognition, and we argue that this is just as important as development of matching algorithms themselves. We propose that this representation could significantly improve recognition rates in everyday settings.

Facial width-to-height ratio in a large sample of Commonwealth Games athletes.

Evidence that facial width-to-height ratio (FWHR) is a sexually dimorphic morphological measure is mixed. Research has also linked FWHR with aggression and other behavioral tendencies, at least in men. Again, other research has found no such relationship. Here, I tested for both possible relationships using a sample of 2,075 male and 1,406 female athletes from the Glasgow 2014 Commonwealth Games. Men showed significantly greater FWHRs than women, but this difference could be attributed to differences in body size. In addition, I found greater FWHRs in men who competed in sports involving physical contact and those stereotyped as more masculine. Again, these results could be attributed to differences in body size between categories. For women, no differences in FWHR were found regarding the amount of contact involved in a sport and how that sport was stereotyped. Finally, the FWHRs of athletes showed no relationship with the amount of aggression and related traits that were judged as required for success in those sports, although FWHRs did correlate with perceived endurance demands in women. Therefore, in a large sample of athletes, the sex difference in FWHR could be attributed to body size, and little support was found for the predicted links between this facial measure and behavior.

Natural variability is essential to learning new faces

ABSTRACT We learn new faces throughout life, for example in everyday settings like watching TV. Recent research has shown that image variability is key to this ability: if we learn a new face over highly variable images, we are better able to recognize that person in novel pictures. Here we asked people to watch TV shows they had not seen before, and then tested their ability to recognize the actors. Some participants watched TV shows in the conventional manner, whereas others watched them upside down or contrast-reversed. Image variability is equivalent across these conditions, and yet we observed that viewers were unable to learn the faces upside down or contrast-reversed—even when tested in the same format as learning. We conclude that variability is a necessary, but not sufficient, condition for face learning. Instead, mechanisms underlying this process are tuned to extract useful information from variability falling within a critical range that corresponds to natural, everyday variation.

Robust Social Categorization Emerges From Learning the Identities of Very Few Faces.

Viewers are highly accurate at recognizing sex and race from faces-though it remains unclear how this is achieved. Recognition of familiar faces is also highly accurate across a very large range of viewing conditions, despite the difficulty of the problem. Here we show that computation of sex and race can emerge incidentally from a system designed to compute identity. We emphasize the role of multiple encounters with a small number of people, which we take to underlie human face learning. We use highly variable everyday ambient images of a few people to train a Linear Discriminant Analysis (LDA) model on identity. The resulting model has human-like properties, including a facility to cohere previously unseen ambient images of familiar (trained) people-an ability which breaks down for the faces of unknown (untrained) people. The first dimension created by the identity-trained LDA classifies both familiar and unfamiliar faces by sex, and the second dimension classifies faces by race-even though neither of these categories was explicitly coded at learning. By varying the numbers and types of face identities on which a further series of LDA models were trained, we show that this incidental learning of sex and race reflects covariation between these social categories and face identity, and that a remarkably small number of identities need be learnt before such incidental dimensions emerge. The task of learning to recognize familiar faces is sufficient to create certain salient social categories. (PsycINFO Database Record

InterFace: A software package for face image warping, averaging, and principal components analysis

We describe InterFace, a software package for research in face recognition. The package supports image warping, reshaping, averaging of multiple face images, and morphing between faces. It also supports principal components analysis (PCA) of face images, along with tools for exploring the “face space” produced by PCA. The package uses a simple graphical user interface, allowing users to perform these sophisticated image manipulations without any need for programming knowledge. The program is available for download in the form of an app, which requires that users also have access to the (freely available) MATLAB Runtime environment.