Physical and Psychological Representations of Faces: Evidence FromMorphing

Models of face recognition and classification often adopt a framework in which faces are represented as points in a multi-dimensional space. This psychological face space organizes the faces according to similarity and makes predictions for representational theories of faces. A variety of image-processing techniques have been used to create novel stimuli in this space that represent the average of a population or make a face appear more distinctive. The current research examined the relation between the stimuli created by these image-processing techniques and the underlying psychological representation as measured by multidimensional scaling (MDS) procedures. Morphing procedures were used to create 16 faces that were embedded in a set of 84 other faces. Similarity ratings between all possible pairs of faces were collected, and the data were analyzed using MDS procedures. Dimensions that emerged from the MDS solution included age, race, adiposity, and facial hair. In the MDS space, the morphs appeared more typical than the parents, as predicted by the geometric model. A number of biases were examined, including the tendency of the morphs to be less typical than predicted, which may be attributed to the effects of density near the center of face space. In addition, age and facial-adiposity biases were found. The results support the use of the face-space framework for models of face recognition, although image-processing techniques that are designed to create novel stimuli in this space may introduce systematic biases.     

Touching on face space: Comparing visual and haptic processing of face shapes

The idea that faces are represented within a structured face space (Valentine Quarterly Journal of Experimental Psychology 43: 161–204, 1991) has gained considerable experimental support, from both physiological and perceptual studies. Recent work has also shown that faces can even be recognized haptically—that is, from touch alone. Although some evidence favors congruent processing strategies in the visual and haptic processing of faces, the question of how similar the two modalities are in terms of face processing remains open. Here, this question was addressed by asking whether there is evidence for a haptic face space, and if so, how it compares to visual face space. For this, a physical face space was created, consisting of six laser-scanned individual faces, their morphed average, 50 %-morphs between two individual faces, as well as 50 %-morphs of the individual faces with the average, resulting in a set of 19 faces. Participants then rated either the visual or haptic pairwise similarity of the tangible 3-D face shapes. Multidimensional scaling analyses showed that both modalities extracted perceptual spaces that conformed to critical predictions of the face space framework, hence providing support for similar processing of complex face shapes in haptics and vision. Despite the overall similarities, however, systematic differences also emerged between the visual and haptic data. These differences are discussed in the context of face processing and complex-shape processing in vision and haptics.     

“Just another pretty face”: A multidimensional scaling approach to face attractiveness and variability

Findings on both attractiveness and memory for faces suggest that people should perceive more similarity among attractive than among unattractive faces. A multidimensional scaling approach was used to test this hypothesis in two studies. In Study 1, we derived a psychological face space from similarity ratings of attractive and unattractive Caucasian female faces. In Study 2, we derived a face space for attractive and unattractive male faces of Caucasians and non-Caucasians. Both studies confirm that attractive faces are indeed more tightly clustered than unattractive faces in people’s psychological face spaces. These studies provide direct and original support for theoretical assumptions previously made in the face space and face memory literatures.     

Encoding differences affect the number and precision of own-race versus other-race faces stored in visual working memory

Other-race faces are discriminated and recognized less accurately than own-race faces. Despite a wealth of research characterizing this other-race effect (ORE), little is known about the nature of the representations of own-race versus other-race faces. This is because traditional measures of this ORE provide a binary measure of discrimination or recognition (correct/incorrect), failing to capture potential variation in the quality of face representations. We applied a novel continuous-response paradigm to independently measure the number of own-race and other-race face representations stored in visual working memory (VWM) and the precision with which they are stored. Participants reported target own-race or other-race faces on a circular face space that smoothly varied along the dimension of identity. Using probabilistic mixture modeling, we found that following ample encoding time, the ORE is attributable to differences in the probability of a face being maintained in VWM. Reducing encoding time, a manipulation that is more sensitive to encoding limitations, caused a loss of precision or an increase in variability of VWM for other-race but not own-race faces. These results suggest that the ORE is driven by the inefficiency with which other-race faces are rapidly encoded in VWM and provide novel insights about how perceptual experience influences the representation of own-race and other-race faces in VWM.

     

Locating attractiveness in the face space: Faces are more attractive when closer to <Emphasis Type="Italic">their</Emphasis> group prototype

Face attractiveness relates positively to the mathematical averageness of a face, but how close attractive faces of varying groups are to their own and to other-group prototypes in the face space remains unclear. In two studies, we modeled the locations of attractive and unattractive Caucasian, Asian, and African faces in participants’ face space using multidimensional scaling analysis. In all three sets of faces, facial attractiveness significantly increased with the absolute proximity of a face to its group prototype. In the case of Caucasian and African faces (Study 1), facial attractiveness also tended to increase with the absolute proximity of a face to the other-group prototype. However, this association was at best marginal, and it became clearly nonsignificant when distance to the own-group prototype was controlled for. Thus, the present research provides original evidence that average features of faces contribute to increasing their attractiveness, but only when these features are average to the group to which a face belongs. The present research also offers further support to face space models of people’s mental representations of faces.     

The organization of conspecific face space in nonhuman primates

Humans and chimpanzees demonstrate numerous cognitive specializations for processing faces, but comparative studies with monkeys suggest that these may be the result of recent evolutionary adaptations. The present study utilized the novel approach of face space, a powerful theoretical framework used to understand the representation of face identity in humans, to further explore species differences in face processing. According to the theory, faces are represented by vectors in a multidimensional space, the centre of which is defined by an average face. Each dimension codes features important for describing a face's identity, and vector length codes the feature's distinctiveness. Chimpanzees and rhesus monkeys discriminated male and female conspecifics’ faces, rated by humans for their distinctiveness, using a computerized task. Multidimensional scaling analyses showed that the organization of face space was similar between humans and chimpanzees. Distinctive faces had the longest vectors and were the easiest for chimpanzees to discriminate. In contrast, distinctiveness did not correlate with the performance of rhesus monkeys. The feature dimensions for each species’ face space were visualized and described using morphing techniques. These results confirm species differences in the perceptual representation of conspecific faces, which are discussed within an evolutionary framework.     

Towards an exemplar model of face processing: the effects of race and distinctiveness.

Valentine (1991a, 1991b) described a theoretical framework for face recognition in which faces are encoded as locations in a multidimensional space. It was argued that this approach could provide a unified account of the effects of distinctiveness, inversion, and race on face recognition. In this paper we evaluate the ability of this theoretical framework to account for the effects of distinctiveness and race in four experiments in which white British and Japanese faces served as stimuli and both white British and Japanese students acted as subjects. In a recognition memory experiment the expected "own-race bias" was observed as a Race of Subject x Race of Face interaction. Distinctive faces were recognized more accurately than typical faces, but the effect of distinctiveness did not interact with the race of face or the race of subject. Typical faces were classified faster than distinctive faces in a task in which intact faces had to be distinguished from jumbled faces, as found in earlier work, and the effect of distinctiveness did not interact with the race of face or race of subject. In contrast, a task in which subjects classified faces according to their race did show a greater effect of distinctiveness for own-race faces. The results are discussed in relation to the two specific models within the multidimensional space framework identified by Valentine (1991a): a purely exemplar-based model and a norm-based coding model. It is argued that these results are more easily accommodated in terms of a purely exemplar-based model. Some conceptual problems in applying the norm-based coding model to the effect of race are discussed.     

Searching for faces in scrambled scenes

While much is known about how faces are recognized, little is known about how a face is first detected. Five face-detection experiments investigated how faces are localized and detected among either scrambled natural backgrounds or inverted faces. The first two experiments revealed that faces pop out among the former but must be searched for serially among the latter. Two subsequent experiments investigated what properties of a face allow for this pop-out. Colour removal or reversal, blurring, and inversion all had small effects on the visual search slope. The only transformation that reduced the parallel nature of face detection was luminance reversal. These results are considered in the light of models of face processing and systems for automatic face detection.     

Face‐space‐R: Towards a unified account of face recognition

Versions of the “face space” are considered and built upon to develop an explicitly defined model of face recognition based on stimulus generalization that is similar to models of animal learning. This face‐space‐R model is implemented using realistic numbers of known faces. The model is able to account for distinctiveness, caricature, and race effects. It also predicts which faces will be falsely recognized and accounts for mirror effects. The application of the model to face learning and development is considered, as well as the effects of brief presentation. By varying parameters of the model, it is possible to match its performance to that of humans, leading to an estimate of the dimensionality of face space (of between 15 and 22 dimensions for same‐race faces).     

Identification of own-race and other-race faces: Implications for the representation of race in face space

Own-race faces are recognized more easily than faces of a different, unfamiliar race. According to the multidimensional space (MDS) framework, the poor discriminability of other-race faces is due to their being more densely clustered in face space than own-race faces. Multidimensional scaling analyses of similarity ratings (Caucasian participants, n = 22) showed that other-race (Chinese) faces are more densely clustered in face space. We applied a formal model to test whether the spatial location of face stimuli could account for identification accuracy of another group of Caucasian participants (n = 30). As expected, own-race (Caucasian) faces were identified more accurately (higher hit rate, lower false alarms, and higher A) than other-race faces, which were more densely clustered than ownrace faces. A quantitative model successfully predicted identification performance from the spatial locations of the stimuli. The results are discussed in relation to the standard MDS account of race effects and also an alternative “race-feature” hypothesis.     

On the other side of the mean: the perception of dissimilarity in human faces

We created a 'face space' using a laser-scan representation of faces. In this space, a caricature can be made by moving a face away from the average face, along the line connecting the particular face to the average face. Here, we move the face along this line in the other direction, proceeding through the mean and 'out the other side'. This results in a face that is 'opposite', in a computational sense, to the original face. We morphed several faces into their anti-faces and sampled the morph trajectory in five discrete steps. We then collected similarity ratings from human participants for all possible pairs of morphed faces to determine how the distances in the 'physical face space' related to the distances in the 'psychological face space'. The data indicate that there is a perceptual discontinuity of face identity as the face crosses over to the 'other side of the mean'. We consider these results in the context of face-space models of human face processing.     

The representations of spacing and part-based information are associated for upright faces but dissociated for objects: Evidence from individual differences

Considerable evidence suggests that qualitatively different processes are involved in the perception of faces and objects. According to a central hypothesis, the extraction of information about the spacing among face parts (e.g., eyes and mouth) is a primary function of face processing mechanisms that is dissociated from the extraction of information about the shape of these parts. Here, we used an individual-differences approach to test whether the shape of face parts and the spacing among them are indeed processed by dissociated mechanisms. To determine whether the pattern of findings that we reveal is unique for upright faces, we also presented similarly manipulated nonface stimuli. Subjects discriminated upright or inverted faces or houses that differed in parts or spacing. Only upright faces yielded a large positive correlation across subjects between performance on the spacing and part discrimination tasks. We found no such correlation for inverted faces or houses. Our findings suggest that face parts and spacing are processed by associated mechanisms, whereas the parts and spacing of nonface objects are processed by distinct mechanisms. These results may be consistent with the idea that faces are special, in that they are processed as nondecomposable wholes.     

Face detection differs from categorization: Evidence from visual search in natural scenes

In this study, we examined whether the detection of frontal, ¾, and profile face views differs from their categorization as faces. In Experiment 1, we compared three tasks that required observers to determine the presence or absence of a face, but varied in the extents to which participants had to search for the faces in simple displays and in small or large scenes to make this decision. Performance was equivalent for all of the face views in simple displays and small scenes, but it was notably slower for profile views when this required the search for faces in extended scene displays. This search effect was confirmed in Experiment 2, in which we compared observers’ eye movements with their response times to faces in visual scenes. These results demonstrate that the categorization of faces at fixation is dissociable from the detection of faces in space. Consequently, we suggest that face detection should be studied with extended visual displays, such as natural scenes.     

Simulating the 'other-race' effect with autoassociative neural networks: further evidence in favor of the face-space model

Other-race (OR) faces are less accurately recognized than same-race (SR) faces, but faster classified by race. This phenomenon has often been reported as the 'other-race' effect (ORE). Valentine (1991 Quarterly Journal of Experimental Psychology A: Human Experimental Psychology 43 161-204) proposed a theoretical multidimensional face-space model that explained both of these results, in terms of variations in exemplar density between races. According to this model, SR faces are more widely distributed across the dimensions of the space than OR faces. However, this model does not quantify nor state the dimensions coded within this face space. The aim of the present study was to test the face-space explanation of the ORE with neural network simulations by quantifying its dimensions. We found the predicted density properties of Valentine's framework in the face-projection spaces of the autoassociative memories. This was supported by an interaction for exemplar density between the race of the learned face set and the race of the faces. In addition, the elaborated face representations showed optimal responses for SR but not for OR faces within SR face spaces when explored at the individual level, as gender errors occurred significantly more often in OR than in SR face-space representations. Altogether, our results add further evidence in favor of a statistical exemplar density explanation of the ORE as suggested by Valentine, and question the plausibility of such coding for faces in the framework of recent neuroimaging studies.     

知觉场理论对面孔知觉年龄偏差效应的解释

使用融合面孔范式和倒置面孔范式来研究面孔知觉的年龄偏差效应,检验知觉场能否作为面孔整体加工的指标,并以此发展年龄偏差的整体加工解释。结果发现：（1）成人和儿童在加工正立面孔时的知觉场均大于加工倒置面孔时的知觉场;（2）在加工正立面孔时,成人加工本年龄面孔比加工他年龄面孔的知觉场更大。上述结果表明：（1）知觉场大小可以作为面孔整体加工的指标,且受面孔朝向的影响;（2）知觉场假设可以解释面孔的年龄偏差效应。     

知觉场理论对面孔知觉年龄偏差效应的解释

使用融合面孔范式和倒置面孔范式来研究面孔知觉的年龄偏差效应,检验知觉场能否作为面孔整体加工的指标,并以此发展年龄偏差的整体加工解释。结果发现：（1）成人和儿童在加工正立面孔时的知觉场均大于加工倒置面孔时的知觉场;（2）在加工正立面孔时,成人加工本年龄面孔比加工他年龄面孔的知觉场更大。上述结果表明：（1）知觉场大小可以作为面孔整体加工的指标,且受面孔朝向的影响;（2）知觉场假设可以解释面孔的年龄偏差效应。     

Exploring the perceptual spaces of faces, cars and birds in children and adults

While much developmental research has focused on the strategies that children employ to recognize faces, less is known about the principles governing the organization of face exemplars in perceptual memory. In this study, we tested a novel, child-friendly paradigm for investigating the organization of face, bird and car exemplars. Children ages 3-4, 5-6, 7-8, 9-10, 11-12 and adults were presented with 50/50 morphs of typical and atypical face, bird and car parent images. Participants were asked to judge whether the 50/50 morph more strongly resembled the typical or the atypical parent image. Young and older children and adults showed a systematic bias to the atypical faces and birds, but no bias toward the atypical cars. Collectively, these findings argue that by the age of 3, children encode and organize faces, birds and cars in a perceptual space that is strikingly similar to that of adults. Category organization for both children and adults follows Krumhansl's (1978) distance-density principle in which the similarity between two exemplars is jointly determined by their physical appearance and the density of neighboring exemplars in the perceptual space.     

Biological sex determines whether faces look real

Judging whether a face is real or artificial can be done relatively rapidly and accurately, even when visual information is substantially impoverished. The perception of animacy in the face also has several interesting properties that may reflect both the underlying “tuning” of face space to preferentially represent real face appearance and the diagnosticity of individual features for categorizing faces as animate or inanimate. In the current study, we examined how sex categories interact with animacy perception by separately characterizing animacy judgements as a function of stimulus sex. We find that stimulus sex affects subjective ratings of animacy and sex categorization of real and artificial faces. Specifically, female faces look more artificial and artificial faces look more female. We discuss our results in terms of the ecology of real and artificial faces and the possible role of visual experience with artificial female faces, and the objectification of female faces.     

Visual adaptation of the perception of “life”: Animacy is a basic perceptual dimension of faces

One critical component of understanding another’s mind is the perception of “life” in a face. However, little is known about the cognitive and neural mechanisms underlying this perception of animacy. Here, using a visual adaptation paradigm, we ask whether face animacy is (1) a basic dimension of face perception and (2) supported by a common neural mechanism across distinct face categories defined by age and species. Observers rated the perceived animacy of adult human faces before and after adaptation to (1) adult faces, (2) child faces, and (3) dog faces. When testing the perception of animacy in human faces, we found significant adaptation to both adult and child faces, but not dog faces. We did, however, find significant adaptation when morphed dog images and dog adaptors were used. Thus, animacy perception in faces appears to be a basic dimension of face perception that is species specific but not constrained by age categories.     

An Investigation of the Contact Hypothesis of the Own-race Bias in Face Recognition

Although previous studies have demonstrated that faces of one's own race are recognized more accurately than are faces of other races, the theoretical basis of this effect is not clearly understood at present. The experiment reported in this paper tested the contact hypothesis of the own-race bias in face recognition using a cross-cultural design. Four groups of subjects were tested for their recognition of distinctive and typical own-race and other-race faces: (1) black Africans who had a high degree of contact with white faces, (2) black Africans who had little or no contact with white faces, (3) white Africans who had a high degree of contact with black faces, and (4) white Britons who had little contact with black faces. The results showed that although on the whole subjects recognized own-race faces more accurately and more confidently than they recognized other-race faces, the own-race bias in face recognition was significantly smaller among the high-contact subjects than it was among the low-contact subjects. Also, although high-contact black and white subjects showed significant main effects of distinctiveness in their recognition of faces of both races, low-contact black and white subjects showed significant main effects of distinctiveness only in their recognition of own-race faces. It is argued that these results support the contact hypothesis of the own-race bias in face recognition and Valentine's multidimensional space (MDS) framework of face encoding.