Face Recognition

Face Identification - Looking at a face and stating who it is

Face Recognition - Recognising the face as one seen before

Face Recall - From memory, verbally describing, or drawing, or forming a mental image of the face

Bottom up theory - visual cues (features) from the stimulus (face) will be sufficient enough to enable recognition

Sadr et al. - Eyebrows were found to be the most important feature

Shepard, Davies and Ellis - Investigated how features were used in the description of unfamiliar faces. Participants were shown unfamiliar face and then had to describe what they had been shown. In these descriptions, the features most often noted were hair, eyes, mouth, eyebrows, chin and forehead in that order. The researchers concluded that unfamiliar faces tend to be recalled using the main features. 

Ellis et al. - With unfamiliar faces we rely on external features e.g. hair, face shape. With familiar faces we tend to rely on internal features e.g. eyes, nose. External features are more noticeable but more likely to change, so internal features are therefore more reliable. 

Early studies of face recognition focused on whether features were processed independently. 

Bradshaw and Wallace - constructed pairs of faces out of Identikit, which differed by fixed numbers of features. They found that volunteers were quicker to identify that faces differed if the number of different facial features on the pairs of faces was greater. i.e. the more differences there were on the pairs of faces, the quicker the participants could respond. The researchers concluded that facial features are processed independently and in sequence.

However, Sergent pointed out that when faces differed in a number of features the overall configuration also differed, and this too could explain why identification was more difficult. Sergent carried out a study that showed that facial features were processed interactively and not independently.

Aim - To investigate if features are processed independently or configuratively.

Method - Participants were shown pairs of faces and the time taken to state whether the faces were the same or different was measured. The faces were in pairs and were of 4 types: the same; differed on one feature only; differed on two features; differed on three features. There were 2 different chins, 2 different eye colours and different internal spaces.

Results - The more features that differed between the faces, the quicker the 'different' response could be made. A difference in chins was the fastest to be detected, and when there was a difference in chin and at least one other difference then differences were detected even quicker, suggesting that features were not processed independently.

Conclusion - Facial features are processed configuratively not independently

Scrambled Faces - When the configuration of features is altered, it is more difficult to recognise a face. If only features were important, it wouldn't matter where they were located on the face, suggesting configuration is also important.

Inverted Faces - Inverted faces produce particular difficulty in terms of recognition. Sergent repeated her experiment with inverted faces and found no differences in the time taken for recognition; these results are consistent with a model of independent feature processing. 

Other studies have suggested that different processing occurs in upright and inverted faces. One explanation for this is that faces are normally recognised holistically; therefore, inverting faces makes recognition difficult because the relationship between the features cannot be detected. With inverted faces, the participants may have to resort to independently processing features, which takes longer.

Pairing different halves of faces - Young + Hay cut pictures of famous faces horizontally and stuck them together to produce a composite, and asked participants to name the top and bottom halves. This proved difficult as the composite seemed to produce a new holistic face.

Prosopragnosia - A disorder where patients are unable to recognise familiar faces and in severe cases, can't recognise their own face in the mirror. They can recognise individual features on faces and feel that they know the person (emotional response), but they have no knowledge of who the person is (no cognitive response). If only features were important, then they should be able to recognise people, suggesting that configuration plays a part too.

Capgras Syndrome - A delusional disorder where patients think that people they know have been replaced by 'doubles'. Such patients recognise the person (cognitive response) but they have no emotional response to the individual as they believe them to be a double. This disorder suggests that semantic and emotional information is important in terms of recognising somebody, as the cognitive response is not enough to enable recognition. 

These disorders, therefore, support the holistic model.

Although features are important, reliance wholly on bottom-up processing for such a complex activity is very unlikely. Bruce and Young proposed a top-down approach to face recognition in which they argued recognising a face is a highly complex process involving stored knowledge of semantic and emotional information (which is already stored in memory).

Top down theories say that visual cues from the environment are not sufficient enough to enable recognition, and that cognitive processing of some sort is also necessary for successful recognition. 

A face is recognised as a whole - features, configuration of the face, the relationship between features, feelings aroused by the face and semantic and contextual information about the person all contribute towards recognition. 

Also referred to as a template model - we have a stored template/pattern for each person we know and when presented with a face try to match this stimulus to our mental template. (Ellis et al.) Bruce referred to these as 'face recognition units'.

Bruce believed that there is a process we go through each time we are presented with a face - this process is sequential.

1. A familiar face is presented and structurally encoded - a representation/description of the face is produced

2. This activates Face Recognition Units (FRUs) - each face known to the viewer has an FRU containing structural information about that person

3. This activates Person Identity Nodes (PIN) - Information about the person e.g. normal context, occupation

4. This activates Name Generation - an individuals name is stored separately from other information about them and is accessed last.

Aim: To investigate the difficulties in face recognition experienced in everyday life and to find support for a holistic theory of face recognition.

Method: A field study was carried out with 22 volunteers who were asked to keep a diary for 8 weeks and record any cases where they failed or had difficulty in recognising someone they knew.

Results: 1,008 cases were recorded. There were no reports of naming an individual without knowing other information about that person. In 190 cases, participants recorded knowing information about someone but not being able to name them. There were 233 reports of participants reporting a feeling of familiarity without being able to access personal identity information or name.

Conclusion: The results support the sequential nature of face recognition proposed by the holistic theory.

Evaluation: This is a particularly valid study because it was carried out in the real world over a period of time.

• There is converging evidence from a wide range of sources - the diary study, clinical studies and lab research that the components of the holistic model are relevant to the recognition of familiar faces. 
• Flude et al. presented evidence from a patient who could name the occupations of 85% of the familiar faces presented to him, but could only name 15%, supporting the sequential nature of the model.
• The model has been criticised with respect to the explanation for unfamiliar-face recognition, which lacks the detailed evidence that has supported the familiar-face recognition aspect of the model
• Bruce + Young assume separate components/processes for familiar and unfamiliar faces. Malone et al gave evidence for this by describing a patient who could recognise famous faces but was highly impaired at recognising unfamiliar faces. Another patient could match unfamiliar faces perfectly well but had difficulty recognising photos of famous people, getting only 5/22 correct. This double dissociation supports the model. 

One of the main problems with the sequential holistic model has been the findings from some patients with prosopragnosia on face-recognition tasks. Although such patients show no conscious recognition of a familiar face, they do register covert recognition.

This has been measured by autonomic responses which have found a response to familiar faces at an unconscious level, where both the FRU and PIN appear to be activated. 

If conscious familiarity is not available, how can a deeper level of familiarity be available unconsciously?

Such findings have led to a modification of the original model, known as the interactive activation and competition model (IAC).