face recognition Flashcards
early model of face recognition
Bruce & Young [1986[
- modular model in that different sub-functions are processed independently
- distinct pathways for recognising familiar faces vs recognising expressions etc
- parallel pathways dealing with facial expressions, facial speech, and ‘visual derived semantic info’ such as sex, age and race
Bruce & Young [1986]
- Different representations constructed for different purposes and for familiar vs. unfamiliar faces.
- For recognition, a familiar face activates a “Face Recognition Unit” – faces previously encountered.
- FRUs are linked to “Person Identity Nodes”, gateways to semantic information about the person.
- PINs are linked to name generation.
early evidence for bruce+young
Memory loss diary study (Young, Hay and Ellis,1985)
- Most common errors:–
- Person not recognized (i.e., ‘blanked’)
- Feeling of familiarity without identity
- Person recognized but no name retrieved
- Person misidentified.*
- Repetition priming found for familiarity decisions but notfor gender or expression decisions (Ellis et al. 1990)
- Familiarity does not influence:
- gender decisions (Bruce, 1986)
- expression analysis (Young et al.1986) since disputed
- Humans can selectively attend to identity or emotion insorting tasks (Etcoff, 1984)
neuropsychological support for Bruce and Young
support for parallelism
Neuropsychological support for parallelism:
- Double dissociation between the processing of facial expression and face recognition. Some have a deficit in identity but not expression and vice versa
identity❌ expression✅ [Bruyer et al., 1983]
identity✅ expression ❌ [Humpherys et al. 1983]
neuropsychological support for Bruce and Young
Neuro-imaging support for parallelism
Different cortical sites are active in the processing of identity versus emotion (don’t worry about which sites!)
- inferiori occipital and lateral fusiform gyro activity:
identity✅ expression❌ [Sergent et al. 1992] - amygdala and superior temporal sulcus activity:
identity❌ expression✅ [Posamentier & Abdi, 2003]
the challenge of semantic priming
- ‘semantic’ priming - a face is responded to faster if it follows a closely related face [eg Prince Charles followed by Diana] compared to an unrelated face. [Bruce and Valentine, 1986]
- → no means to account for this using Bruce &Youngs model
McClelland’s study [1981]
proposed parallel distributed networks that have interactive activation and competition built in as basic processes.
- Concepts and category learning (e.g., Jets and Sharks in West Side Story) via an IAC mode
IAC Model of Jets vs. Sharks
- Semantic information is‘pooled’.
- Knowledge is represented in pools.
- Relationships between Different bits of knowledge are represented in the connections between the pools.
- Connections within a pool are mutually inhibitory
- connections between pools are mutually facilitatory.
IAC model of face processing
- FRUs signal face familiarity, PINs are modality-freegateways to semantic information.
- Details of connectivity and spread of activation and inhibition clarified.
- No separate nodes for names; simply part of semantic information.
- Can explain more empirical data than earlier models
- Relative timing of familiarity, semantic access, and naming (familiarity faster than semantics which isfaster than naming)
- Repetition priming: Laurel’s face primes Laurel’s face.
- Semantic priming: Laurel primes Hardy.
- Cross-modal semantic priming: e.g. Laurel’sspoken name primes Hardy’s face etc.
what do we use to recognise faces?
- human beings are exceptionally capable of recognizing individual faces.
- The challenge of reliably individuating aces is made apparent by the fact that all faces share a basic configuration.
- Every individual face consists of facial features such as eyes, nose, and a mouth that have the same first-order relations such as two eyes above a nose and mouth.
- Although these features are ample for rendering the percept of ‘a’ face, they are wholly inadequate in rendering a percept of ‘that’ face.
what about second order relationships ?
- While there are instances of features that are rather distinctive and sufficiently accurate in signaling the identity of an individual, they are rare.
- Thus, it has been suggested that the efficacy of face coding for the purposes of recognition must exploit second-order relations i.e., the Fine-grained spatial interrelationship between“features”
- e.g. Richler (2009): “…subtle differences inspatial relations between face features beingencoded…
- ”e.g. Tanaka and Gordon (2011): “…encodingof metric distances between features”.
second order account challenges
Hoel, George, Eaves and Rasek [2002] : stretching, squashing, shearing alter second order relationships but faces remain easily identifiable
* Hoel, George and dunsmore [1999]: negative faces are very difficult to recognise despite preserving configural information
where does face processing happen
Inferior temporal cortex Superior temporal sulcus
* - It is widely distributed*
- Some core aspects localizable in the superior temporal sulcus and the inferior temporal cortex
- This is where we see face selective neurons
hierarchical accounts of processing
- cells in the inferior temporal cortex are selective to complex stimuli giving credence to hierarchical theories of object perception
- according to this view early visual cortex codes elementary features such as line orientation and colour. → outputs are combines to form detectors of higher-order features such as corners or T-junctions. cells at the highest level in the hierarchy code specific shapes such as hands or faces.
