Face Recognition Flashcards
Face vs object recognition
purpose: to identify a specific individual vs to categorise, individuation and fine discriminations required
faces all have same parts, so recognition by components does not help
great amount of expertise with faces
dissociation after brain damage
Prosopagnosia = deficits in face recognition (with intact object recognition)
Visual object agnosia = deficits in object recognition (with intact face recognition)
–> objects recognised by integrating parts, faces recognised holistically
different brain regions:
- fusiform face area (FFA), occipital face area (OFA), superior temporal sulcus (STS) for face
- lateral and inferior occipital lobe, inferior temporal lobe (including fusiform gyrus) for object
configural processing of faces
e. g. things that have a face-like configuration are seen as faces
e. g. scrambling parts impairs face recognition
holistic process: all features are perceived together as a unified gestalt
first-order relations: eyes above nose, nose above mouth
second-order relations: exact metric distances between features
holistic processing of faces
- composite effect
It is difficult to judge whether the top part of a face is the same or different, when it’s presented with different bottom halves.
- When the face is upright and the halves are aligned
–>due to holistic processing
- Much easier when the parts are mis-aligned, or the faces are inverted - part-whole effect
It is easier to recognise a face feature when it’s presented within a face.
-Subjects learnt intact or scrambled faces, then they had to identify the individuals, or isolated features - face inversion effect
- Upside-down faces are harder to recognise, slower and less accurate
- Parts are hard to discriminate in inverted faces
holistic processing in prosopagnosic patients
better at discriminating features in isolation than in the face
Young man with long-standing developmental prosopagnosia
Had particular difficulty discriminating features when they were presented in
upright faces.
–> Holistic processing is mandatory, even when it is counter-productive
is face recognition really configural
Burton et al (2015) argue that configural information (ie. Second-order relations) cannot explain our ability to recognise familiar faces
- large configural changes leave recognition unharmed
- Within-person variation is as big, if not bigger, than between-person variation - non-configural changes harm recognition
- Very hard to recognise faces from photographic negatives (same configural relations).
- Drawings traced from photographs are quite hard to recognise, but adding some shading helps.
- Suggests surface texture is important - Face shape vs face texture – texture dominates recognition
expertise hypothesis
Face perception is an example of visual expertise in individuating a visually homogenous class of objects
• We are all face recognition experts.
• People who have expertise with other types of stimuli (e.g., birds, cars, dogs, ‘greebles’, process these stimuli in a more holistic way,
akin to faces.
• Experts recruit the same brain regions (Fusiform Face Area [FFA] and other face areas) when they look at their objects of expertise.
–> there is nothing intrinsically ‘special’ about face recognition
Expertise leads to a change in the entry level of object recognition
Usually, naming is faster at the basic category level (e.g. dog, bird) than at the subordinate level (e.g., Labrador, robin)
• However, people with acquired expertise (e.g., dog show judges) are able to
name individual exemplars as fast as basic level.
• This shift also observed with laboratory-trained stimulus individuation of artificial objects (Greebles).
Expertise leads to holistic coding of objects of expertise
- Inversion effects for objects of expertise similar to inversion effects for faces (Diamond & Carey, 1986)
- Whole vs part advantage for objects of expertise - e.g., individual parts of Greebles, (Gauthier & Tarr, 1998)
- Sensitivity to configural relations – e.g., composite task or changes in 2nd order relations.
–> However, some of these effects have not been replicated and are disputed (Robbins & McKone, 2007)
Objects of expertise recruit the FFA and OFA
- Trained Greeble experts had increased activation in the right FFA and right OFA compared to pre-training.
- FFA was activated by upright, but not inverted Greebles, similarly to the activation in response to faces.
- Bird experts activated the FFA when viewing faces and birds, but not cars.
- Car experts activated the FFA when viewing faces and cars, but not birds.
- Behavioural expertise correlated with the strength of activation.
There is interference between stimuli tapping into shared
expertise processes
• Car experts show less holistic processing of faces when they are simultaneously processing cars (using the composite task).
• Electrophysiological markers of face processing (the N170
component of the evoked potential) are modulated by interference from holistic processing of cars in car experts.
Familiarity matters
We are extremely good at face recognition for familiar individuals, despite huge variability in images
We are pretty poor at face recognition (and even face matching) for unfamiliar individuals, even with minimal variability
Most of the research on face recognition fails to differentiate between familiar and unfamiliar faces
- Focuses on how we can tell individuals apart
- Focuses on tasks that are proxies for actual recognition (e.g., composite effect, inversion effect)
familiar face recognition
Jenkins and Burton (2011)
• We abstract a statistical average of all the instances of a familiar face.
• This average becomes stable after averaging as little as 8 photographs.
• It is stable against contamination from other faces.
Capturing intra-individual variability
The variability within an individual is critical (Burton, 2013)
- It’s not just “noise”
- It provides a “confidence interval” for that person
- It probably varies with the degree of familiarity
A stable representation of a familiar face includes both the central tendency (statistical average) and this variability.
No information about this variability for unfamiliar faces, so unfamiliar face recognition has to rely on picture matching
