There are two main explanations for face recognition. Firstly, there is the Feature-analysis theory, a bottom-up theory. It suggests that analysing individual features is the most important factor in face recognition. The visual cues from a face we look at are the most important information for recognition, and therefore we need to focus on the detail of the face, looking mainly at the separate features. Visual cues include the way the light and shade appear on a face and also the texture of hair and skin. All of the visual cues combine to let us observe the broader features of the face like the shape of the eyes or mouth.
Ellis et al. (1979) showed that we tend to use internal features when recalling faces of familiar people, supporting the feature analysis theory of face recognition when recalling a face. This studies’ findings can be refuted as Shepherd et al. (1981) found the opposite to be true. He investigated the importance of different features, with participants being shown unfamiliar faces for a brief period of time. They were then asked to describe the faces from memory, with the features most often used being hair, eyes, nose, mouth, eyebrows, chin and forehead, in that order, showing internal features were used to recall unfamiliar faces.
Secondly, there is the Holistic-form theory. This is an alternative to the feature analysis approach. Although visual cues and facial features are important in describing faces and do have a role in face recognition, relying just on bottom-up processing for this very complex process is unlikely. Bruce and Young (1986) proposed a top-down approach, suggesting that face recognition requires the use of stored semantic and emotional information, and is more complex than simply putting together different features. According to the holistic approach, a face is recognised as a whole, looking at the relationship between features, feelings caused by the face and semantic information about the person. Young and Hay (1986) showed the importance of layout in the processing of faces.
Pictures of famous people’s faces were cut in half horizontally, and participants’ recognition of the people in the two separate halves was tested. The halves were then put together in non-matching pairs and participants were asked to name the person in each half. Again, participants’ recognition was tested. It was discovered that recognition time from the halves was much longer when two were put together. Recognition was difficult because the two halves combined made a completely new holistic face, making it harder to recognise the separate halves. This shows that visual cues and individual features are not the only factor in face recognition, as the overall layout of the face is equally, if not more important.
Empirical evidence suggests that face recognition is a very difficult activity, concerning both cognitive and emotional processes. Cases show that face recognition does not simply rely on features, because people suffering from Prosopagnosia and Capgras can name and describe individual features of familiar faces. This as well as other evidence points to the holistic form theory of face recognition rather than the feature-analysis theory.
Bruce and Young (1986) proposed a model of face recognition based on holistic form theory, shown below; 1. A familiar face is seen – A friend, relative, famous person, etc. 2. The face is structurally encoded – A mental description or representation of the face is produced from the stimulus. 3. The Face Recognition Unit (FRU) is activated – Each face known to the viewer has an FRU containing structural information about the face. 4. The Person Identity Node (PIN) is then activated – Information about the person, e.g. occupation, usual context, interests, whether liked or disliked. 5. Name Generation then occurs – An individual’s name is stored separately from other information about them and is accessed last.
According to Bruce and Young, a face is encoded structurally; meaning that visual information and the look of the face is processed. If this information matches an existing FRU, then the FRU is activated, which contains physical information about the person. Activation of the FRU then activates the PIN, giving semantic information about the person such as occupation, interests, where the person is normally seen and also whether or not we like the person. The final stage is then name generation. Names are stored separately to the FRUs and PINs, but can only be accessed through the PIN. This explains situations where lots of information about a person we meet is known, but their name can’t be remembered.
This model is good because it is backed up by empirical evidence. Young, Hay and Ellis (1985) tested the holistic model, making participants fill in a daily diary of problems in face recognition. They concluded that their findings supported the sequence proposed by the holistic model where names can only be accessed if semantic information has been accessed first. This experiment is excellent as it is valid, being a field study and also following every day activities (recognising faces).
A study linked to my chosen area of research is Sadr et al. (2003). Participants were asked to identify fifty celebrity face images in three conditions: unaltered image, image lacking eyes, image lacking eyebrows. Performance on images lacking eyebrows was found to be significantly worse than with both the unaltered images and also those lacking eyes. These results suggest that a particular feature – eyebrows – may contribute in an important way to face recognition. This study has been designed well, with a large group of faces, and a control condition. There is however, the problem that some people may recognise some celebrities faster than others as different people find different celebrities more or less famous.
There has been a large range of research into face recognition, but importance of features in face recognition is one area that has fewer studies into it than for example, the processes involved. Where Sadr et al. looked at features in the upper part of the face, the eyes and eyebrows; it would be interesting to test the difference in performance of recognition with different features covered and in a different population.