As one of the most successful application of image analysis and understanding, face recognition has received significant attention throughout the past several years. It is of particular interest in a wide variety of applications. Law enforcement applications seem to be the most important because verifying a personal identity (such as the face) can sometimes be a matter of life or death (Eysenck and Keane, 2010). In the context of social interactions, it is important to recognize other people’s facial expressions to judge them and know our impact on them. Faces also provide a wealth of information that facilitates social communication (Haxby, Hoffman and Gobbini,2000). Understanding human face recognition is also important in developing successful face recognition systems (e.g. surveillance cameras, face data bases). Though we currently are unable yet to endow machines with the capability of the human visual system, it is a good reference point from which to start (Eysenck et al, 2010).
Many biological, physiological and psychological researchers have been trying to understand face recognition. Researchers have been in constant debate about whether face recognition was a special mechanism that is different from the recognition of other objects (the domain specificity hypothesis) or whether it was a general one that responds to face stimuli the same way it does to on non-face stimuli (the expertise hypothesis). The domain-specificity advocate claim that faces, contrary to non-face objects, are processed holistically and that specialized cognitive and neural functioning in the cortical region are activated when faces are processed (Kanwisher and Yovel, 2006). While the domain-general supporters argue that face recognition only seems to be special because people have greater expertise in discriminating faces. Both theories rely on evidence from behavioral, neuropsychological and biological studies to verify their assumptions.
Evidence for the domain specificity hypothesis
Substantial behavioral studies have shown differences in the ways faces and non-face objects are processed, suggesting that special mechanisms are involved in processing faces (Duchaine and Nakayama, 2006). Humans process faces in a holistic inversion-sensitive manner; it involves strong integration across the whole object rather than parts of it (Eysenck et al, 2010). Support for this idea comes from studies on the part-whole and composite effects as well as the inversion effect.
In the part-whole effect, individual face parts (such as the mouth) are recognized better when the entire face is presented than when single parts are presented in isolation, whereas recognition of parts of other objects (houses for example) was the same whether presented as a whole or as separate parts (Eysenck et al, 2010). Yavel, Paller and Levy (2005) stated that that “the probability of correctly identifying a whole face is greater than the sum of the probabilities of matching each of its component face halves” (p. 2111). Humans are more accurate in discriminating faces than discriminating each parts of the same face independently
In the face-inversion effect, faces become difficult to identify when turned upside down, whereas this inversion does not affect the discrimination of other non-face objects. Eysenck et al (2010) explains that the reason behind the inversion effect is that normally oriented faces are perceived as a unique pattern and processed holistically while inverted faces and objects are perceived as components and processed on a parts-based manner (Moscovitch Winocur and Behrmann 1997). Scapmello and Yarmey (1970) showed that the inversion effect was higher for human face recognition than for recognition of inverted buildings. Similarly, Yin (1970) found that recognition of normal oriented photographs was higher than inverted ones by more than 25 percent, while it was reduced to two to ten percent in other objects’ photographs.
In the composite affect, identifying half a face is impaired when aligned with another inconsistent half. Not only that, reaction time increases when identifying two half faces of different people that are aligned than if the two half-faces are misaligned (Eysenck et al, 2010). However, a composite effect was absent when identifying objects of inverted faces (Moscovitch et al., 1997). Accordingly the effects mentioned above have been considered as evidence of the unique diagnostic nature of face recognition.
Evidence from brain damage studies has revealed a great deal about the structure of face perception. By studying the behavioral performance of impaired and normal brains, scientist found that face perception functions separately from other non-face perceptions and can be damaged independently as well (Bruce and Humphreys, 1994). This is the case in prosopagnosia, a mental illness that impairs the patient’s ability to recognize previously familiar faces while maintaining the ability to recognize objects (Eysenck et al, 2010). Prosopagnosic patients apparently have impairments that are not the result of deficits in a general visual-recognition mechanism but rather a specific one. Brain prosopagnosia can result from a sudden or developmental brain damage; however, patients in both cases demonstrate relatively poor face identification but excellent identification of objects of expertise which indicate a disassociation between the two. For example, Patient R.M. (a car expert) suffered from prosopagnosia after an aneurysm where he could not recognize familiar faces but still could remarkably recognize cars models and years (Sergent, and Signore, 1992).
Even more striking evidence comes from patients with the opposite syndrome, where patients show deficiency in recognizing objects while their face recognition is not affected, such as the case of MX, a farmer who could not recognize his cows but could recognize faces (Assal, Favre, and Anderes, 1984). Another convincing case is patient CK, a collector of toy soldiers who lost his ability to discriminate his toy soldiers but retained his face recognition ability (Moscovitch. et al, 1997). In addition C.K.’s face recognition was more disrupted by the inversion and part-whole effects that than those of normal subjects, which confirms the idea that face processing is a holistic inversion-sensitive mechanisms (Kanwisher, 2000).
The clinical studies detailed above suggest that there is a specific area in the human visual system that is selective for faces. These studies provide an important complement to neuropsychological and neuroimaging functional studies which latest found considerable a particular area in the temporal cortex that gets more active when processing faces than when processing other objects. It is the fusiform face area (FFA) (Bruce et al, 1994). Several Functional magnetic resonance imaging (fMRI) studies have found that the FFA responded at least twice as strongly for faces than for non-face stimuli, such as animals without heads (Kanwisher, Stanley, and Harris, 1999). Kanwisher (2000) found that the FFA’s response to viewing human faces was four times higher than when viewing human hands. Few studies have found other face-selective regions (such as the superior temporal sulcus and the occipital face area (OFA); however, activation of these areas was not strongly detectable when processing faces. They were also found to respond when performing other non-face recognition (Kanwisher, et al, 2006).
Interestingly, recent studies found that some non-face stimuli has activated different regions in the brain, suggesting a specific system exists for recognizing different objects and that it is subdivided into smaller systems responsive to a specific category of objects (Tove?e, 1998). Aguirre, Zarahn, and D’Esposito (1998) found that a specific region of the lingual gyrus was responsible for recognizing buildings. Grill-Spector, Knouf and Kanwisher (2004) found that many non-face stimuli (including cars viewed by car experts) have activated the ventral Ocipitotemporal cortex.
Neuroimaging analysis has been of great impact in determining the classification of faces and objects in the FFA area and other areas in the brain. However, fMRI techniques by themselves are not adequate enough to identify the actual activated location in the brain (McKone, Kanwisher and Duchaine, 2006). fMRI capacity to isolate the area where the actual computation takes place is quite limited.
Evidence for the domain general hypothesis
The above findings have led many scientists to conclude that the human brain contains specific mechanisms dedicated for handling faces. Other researchers, however, disagree on the uniqueness of face processing. They claim that the same face processing mechanisms are involved in handling any type of visual stimuli for which we have acquired significant visual expertise (the Expertise Hypothesis) (McKone et al, 2006). And since humans have much greater experience in memorizing faces compared to recognition of other exemplars such as animals or cars, therefore face processing falsely appears to be “special” (Kanwisher, 2000).
The expertise hypothesis advocates claim the FFA may not be explicit for faces, but rather for operations that we generally execute when perceiving faces. They predict that the FFA should be strongly activated when presenting objects of expertise and/or familiar object than when novel objects are presented (McKone et al, 2006). Therefore, they have used brain imaging techniques to investigate the neural mechanisms involved in processing expertise stimulus (Atkinson and Adolphs, 2011).
Gauthier, Tarr, Anderson, Skudlarski and Gore (1999) conducted a series of behavioral studies where they trained subjects to become experts with a novel class of objects called ‘greebles’-photorealistically rendered 3D objects that share a typical pattern. Subjects were scanned using fMRI before, during and after the training where they reached the required level of expertise. Results showed higher fMRI response in the right hemisphere face areas in greebles experts than in novices during passive viewing of greebles, suggesting that acquiring expertise leads to the same neural response of processing faces. Gauthier et al. (1999) also found higher activation in the right FFA when identifying upright greebles as compared to identifying inverted ones. They explain that while becoming skilled with greebles, subjects shift from regular feature-based processing to a more configural one, a property that discriminates face recognition from object recognition.
Gauthier, Skudlarski, Gore and Anderson (2000) also measured fMRI responses in cars and birds professionals when identifying related materials. Results showed that both the right FFA and the right OFA were more highly activated when matching stimuli within the participants’ proficiency than outside their proficiency. Results also indicated that level of response in the right FFA significantly correlated with the level of expertise when participants performed location-matching tasks.
Some opponents of the expertise theory claim that the expertise effect found is due to some resemblance in the stimuli; birds have faces and three-quarter frontal views of some cars can be configured as faces (McKone et al, 2006). Greebles have faces and human names (Kanwisher, 2000). The expertise effect also may be a result of a biased attentional modulation; experts may pay more attention to the material that falls under their interest (Wojciulik et al., 1998). And since the response in the FFA can be strongly modulated by different amount of visual attention (Kanwisher, 2000), it is possible that the effects do not reflect uniqueness of the FFA in processing objects of proficiency but rather an increase in attending to these stimuli (McKone et al, 2006). Furthermore, the trails in the experiment of Gauthier et al (1999) were presented to participants as blocks of similar objects (e.g. cars) which allowed them to anticipate the images that will be shown in the same block, further affecting their attention (Wojciulik et al., 1998). To exclude these possibilities, Xu (2005) conducted an experiment using different car images that do not look like faces and presenting them in an event-related design that minimizes attentional modulation. Results indicated an expertise response in the right FFA suggesting it process non-face stimuli of expertise.
Few other studies have found evidence against the face-specifity hypothesis. For example Damasio, Damasio, and VanHoesen (1982) found that prosopagnosia was often accompanied by failure to identify similar non-face objects. However, it is difficult to identify is whether this deficit is caused by damage in the face area or nearby cortical areas that are frequently damaged together (Kanwisher, 2000). On the inversion effect, Diamond and Carey (1986) found that dog experts showed the same inversion disruption for inverted photographs of dogs as they did for face inversion. However, when using different stimulus (landscapes), a lesser inversion effect was revealed, probably because the spatial relations between different feature of the picture have made it easier to identify the picture even when inverted. Nevertheless there has been only one replication of Diamond and Carey (1986) experiments which failed to validate the original result (Robbins 2005). Furthermore, it is also not clear whether the inversion effect found for non-face stimuli is a result of a unitary mechanism that works for both faces and non-faces (e.g. dogs) or of two different mechanisms that work in similar ways and cause the inversion effect in dogs and faces (Kanwisher, 2000).
Early studies of children’s face perception claimed that children can only develop a mature holistic face processing when reaching the age of ten, explaining that children by that age would be experienced enough with faces, which supports the expertise hypothesis (Carry, 1981). However, the latest research has found that children as young as four have shown evidence of the inversion effect, the composite effect, and the part-whole effect (Pellicano and Rhodes, 2003). Cohen and Cashon (2001) have shown that even seven-month-old babies show signs of holistic face processing. Thus, developmental studies do not offer support for the expertise view
In this essay, we have argued that the majority of evidence from cognitive, neural and behavioral means demonstrate that face perception is distinct from the perception of other objects. We have also presented opposing evidence that denies this uniqueness, presuming a common processing system that includes a specialized sub-region for faces as well as non-face objects, including objects of expertise. All the recent contradictory research on face processing has added additional uncertainty to the domain specific-domain general debate with one piece of evidence refuting the other. As indicated before, even though several studies have found that identifying non-face objects of expertise activates the FFA area, other studies have found that non-face stimuli of expertise was correlated with higher activation in regions outside the FFA, such as parahippocampal cortex (Gauthier et al, 1991).
Identifying human nature has hardly been straightforward in science, and when it comes to the mind and human perception, this inconsistency is amplified. “The face recognition system is highly specialized and capable of extremely fine discriminations” explains Tove?e, (1998, p1224). Humans remain better experts in identifying faces than any other exemplars, such as hands or body shapes. The classic remark is that thieves regularly cover their faces rather than any other body parts. (McKone et al, 2006). But whether face processing functioning is exclusive or is shared by the recognition of other object classes is the matter of evidence, which calls for more research to examine the roles of expertise in the stimulating face-specific regions of the cortex. Xu (2005) explains that examining how other objects classes that do not resemble faces and do not contain face parts are recognized and represented in the brain is the best approach to study the expertise hypothesis. We also need further insights into child and infant’s face recognition to better understand patterns of face recognition and whether it is an innate process or one that comes by expertise (McKone et al, 2006). Only then will we be able to confirm whether face recognition is truly “special.”