Week 4 Lecture 4 - face recognition Flashcards
What is prosopagnosia?
inability to recognise faces
What are the demands of face recognition?
Generally same as for object recognition
- recognition in context
- object invariance
- specificity
But faces require higher specificity –> in most situations we need to recognise a specific individual face
- within-category discrimination
Ellis, Shepherd and Davies (1975) studied the effectiveness of the Photofit system.
What did they find?
- Results indicated that this was not a very good way of forming representations of the face
- People have difficulty reproducing likenesses of even familiar faces
- Photofit has the inherent belief that we process local features individually
How are faces thought to be represented?
configurally
Faces are thought be be represented configurally.
What can this mean?
- That the spatial relationships between features are as important as features themselves
- That face features interact with one another
- That faces are processed holistically
A study by Tanaka and Farah was based on the premise:
If parts of faces represented separately then memory for these parts presented in isolation should be as good as when presented within a face
What was the method for this study?
- Learned names to go with a face
- Later asked to pick out ‘Larry’ from two alternatives
- Alternatives were either features presented alone or the whole face presented in the learning context
A study by Tanaka and Farah was based on the premise:
If parts of faces represented separately then memory for these parts presented in isolation should be as good as when presented within a face
What did they find?
- Features learned in the context of a normal face were better tested in that context
- Features learned in the context of a scrambled faces were better tested in isolation
- Concluded that the representation of whole faces based, at least in part, on a holistic representation
A study was conducted by Young et al (1987) into upright and inverted faces.
What was the method when faces were presented on upright??
- half faces of celebrities paired to make a new face in either a composite (aligned) condition or non-composite (misaligned) condition
- ppts asked to name top half or bottom half
- RT measured
A study was conducted by Young et al (1987) into upright and inverted faces.
What were the results when faces were presented only upright?
- Significantly slower to name composite than non-composite stimuli
- Perception of a novel facial configuration interfered with the identification of the constituent parts
A study was conducted by Young et al (1987) into upright and inverted faces.
What were the results when faces were presented both upright and inverted?
- Interference from the configurational information only found for upright faces
- no significant difference in reaction times for composite and non-composite images when picture was presented inverted
What can the Thatcher illusion be explained by?
explained by lack of configural processing for inverted faces
What do face representations seem to preserve in terms of surface characterisitcs?
pigmentation
Bruce and Langton looked at the effects of negation on face recognition.
What did they find?
- turning face upside down effects recognition in one way
- making the image negative effects recognition in a second different way
- using both an inverted negative image combines effects and makes recognition even lower
What are 2 possible sources of information that are adversely affected by negation?
- Pigmentation = skin and hair colour &
variations in these - Pattern of shading and shadow which may help specify 3D structure of face
What did a study into recognition of 3D heads find?
When images lacked pigmentation effects of negation were much reduced
A study by Hanna et al (1992) looked at recognition of line drawings.
What did they find?
Line drawings are difficult to recognise unless information about pigmentation and / or shading preserved in some way
Are faces all equally easy to recognise?
no
Are distinctive or typical faces recognised more accurately and quickly?
distinctive
Valentine and Bruce shown pictures of famous and unfamiliar faces and asked to rate ‘how well it would stand out in a crowd
What did they find and conclude?
- Distinctive faces are easier to identify than typical faces
- Typical faces are more easily classified as
faces than distinctive faces - Valentine (1991) accounted for such effects using a model of ‘face space’ where distinctive faces are represented in more sparsely populated regions
What is Bruce and Young’s (1986) model of face recognition?
4 stages in recognition of familiar faces
1. form structural description of seen face –> structural encoding
2. match to stored representations of known faces –> Face recognition units (FRUs)
each FRU contains structural information about a face known to the viewer
3. access semantic information –> Person Identity Nodes (PINs)
4. recall person’s name –> name retrieval
Is Bruce and Young’s (1986) model of face recognition supported by evidence?
yes –> but underspecified
What is 2 problems identified with Bruce and Young’s (1986) model of face recognition?
- independence
- covert recognition
How is independence a problem for Bruce and Young’s (1986) model of face recognition?
- Schweinberger & Soukup (1998) found that RTs for identity judgments were independent of variations in expression and facial speech
- But RTs for expression and facial speech were influenced by variations in identity
- Asymmetrical dependencies between different components of face perception?
How is covert recognition a problem for Bruce and Young’s (1986) model of face recognition?
- Bauer (1984) - some prosopagnosic patients with no OVERT recognition of famous faces showed COVERT responses
Suggests either:
- there is “leakage” from the rest of the cognitive system in face recognition
- or there are two face processing routes, one conscious and one unconscious
How does prosopagnosic patient PH provide evidence for covert recognition?
- At chance in selecting famous face from pair (18 / 36 correct)
1.) same / different matching
- Better at deciding that two views of a famous face belonged to same person than two pictures of an unfamiliar face
2.) face name interference
- Showed interference from distracting faces when asked to classify names into semantic categories
3.) semantic priming
- Faster recognition of a familiar faces if the face or name of a related person has just been seen.
- Priming from famous face does not overtly recognise onto the name of a related person
4.) learning of true and untrue occupations and names
- PH Learns true names and occupations of famous faces significantly faster than untrue ones.
- True even for faces famous since accident
What model of face recognition can account for covert recognition?
Burton, Bruce and Johnston (1990) –
connectionist ‘interactive activation’ (IAC)
model
What is the IAC model?
- uses feature units / FRUs / NRUs /
PINs/ SIUs connected by bidirectional
excitatory links
semantic priming:
- caused by feedback from semantic units to PINs to FRUs which ‘warms up’ PINs and FRUs of related people
- e.g., William FRU - William PIN - semantic system “royal” - back to Kate PIN - Kate FRU
- familiarity decision taken at PINs. On the
basis of activation received from faces, names, voices etc.
What does the IAC model not do?
doesn’t stimulate name retrieval
