Face recognition is an essential social tool for humans. It allows us to identify members of our species, and in doing so allows us to recognise those who we need to socialise with to aid our survival (our mother during our childhood, for instance). Not only do faces provide identity, they also provide information about someone's emotional state through their expression. Expression can also be a form of communication (e.g., exchanging a smile with someone when seeing each other across a busy street). By studying the Bruce and Young model of face recognition, we can gain an insight into the possible cognitive systems which serve the recognition of faces.
What is the Bruce and Young model of face recognition?
Bruce and Young’s model of face recognition is a classic “box and arrow” model. As with all cognitive models, the intention is to model functions and processes without trying to identify the actual mechanisms or brain locations involved. This imposes a limit on what such a model can tell us. These cognitive models do not tell us about which areas of the brain are involved in relevant processing, but the location of processing does not make any difference to the model of functioning. The mechanism itself is not relevant to the functions either. Bruce and Young’s model is based on observations of both patients and normals. They postulated from such evidence that face recognition is a modular series of processes, or in other words, is done in stages.
Structural encoding, view-centred descriptions, and expression-independant descriptions
The first stage of their model is called the “structural encoding” stage. This stage represents the encoding of visual information from a face into information that can be used by the other stages of the face recognition system. There are two separate processes here – “view-centred descriptions” and “expression-independent descriptions”. The expression-independent descriptions takes its input from the view-centred descriptions processes. These stages allow for the features of a face to be identified when viewed from any angle. This is a good part of the model due to the immense impracticability of a system which matches visual input to templates of what objects would look like from every possible angle. Instead, the Bruce and Young model has what could be described as a three-dimensional representation of known faces contained within it. This is derived from identifying features.
Parallel processing: a separation of funtion
There is a series of parallel processes after the structural encoding stage. The “expression analysis” stage takes its input from the view-centred descriptions processes. This stage, as one would imagine, analyses facial expressions. Evidence for there being a separate expression analysis set of processes comes from patients with specific brain damage that means they recognise faces, but they cannot interpret expressions. They can see the movements of the facial features, but this movement has no meaning to them. Another stage also receiving input from the view-centred descriptions processes is the “facial speech analysis” stage of processing. Evidence for this being a separate set of processes comes from patients who cannot lip-read, yet can perform all other facial recognition tasks without apparent deficit.
There is one further branch of parallel processing in the Bruce and Young model – “directed visual processing”. This is concerned with some more general facial processing, such as distinguishing between different faces. This set of processes takes its input from both the stages of structural encoding. All of these four streams of parallel face processing feed into the general cognitive system. Some of the streams get input back from the general cognitive system as well.
Recognition and identification stages
Bruce and Young’s model identifies a stage of processing involving “face recognition units”, or FRUs. These are individual nodes, each associated with a familiar face. Activation from detected facial features is fed into the FRU system, within which there is interactive activation and inhibition. The node which reaches a threshold activation level is the one that corresponds the face being observed, and it is then recognised. The FRUs interact with a further stage of processing involving “person identity nodes”, or PINs. The PINs receive input from the FRUs, thus when a face is recognised, the appropriate PIN provides information about that person. There is two-way interaction between the FRUs and the PINs as many experiments have shown that tasks which activate the PIN for a person creates some residual activation in the associated FRU, thus a subject’s reaction time for recognising the face in question is faster. After the PIN processes, there is a name generation process. It has been found that some people with brain damage can recognise faces insofar as they can give information about the person (from the PIN), but they just cannot recall the name. This is a form of prosopagnosia.
Does the Bruce and Young model stand up to its criticism?
One general criticism of the Bruce and Young model is to do with the observations that inspired it. A patient’s inability to do something may not result from damage destroying or cutting off particular dedicated area, but that damage may have occurred so that the only "easier processes" can be performed and the deficit in question happens to be more difficult. Evidence to support the modularity of the Bruce and Young model would need to come from sets of patients with “double dissociations”. Double dissociations are situations where there are two cognitive abilities A and B where there is a patient that can do A but not B, and another patient who can do B, but not A. This points to these two cognitive processes being separate. There are some parts of the Bruce and Young model that do survive this criticism, such as the facial expression analysis, as patients can sometimes recognise faces, but expressions mean nothing to them, and there are also some patients who cannot recognise faces, yet can recognise facial expressions. An experiment to show this involved such a patient being presented with three faces, then the patient was asked to pick out the faces he had just seen from a collection of six faces. The patient failed to do so. The experiment then went on to ask the same thing, except the patient was asked to pick out the same facial expressions. They were successful. This example of double dissociation shows that the FRU and facial expression analysis parts of the model are distinct.
An effect called “covert recognition” provides evidence that the FRUs and PINs are separate to the name generation process. It has been found that prosopagnosics are able to identify the occupation of a famous face, but cannot recall the person’s name. This could imply that there is some functioning in the FRUs and PINs, but there is difficulty with name generation.
Development of the Bruce and Young model
Burton et al. (1990) developed the FRU, PIN, and cognitive system parts of the Bruce and Young model, because as it stood, it did not quite capture normal function. Burton et al.’s model is an interactive activation model. It has excitatory and inhibitory connections between nodes in each part. There are nodes in the FRUs which correspond directly to face input and face recognition, PIN nodes correspond to information about people, whereas the relative areas of then cognitive system, consist of semantic identification nodes, or SIUs, which give meaning to a person’s identity. This interactive activation model explains the phenomena of semantic priming (where the input of a certain person’s face, leads to some activation of semantically related FRUs), and identity priming (where the repeated presentation of the same face strengthens the FRU, as in Hebbian learning).
The failure of the Bruce and Young model
The biggest failing of the Bruce and Young model is that it does not tell us all that much. It tells us just what sorts of processes happen in face recognition, but that is it. This does not seem that useful or even particularly incisive. The model is no more than a surface description of what happens when we recognise a face. It does fit in with observations, so we can reasonably believe that these are the processes that occur, but with no attempt to describe how these processes work, the model’s only valuable contribution is to be a basis for further study.
Is there a neurological basis for face recognition?
The recognition of faces is not really such a special thing when considered in terms of neurology. In fact, neuronally, it is hardly different to recognising any other object. Object recognition happens within the temporal cortices of the brain, as part of the ventral stream of visual processing. Objects are recognised by comparing the various responses of a population of neurons in the latter parts of the ventral stream, to stored object and mnemonic representations. For complex objects such as faces, this is done in the perirhinal cortex, which recieves input form the visual system and memory system, thereby associating a visual stimulus with a stored identity of what that stimulus is. Prosopagnosics have been shown (through autopsy and more recently neuroimaging techniques) to have some damage in the temporal lobe - suggesting that this is where face recognition occurs.
Based on the lectures of Prof. Peter McLeod, Department of Experimental Psychology, University of Oxford
M.J. Buckley and M. Gaffan (2000), The hippocampus, perirhinal cortex and memory in the monkey, In; J.J. Bolhurs (Ed.) "Brain, Perception, Memory: Advances in Cognitive Neuroscience", Oxford University Press