Neurocomputational bases of object and face recognition

Identification of partially presented meaningless patterns: effect of completeness and distinctiveness

The role of parts versus that of wholes in a visual perception has been debated for a century as two opposite approaches, namely, an analytic and holistic. In two psychophysical experiments we investigated whether the stimulus completeness or distinctiveness is essential for identification of the partially presented patterns under brief presentation conditions. For this purpose, a special class of stimuli was constructed in such a way that the patterns could be divided into informative and redundant parts. The first experiment clearly demonstrated the importance of the redundant part for effective pattern identification for the majority of subjects. The second experiment revealed the direct dependence of identification accuracy of the patterns on their completeness (2, 3, 4, 5, or 6 elements). Familiarisation of subjects with the test stimuli influenced the strength of this dependence.

What makes faces special?

What may be special about faces, compared to non-face objects, is that their neural representation may be fundamentally spatial, e.g., Gabor-like. Subjects matched a sequence of two filtered images, each containing every other combination of spatial frequency and orientation, of faces or non-face 3D blobs, judging whether the person or blob was the same or different. On a match trial, the images were either identical or complementary (containing the remaining spatial frequency and orientation content). Relative to an identical pair of images, a complementary pair of faces, but not blobs, reduced matching accuracy and released fMRI adaptation in the fusiform face area.

Processing of facial identity and expression: a psychophysical, physiological, and computational perspective

A deeper understanding of how the brain processes visual information can be obtained by comparing results from complementary fields such as psychophysics, physiology, and computer science. In this chapter, empirical findings are reviewed with regard to the proposed mechanisms and representations for processing identity and emotion in faces. Results from psychophysics clearly show that faces are processed by analyzing component information (eyes, nose, mouth, etc.) and their spatial relationship (configural information). Results from neuroscience indicate separate neural systems for recognition of identity and facial expression. Computer science offers a deeper understanding of the required algorithms and representations, and provides computational modeling of psychological and physiological accounts. An interdisciplinary approach taking these different perspectives into account provides a promising basis for better understanding and modeling of how the human brain processes visual information for recognition of identity and emotion in faces.

Faces are "spatial"--holistic face perception is supported by low spatial frequencies

Faces are perceived holistically, a phenomenon best illustrated when the processing of a face feature is affected by the other features. Here, the authors tested the hypothesis that the holistic perception of a face mainly relies on its low spatial frequencies. Holistic face perception was tested in two classical paradigms: the whole-part advantage (Experiment 1) and the composite face effect (Experiments 2-4). Holistic effects were equally large or larger for low-pass filtered faces as compared to full-spectrum faces and significantly larger than for high-pass filtered faces. The disproportionate composite effect found for low-pass filtered faces was not observed when holistic perception was disrupted by inversion (Experiment 3). Experiment 4 showed that the composite face effect was enhanced only for low spatial frequencies, but not for intermediate spatial frequencies known be critical for face recognition. These findings indicate that holistic face perception is largely supported by low spatial frequencies. They also suggest that holistic processing precedes the analysis of local features during face perception.

Learning from humans: computational modeling of face recognition

In this paper, we propose a computational architecture of face recognition based on evidence from cognitive research. Several recent psychophysical experiments have shown that humans process faces by a combination of configural and component information. Using an appearance-based implementation of this architecture based on low-level features and their spatial relations, we were able to model aspects of human performance found in psychophysical studies. Furthermore, results from additional computational recognition experiments show that our framework is able to achieve excellent recognition performance even under large view rotations. Our interdisciplinary study is an example of how results from cognitive research can be used to construct recognition systems with increased performance. Finally, our modeling results also make new experimental predictions that will be tested in further psychophysical studies, thus effectively closing the loop between psychophysical experimentation and computational modeling.

Electrophysiological correlates of facial decision: Insights from upright and upside-down Mooney-face perception

We investigated the ERP correlates of the subjective perception of upright and upside-down ambiguous pictures as faces using two-tone Mooney stimuli in an explicit facial decision task (deciding whether a face is perceived or not in the display). The difficulty in perceiving upside-down Mooneys as faces was reflected by both lower rates of "Face" responses and delayed "Face" reaction times for upside-down relative to upright stimuli. The N170 was larger for the stimuli reported as "faces". It was also larger for the upright than the upside-down stimuli only when they were reported as faces. Furthermore, facial decision as well as stimulus orientation effects spread from 140-190 ms to 390-440 ms. The behavioural delay in 'Face' responses to upside-down stimuli was reflected in ERPs by later effect of facial decision for upside-down relative to upright Mooneys over occipito-temporal electrodes. Moreover, an orientation effect was observed only for the stimuli reported as faces; it yielded a marked hemispheric asymmetry, lasting from 140-190 ms to 390-440 ms post-stimulus onset in the left hemisphere and from 340-390 to 390-440 ms only in the right hemisphere. Taken together, the results supported a preferential involvement of the right hemisphere in the detection of faces, whatever their orientation. By contrast, the early orientation effect in the left hemisphere suggested that upside-down Mooney stimuli were processed as non face objects until facial decision was reached in this hemisphere. The present data show that face perception involves not only spatially but also temporally distributed activities in occipito-temporal regions.

Portraits or People? Distinct Representations of Face Identity in the Human Visual Cortex

Humans can identify individual faces under different viewpoints, even after a single encounter. We determined brain regions responsible for processing face identity across view changes after variable delays with several intervening stimuli, using event-related functional magnetic resonance imaging during a long-term repetition priming paradigm. Unfamiliar faces were presented sequentially either in a frontal or three-quarter view. Each face identity was repeated once after an unpredictable lag, with either the same or another viewpoint. Behavioral data showed significant priming in response time, irrespective of view changes. Brain imaging results revealed a reduced response in the lateral occipital and fusiform cortex with face repetition. Bilateral face-selective fusiform areas showed view-sensitive repetition effects, generalizing only from three-quarter to front-views. More medial regions in the left (but not in the right) fusiform showed repetition effects across all types of viewpoint changes. These results reveal that distinct regions within the fusiform cortex hold view-sensitive or view-invariant traces of novel faces, and that face identity is represented in a view-sensitive manner in the functionally defined face-selective areas of both hemispheres. In addition, our finding of a better generalization after exposure to a 3/4-view than to a front-view demonstrates for the first time a neural substrate in the fusiform cortex for the common recognition advantage of three-quarter faces. This pattern provides new insights into the nature of face representation in the human visual system.

Familiarity enhances invariance of face representations in human ventral visual cortex: fMRI evidence

Face recognition across different viewing conditions is strongly improved by familiarity. In the present study, we tested the hypothesis that the neural basis of this effect is a less view-dependent representation of familiar faces in ventral visual cortex by assessing priming-related fMRI repetition effects. 15 healthy volunteers made male/female judgements on familiar (famous) and unfamiliar (novel) faces preceded by the same image, a different image of the same face, or another (unprimed) face. Reaction times revealed priming by same and different images independent of familiarity and more pronounced for same than different images. In the imaging data, a main effect of prime condition was found in bilateral fusiform and orbitofrontal regions. A right anterior fusiform region expressed stronger response decreases to repetition of familiar than unfamiliar faces. Bilateral mid-fusiform areas showed stronger response decreases to repetition of same than different images. A regions-of-interest analysis focussing specifically on face responsive regions suggested differences in the degree of image dependency across fusiform cortex. Collapsing across familiarity, there was greater image dependency of repetition effects in right than left anterior fusiform, replicating previous imaging findings obtained with common objects. For familiar faces alone, there was greater generalisation of repetition effects over different images in anterior than middle fusiform. This suggests a role of anterior fusiform cortex in coding image-independent representations of familiar faces.

Face recognition is affected by similarity in spatial frequency range to a greater degree than within-category object recognition

Previous studies have suggested that face identification is more sensitive to variations in spatial frequency content than object recognition, but none have compared how sensitive the 2 processes are to variations in spatial frequency overlap (SFO). The authors tested face and object matching accuracy under varying SFO conditions. Their results showed that object recognition was more robust to SFO variations than face recognition and that the vulnerability of faces was not due to reliance on configural processing. They suggest that variations in sensitivity to SFO help explain the vulnerability of face recognition to changes in image format and the lack of a middle-frequency advantage in object recognition.

Face perception: domain specific, not process specific

Evidence that face perception is mediated by special cognitive and neural mechanisms comes from fMRI studies of the fusiform face area (FFA) and behavioral studies of the face inversion effect. Here, we used these two methods to ask whether face perception mechanisms are stimulus specific, process specific, or both. Subjects discriminated pairs of upright or inverted faces or house stimuli that differed in either the spatial distance among parts (configuration) or the shape of the parts. The FFA showed a much higher response to faces than to houses, but no preference for the configuration task over the part task. Similarly, the behavioral inversion effect was as large in the part task as the configuration task for faces, but absent in both part and configuration tasks for houses. These findings indicate that face perception mechanisms are not process specific for parts or configuration but are domain specific for face stimuli per se.

Representation of Regular and Irregular Shapes in Macaque Inferotemporal Cortex

We determined the degree to which the response modulation of macaque inferior temporal (IT) neurons corresponds to perceptual versus physical shape similarities. IT neurons were tested with four groups of shapes. One group consisted of variations of simple, symmetrical (i.e. regular) shapes that differed in nonaccidental properties (NAPs, i.e. viewpoint-invariant), such as curved versus straight contours. The second and third groups were composed of, respectively, simple and complex asymmetrical (i.e. irregular) shapes, all with curved contours. A fourth group consisted of simple, asymmetrical shapes, but with straight (corners) instead of curved contours. The neural modulations were greater for the shapes differing in NAPs than for the shapes differing in the configuration of the convexities and concavities. Multidimensional scaling showed that a population code of the neural activity could readily distinguish the four shape groups. This pattern of neural modulation was strongly manifested in the results of a sorting task by human subjects but could not be predicted using current image-based models (i.e. pixel energies, V1-like Gabor-jet filtering and HMAX). The representation of shape in IT thus exceeds a mere faithful representation of physical reality, by emphasizing perceptually salient features relevant for essential categorizations.

BOLD repetition decreases in object-responsive ventral visual areas depend on spatial attention

Functional imaging studies of priming-related repetition phenomena have become widely used to study neural object representation. Although blood oxygenation level-dependent (BOLD) repetition decreases can sometimes be observed without awareness of repetition, any role for spatial attention in BOLD repetition effects remains largely unknown. We used fMRI in 13 healthy subjects to test whether BOLD repetition decreases for repeated objects in ventral visual cortices depend on allocation of spatial attention to the prime. Subjects performed a size-judgment task on a probe object that had been attended or ignored in a preceding prime display of 2 lateralized objects. Reaction times showed faster responses when the probe was the same object as the attended prime, independent of the view tested (identical vs. mirror image). No behavioral effect was evident from unattended primes. BOLD repetition decreases for attended primes were found in lateral occipital and fusiform regions bilaterally, which generalized across identical and mirror-image repeats. No repetition decreases were observed for ignored primes. Our results suggest a critical role for attention in achieving visual representations of objects that lead to both BOLD signal decreases and behavioral priming on repeated presentation.

Forward masking of faces by spatially quantized random and structured masks: On the roles of wholistic configuration, local features, and spatial-frequency spectra in perceptual identification

The forward masking of faces by spatially quantized masking images was studied. Masks were used in order to exert different types of degrading effects on the early representations in facial information processing. Three types of source images for masks were used: Same-face images (with regard to targets), different-face images, and random Gaussian noise that was spectrally similar to facial images. They were all spatially quantized over the same range of quantization values. Same-face masks had virtually no masking effect at any of the quantization values. Different-face masks had strong masking effects only with fine-scale quantization, but led to the same efficiency of recognition as in the same-face mask condition with the coarsest quantization. Moreover, compared with the noise-mask condition, coarsely quantized different-face masks led to a relatively facilitated level of recognition efficiency. The masking effect of the noise mask did not vary significantly with the coarseness of quantization. The results supported neither a local feature processing account, nor a generalized spatial-frequency processing account, but were consistent with the microgenetic configuration-processing theory of face recognition. Also, the suitability of a spatial quantization technique for image configuration processing research has been demonstrated.

Configural face encoding and spatial frequency information

Configural relations and a critical band of spatial frequencies (SFs) in the middle range are particularly important for face recognition. We report the results of four experiments in which the relationship between these two types of information was examined. In Experiments 1, 2A, and 2B, the face inversion effect (FIE) was used to probe configural face encoding. Recognition of upright and inverted faces and nonface objects was measured in four conditions: a no-filter condition and three SF conditions (low, medium, and high frequency). We found significant FIEs of comparable magnitudes for all frequency conditions. In Experiment 3, discrimination of faces on the basis of either configural or featural modifications was measured under the same four conditions. Although the ability to discriminate configural modifications was superior in the medium-frequency condition, so was the ability to discriminate featural modifications. We conclude that the band of SF that is critical for face recognition does not contribute preferentially to configural encoding.

The body-inversion effect

Researchers argue that faces are recognized via the configuration of their parts. An important behavioral finding supporting this claim is the face-inversion effect, in which inversion impairs recognition of faces more than nonface objects. Until recently, faces were the only class of objects producing the inversion effect for untrained individuals. This study investigated whether the inversion effect extends to human body positions, a class of objects whose exemplars are structurally similar to each other. Three experiments compared the recognition of upright and inverted faces, houses, and body positions using a forced-choice, same/different paradigm. For both reaction time and error data, the recognition of possible human body postures was more affected by inversion than the recognition of houses. Further, the recognition of possible human body postures and recognition of faces showed similar effects of inversion. The inversion effect was diminished for impossible body positions that violated the biomechanical constraints of human bodies. These data suggest that human body positions, like faces, may be processed configurally by untrained viewers.

Face recognition is robust with incongruent image resolution: relationship to security video images

Identifying a criminal captured on conventional security video typically requires matching poor-quality video footage against a high-quality photograph. The authors examined the consequence of such a large discrepancy in image quality. Recognition and matching performance of this incongruent-quality condition was compared with that of a congruent one, in which a high-quality photograph was reduced to a low-quality video. Recognition memory was little affected by this manipulation, whereas matching performance of the incongruent condition enjoyed occasional advantage. The results show that person identification can tolerate a large discrepancy between image qualities of matching stimuli when one of the images is of poor quality.

Effects of geometric distortions on face-recognition performance

The importance of 'configural' processing for face recognition is now well established, but it remains unclear precisely what it entails. Through four experiments we attempted to clarify the nature of configural processing by investigating the effects of various affine transformations on the recognition of familiar faces. Experiment 1 showed that recognition was markedly impaired by inversion of faces, somewhat impaired by shearing or horizontally stretching them, but unaffected by vertical stretching of faces to twice their normal height. In experiment 2 we investigated vertical and horizontal stretching in more detail, and found no effects of either transformation. Two further experiments were performed to determine whether participants were recognising stretched faces by using configural information. Experiment 3 showed that nonglobal vertical stretching of faces (stretching either the top or the bottom half while leaving the remainder undistorted) impaired recognition, implying that configural information from the stretched part of the face was influencing the process of recognition--ie that configural processing involves global facial properties. In experiment 4 we examined the effects of Gaussian blurring on recognition of undistorted and vertically stretched faces. Faces remained recognisable even when they were both stretched and blurred, implying that participants were basing their judgments on configural information from these stimuli, rather than resorting to some strategy based on local featural details. The tolerance of spatial distortions in human face recognition suggests that the configural information used as a basis for face recognition is unlikely to involve information about the absolute position of facial features relative to each other, at least not in any simple way.

Attention, spatial representation, and visual neglect: simulating emergent attention and spatial memory in the selective attention for identification model (SAIM)

The selective attention for identification model (SAIM) is presented. This uses a spatial window to select visual information for recognition, binding parts to objects and generating translation-invariant recognition. The model provides a qualitative account of both normal and disordered attention. Simulations of normal attention demonstrate 2-object costs and effects of object familiarity on selection, global precedence, spatial cueing, and inhibition of return. When lesioned, SAIM demonstrated either view- or object-centered neglect or spatial extinction, depending on the type and extent of lesion. The model provides a framework to unify (a) object- and space-based theories of normal selection, (b) dissociations within the syndrome of unilateral neglect, and (c) attentional and representational accounts of neglect.

Differential effects of viewpoint on object-driven activation in dorsal and ventral streams

Using fMRI, we showed that an area in the ventral temporo-occipital cortex (area vTO), which is part of the human homolog of the ventral stream of visual processing, exhibited priming for both identical and depth-rotated images of objects. This pattern of activation in area vTO corresponded to performance in a behavioral matching task. An area in the caudal part of the intraparietal sulcus (area cIPS) also showed priming, but only with identical images of objects. This dorsal-stream area treated rotated images as new objects. The difference in the pattern of priming-related activation in the two areas may reflect the respective roles of the ventral and dorsal streams in object recognition and object-directed action.

Relating priming and repetition suppression

We present a prototype of a recently proposed two stage model of the entorhinal-hippocampal loop. Our aim is to form a general computational model of the sensory neocortex. The model--grounded on pure information theoretic principles--accounts for the most characteristic features of long-term memory (LTM), performs bottom-up novelty detection, and supports noise filtering. Noise filtering can also serve to correct the temporal ordering of information processing. Surprisingly, as we examine the temporal characteristics of the model, the emergent dynamics can be interpreted as perceptual priming, a fundamental type of implicit memory. In the model's framework, computational results support the hypothesis of a strong correlation between perceptual priming and repetition suppression and this correlation is a direct consequence of the temporal ordering in forming the LTM. We also argue that our prototype offers a relatively simple and coherent explanation of priming and its relation to a general model of information processing by the brain.

Is there a linear or a nonlinear relationship between rotation and configural processing of faces?

Research suggests that inverted faces are harder to recognise than upright faces because of a disruption in processing their configural properties. Reasons for this difficulty were explored by investigating people's ability to identify faces at intermediate angles of rotation. Participants were asked to discriminate blurred famous and unfamiliar faces presented at nine angles. Blurred faces were used to minimise featural processing strategies, and to assess the effects of rotation that are specific to configural processing. The results indicate a linear relationship between angle of rotation and recognition accuracy. It appears that configural processing becomes gradually more disrupted the further a face is oriented away from the upright. The implications of these findings for competing explanations of the face-inversion effect are discussed.

Frames of reference in unilateral neglect and visual perception: a computational perspective

Neurological patients with unilateral neglect fail to orient and respond to stimuli on one side, typically the left. A key research issue is whether neglect is exhibited with respect to the left side of the viewer or of objects. When deficits in attentional allocation depend not merely on an object's location with respect to the viewer but on the object's intrinsic extent, shape, or movement, researchers have inferred that attention must be operating in an object-based frame of reference. Simulations of a view-based connectionist model of spatial attention prove that this inference is not logically necessary: Object-based attentional effects can be obtained without object-based frames. The model thus explains away troublesome phenomena for view-based theories of object recognition.

Effect of photographic negation on matching the expressions and identities of faces

In four experiments, participants made speeded same-different responses to pairs of face photographs showing the same woman or different women with the same expression or different expressions. Compared with responses to positive pairs, negative pairs were matched more slowly on identity than on expression. A secondary finding showed that face expressions (same, different) influenced identity responses, and identities influenced expression responses, equally for positive and negative pairs. The independence of this irrelevant-dimension effect from the contrast effect supports the conclusion required by the main finding that negation slows perceptual encoding of surface-based information used for identification more than it does encoding of edge-based information used for expression recognition.

Human and automatic face recognition: a comparison across image formats

Human subjects perform poorly at matching different images of unfamiliar faces. When images are taken by different capture devices (cameras), matching is difficult for human perceivers and also for automatic systems. We test an automatic face recognition system based on principal components analysis (PCA) and compare its performance with that of human subjects tested on the same set of images. A number of variants of the PCA system are compared, using different matching metrics and different numbers of components. PCA performance critically depends on the choice of distance metric, with a Mahalanobis metric consistently outperforming a Euclidean metric. Under optimal conditions, the automatic PCA system exceeds human performance on the same images. We hypothesise that unfamiliar face recognition may be mediated by processes corresponding to rather simple functions of the inputs.

Prosopagnosia as a deficit in encoding curved surface

RP is a case of "developmental" prosopagnosia who, according to brain-imaging segmentation data, shows reduction in volume of a limited set of structures of the right hemisphere. RP is as accurate as control subjects in tasks requiring the perception of nonface objects (e.g., matching subordinate labels to exemplars, naming two-tone images), with the exception of one perceptual task: The matching of different perspectives of amoebae-like stimuli (i.e., volumes made of a single smooth surface). In terms of speed ("efficiency") of responses, RP's performance falls clearly outside the normal limits also in other tasks that include "natural" but nonface stimuli (i.e., animals, artia facts). Specifically, RP is slow in perceptual judgments made at very low (subordinate) levels of semantic categorization and for objects and artifacts whose geometry present much curved features and surface information. We conclude from these analyses that prosopagnosia can be the result of a deficit in the representation of basic geometric volumes made of curved surface. In turn, this points to the importance (necessity) for the normal visual system of such curved and volumetric information in the identification of human faces.

Beyond localisation: a dynamical dual route account of face recognition

After decades of research the notion that faces are special is still at the heart of heated debates. New techniques like brain imaging have advanced some of the arguments but empirical data from brain-damaged patients like paradoxical recognition effects have required more complex explanations aside from localisation of the face area in normal adults. In this paper we focus on configural face processes and discuss configural processes in prosopagnosics in the light of findings obtained in brain-imaging studies. In order to account for data like paradoxical face recognition effects we propose a dual route model of face recognition. The model is based on the distinction between two separate aspects of face recognition, detection and identification, considered as dynamical and interrelated. In this perspective the face detection system appears as the stronger candidate for face-specific processes. The face identification system on the other hand is part of the object recognition system but derives its specificity in part from interaction with the face-specific detection system. The fact that face detection appears intact in some patients provides us with a possible explanation for the interference of configural processes on feature-based identification.

Paradoxical configuration effects for faces and objects in prosopagnosia

Selective impairment in recognition of faces (prosopagnosia) has been advanced as an argument for a brain module dedicated to face processing and focusing on the specific configural properties of faces. Loss of the inversion effect supposedly strengthened the argument ([10]: de Gelder B, Bachoud-Levi AC, Degos JD. Inversion superiority in visual agnosia may be common to a variety of orientation polarised objects besides faces. Vision Research, 1998;38:2855-61; [20]: Farah MJ, Wilson K, Drain H, Tanaka J. The inverted face inversion effect in prosopagnosia: Evidence for mandatory, face-specific perceptual mechanisms. Vision Research 1995b;35:2089-93). The present study of prosopagnosic patient LH reports that he has lost the normal pattern of superior performance with upright faces and objects and shows instead paradoxical inversion effect for faces but also for objects. Experiment 2 investigated whether LH's use of features based route for processing upright objects would be hindered by the whole-based encoding when processing upright objects. The data show the same context effect for objects as was found for faces. Therefore the inversion effect does not present decisive evidence for the existence of a face module. Moreover, the importance of configuration-based recognition known to be crucial for face processing, must also be taken seriously for object recognition.

Perceptual priming is not a necessary consequence of semantic classification of pictures

Four experiments investigated how repetition priming of object recognition is affected by the task performed in the prime and test phases. In Experiment 1 object recognition was tested using both vocal naming and two different semantic decision tasks (whether or not objects were manufactured, and whether or not they would be found inside the house). Some aspects of the data were inconsistent with contemporary models of object recognition. Specifically, object priming was eliminated with some combinations of prime and test tasks, and there was no evidence of perceptual (as opposed to conceptual or response) priming in either semantic classification task, even though perceptual identification of the objects is required for at least one of these tasks. Experiment 2 showed that even when perceptual demands were increased by brief presentation, the inside task showed no perceptual priming. Experiment 3 showed that the inside task did not appear to be based on conceptual priming either, as it was not primed significantly when the prime decisions were made to object labels. Experiment 4 showed that visual sensitivity could be restored to the inside task following practice on the task, supporting the suggestion that a critical factor is whether the semantic category is preformed or must be computed. The results show that the visual representational processes revealed by object priming depend crucially on the task chosen.

Organization of face and object recognition in modular neural network models

There is strong evidence that face processing in the brain is localized. The double dissociation between prosopagnosia, a face recognition deficit occurring after brain damage, and visual object agnosia, difficulty recognizing other kinds of complex objects, indicates that face and non-face object recognition may be served by partially independent neural mechanisms. In this paper, we use computational models to show how the face processing specialization apparently underlying prosopagnosia and visual object agnosia could be attributed to (1) a relatively simple competitive selection mechanism that, during development, devotes neural resources to the tasks they are best at performing, (2) the developing infant's need to perform subordinate classification (identification) of faces early on, and (3) the infant's low visual acuity at birth. Inspired by de Schonen, Mancini and Liegeois' arguments (1998) [de Schonen, S., Mancini, J., Liegeois, F. (1998). About functional cortical specialization: the development of face recognition. In: F. Simon & G. Butterworth, The development of sensory, motor, and cognitive capacities in early infancy (pp. 103-116). Hove, UK: Psychology Press] that factors like these could bias the visual system to develop a processing subsystem particularly useful for face recognition, and Jacobs and Kosslyn's experiments (1994) [Jacobs, R. A., & Kosslyn, S. M. (1994). Encoding shape and spatial relations-the role of receptive field size in coordination complementary representations. Cognitive Science, 18(3), 361-368] in the mixtures of experts (ME) modeling paradigm, we provide a preliminary computational demonstration of how this theory accounts for the double dissociation between face and object processing. We present two feed-forward computational models of visual processing. In both models, the selection mechanism is a gating network that mediates a competition between modules attempting to classify input stimuli. In Model I, when the modules are simple unbiased classifiers, the competition is sufficient to achieve enough of a specialization that damaging one module impairs the model's face recognition more than its object recognition, and damaging the other module impairs the model's object recognition more than its face recognition. However, the model is not completely satisfactory because it requires a search of parameter space. With Model II, we explore biases that lead to more consistent specialization. We bias the modules by providing one with low spatial frequency information and the other with high spatial frequency information. In this case, when the model's task is subordinate classification of faces and superordinate classification of objects, the low spatial frequency network shows an even stronger specialization for faces. No other combination of tasks and inputs shows this strong specialization. We take these results as support for the idea that something resembling a face processing "module" could arise as a natural consequence of the infant's developmental environment without being innately specified.

Contrast polarity and face recognition in the human fusiform gyrus

Functional imaging has revealed face-responsive visual areas in the human fusiform gyrus, but their role in recognizing familiar individuals remains controversial. Face recognition is particularly impaired by reversing contrast polarity of the image, even though this preserves all edges and spatial frequencies. Here, combined influences of familiarity and priming on face processing were examined as contrast polarity was manipulated. Our fMRI results show that bilateral posterior areas in fusiform gyrus responded more strongly for faces with positive than with negative contrast polarity. An anterior, right-lateralized fusiform region is activated when a given face stimulus becomes recognizable as a well-known individual.