How do we recognise faces?

Citation

Wells, D. (2000), "How do we recognise faces?", Nutrition & Food Science, Vol. 30 No. 3. https://doi.org/10.1108/nfs.2000.01730caf.002

How do we recognise faces?

How do we recognise faces?

When one considers that an individual recognises hundreds or even thousands of faces that are familiar it is easy to understand that occasionally it is difficult to put a name to a familiar face. Consider the number of people known to one personally in private and professional life as well as the many that are presented to us in the media such as television, newspapers and films. That we can put names to a lot of faces is remarkable given that faces have the same characteristics of eyes, nose, mouth and chin. Yet these features can be arranged in such a wide variety of ways that individuals, with the exception of identical twins, can be distinguished from one another and named.

Face recognition appears to involve a special mechanism not shared by recognition of other objects. Recognition of familiar faces appears to be a holistic process in that we recognise the whole arrangement of features rather than considering each feature individually.

Tonaka and Farah (1993) conducted an experiment to find out whether faces are represented in memory as a complete face or as separate components. They tested subjects with complete faces and faces in which the features were rearranged or scrambled. Subjects were asked to learn proper names for the complete faces. They were then asked to recognise features from these faces when they were rearranged or replaced by a feature from a different face. Tonaka and Farah found that memory for face features was poorer when tested in isolation than when they were in a complete, upright face. This supports the claim that normally arranged faces are processed holistically and they are not analysed into separate features.

Putting a name to a familiar face has important social influence. It also has an effect in helping us to understand current issues which affect us all. The face is the most visible clue to a person's identity. Putting a name to a familiar face in the media leads to a better understanding of current affairs that helps us to appreciate the various factors that influence our daily lives. Various studies have suggested that there are stages in the process of identifying and putting a name to a familiar face.

Young et al. (1985) asked 22 subjects over a period of eight weeks to keep a record of their inability to put a name to a familiar face. There were 1,008 such reported incidents. But there was not one incident where a subject was able to name a familiar face without knowing some personal details about that person. This shows that names cannot be accessed without prior knowledge of some details of the individual's personal history. In contrast there were 190 incidents when the subject could remember some personal details without putting a name to the face.

Young et al. (1985) proposed that when we encounter a face it is first processed by face recognition units. These units contain stored information about the appearance of familiar faces. If the face matches one of these units it will be further processed by the personal identity nodes (PINs). The PINs contain information about the personal identity of the owner of the face, such as occupation or occasions when previously encountered. The PINs lead to the additional information stores from which the name of the individual is accessed. According to Young et al. (1985), in simple terms, the route to naming a familiar face is familiarity - personal identity - name. The face recognition units and the PINs have direct access to the cognitive system. Decisions about whether a name can be put to a face comes from a link between the various units in the system and the cognitive system.

Further evidence about the order of this process comes from subjects' responses when asked to say if the face is familiar or if the name belongs, for example, to a politician or a film star. Responses are quicker to the familiarity question than to the one about occupation. This suggests that the process has stopped at the face recognition unit. Johnston and Bruce (1990) found that subjects were more likely to say that two well-known faces were familiar (John Lennon and John Kennedy) and that they were both dead rather than knowing that they shared a first name of John. The dead or alive question was derived from the PINs but there the process stopped and no names could be derived from the additional information stores. If there is a block between PIN and additional information store, the person is recognised and some personal information about them accessed but their names will not become available. Failure to put a name to a face can therefore occur at any stage in the sequence.

Neuropsychological evidence also supports these findings. Patients suffering from prosopagnosia suffer from a deficiency in face processing but not memory. Prosopagnosic patients are unable to recognise familiar faces, perhaps even their own when seen in a mirror. But they can still recognise familiar people from their names and the sounds of their voices. Flude et al. (1989) studied one such patient who was able to describe the occupation of 85 per cent of very familiar people when shown their photographs but could only recall 15 per cent of their names.

Bruce and Young (1986) extended the Young et al. (1985) model of face recognition. The route by which familiar faces are named was extended to include a separate stage named structural encoding. The Bruce and Young (1986) model also includes an expression-independent description and facial speech analysis as well as a unit called directed visual processing. This latter unit allows certain processes to be addressed selectively before final identification. An example would be looking for a tall, bald-headed man in a crowd before identifying your tall, bald-headed husband. Expression analysis and facial speech analysis may also be used in naming familiar faces. Recognition of familiar faces then proceeds along the stages proposed by Young et al. (1985), namely face recognition units to personal identity nodes and finally to name generation.

However, Bruce and Young's (1986) model had some shortcomings. For example, it failed to specify some of the components and processes involved in recognition of faces. Bruce and Young (1986) admitted that the cognitive system in general serves to effect all those processes not included in the other components of the system, pointing out that face recognition units and PINs have a direct access to the cognitive system. The model also failed to account for the fact that an amnesic patient studied by de Haan et al. (1991) was able to match names and faces of 88 per cent of famous faces she was shown but was unable to recall any autobiographical information about them. The fact that her PINs were damaged should have prevented her matching faces and names.

Burton et al. (1990) and Burton and Bruce (1992) revised and developed the Bruce and Young (1986) model. In this revised model there are three pools of information:

1. 1.

   face recognition units;

2. 2.

   person identity nodes; and

3. 3.

   semantic information units.

Decisions about a person's familiarity is made at the PINs rather than the face recognition units. A face is recognised as familiar when the level of activity in the PIN reaches a certain threshold. The face recognition units are activated by the presentation of a familiar face. The name of the face and other information about the individual are derived from the semantic information units again when activation reaches a certain threshold level. There is no separate store for names alone. This model explains the findings of de Haan et al. (1991) with their amnesic patient because face and name information can be linked without access to autobiographic information.

A computer simulation with an interactive activation and competition connectionist architecture of this revised model has been produced. The model contains some of the units of the Bruce and Young (1986) model. The PINs are represented as identity specific nodes rather than information stores. The PINs receive information from the face recognition units as well as from name input units and semantic information units. The semantic information units specify biographical information such as nationality, occupation and leisure pursuits. The PINs, not the face recognition units, decide whether or not a person is familiar. The face recognition units signal the degree of resemblance between the face input and the stored representation of face appearance. Within each pool there are inhibitory actions which means that only one unit within each pool becomes active after the presentation of a face. The decision that a person is familiar is made when activation in the PINs reaches a certain threshold level of activation. The name input units then put a name to the presented face.

Occasionally, the name of a familiar person is on the "tip of a subject's tongue". Brennan et al. (1990) conducted an experiment to see when subjects are asked to name a famous person from a description, as in the game Trivial Pursuit, and the subjects were in the tip-of-the-tongue state, what would help them retrieve the person's name. They found that repeating the question did not relieve the name block very much. Nor did showing a photograph of the celebrity. But giving the initials of the person's name led to its retrieval in almost half the cases. Bruce and Young (1986) concluded that the subjects were at the PIN stage but the name eluded them. Seeing the initials gave a substantial proportion of them the clues to the missing name.

Even familiar faces are more difficult to name when their photographs are presented to subjects upside-down. It appears that upside-down faces are not recognised holistically but are recognised as a set of independent components. This means that the eyes, nose, mouth and chin are considered separately and not as an integral face as upright faces are recognised. Yin (1969) compared recognition of familiar faces with other familiar objects such as buildings. He concluded that, while faces are recognised more accurately than other subjects when they are upright, faces are recognised least accurately when they are upside-down. This can be tested by turning upside-down a photograph of familiar faces one can name, and seeing how difficult it is to recognise the individual faces.

Recognition of familiar faces is also more difficult when they are presented in photographic negation. Bruce and Langon (1994) compared the effect of negation with those of inverting face images. Subjects were presented with photographs of 28 famous faces and asked to name them. There was a 95 per cent correct response. When the photographs were inverted, correct response rate fell to 70 per cent. When photographic negatives were presented, only 55 per cent of the faces were correctly identified. When the negatives were inverted, response rate fell to 25 per cent.

Bruce and Langton (1994) concluded that negation has an even more detrimental effect on recognition of familiar faces than inversion. The reason for this dramatic effect is because images appear unfamiliar in negation. Light hair and skin appear dark and vice versa. Brunettes become blond and blonds become brunettes. Shading patterns are also wrong. What is shading in a positive may appear to be an unsightly lump in a negative.

Young et al. (1985) took photographs of well-known people, sliced them in half horizontally and rearranged the top half with the bottom half of a different but equally well known face. Subjects found it more difficult to recognise these composite pictures than they did to recognise the top halves on their own. Young et al.'s (1985) explanation for this finding was that the composite produces a new configuration which is more difficult to identify. When these "split" photographs were turned upside-down, subjects were more accurate at naming the halves of the faces than when the composites were upright. Again this suggests that upside-down photographs are judged by individual features rather than holistically.

It seems that putting a name to a familiar face depends on activity in certain regions of the right hemisphere of the brain. Hay and Young (1982) thought there may be special areas of the cortex dedicated to the processing of information about faces which are not concerned with the processing of other objects. This assumes that face processing is a unique process. New-born babies follow a schematic face with eye and head movement rather than follow a scrambled or upside-down face. New born babies may also have the ability to recognise and respond to particular facial expressions of their carers.

Johnson and Morton (1991) argue that there is an innate ability to pay attention to faces to direct neural mechanisms to learn about the special characteristics of faces. Neuropsychological evidence comes from prosopagnosic farmers who typically have damage to the right hemisphere of the brain. These patients have difficulty recognising human faces but can recognise individual sheep and cows. This suggests that human face recognition may be different from the recognition of other animals and objects.

Progress in the field of human face recognition has progressed rapidly in the last ten years. We now know more about the processes involved in face recognition compared with recognition of most other objects. It is a complex area of psychology but then one of the most important and complex of visual objects is the human face.

Dilys WellsEditor