the changing face of face research vicki bruce school of psychology newcastle university
TRANSCRIPT
STRUCTURAL ENCODING
FACE RECOGNITION UNITS
PERSON IDENTITY NODES
NAME GENERATION
COGNITIVE SYSTEM
EXPRESSION ANALYSIS
FACIAL SPEECH ANALYSIS
DIRECTEDVISUAL PROCESSING
Bruce & Young (1986)
(Selective) developments since 1986
• Simple ‘box and arrow’ outline replaced in 1990s by computer model – Interactive Activation with Competition
• Much better ideas about the kinds of visual representations that form the core of the ‘FRUS’ or equivalent
• Development of cognitive neuroscience models (Haxby and many others)
• Emergence of ‘social cognition’ and central role played by gaze
Simple ‘box and arrow’ outline replaced in 1990s
by computer model – Interactive Activation
with Competition
Burton, Bruce and Johnston (1990)
• IAC - Interactive activation with competition (cf early McClelland & Rumelhart)
• Pools of units for features, FRUs, PINS, SIUs
• Excitation between pools, inhibition within pools
• Familiarity decisions when PIN reaches threshold
Provides good simulations of• Repetition priming - via strengthened
connections (so long-lasting, but not cross domain)
• Associative priming - via temporary activation (so short-lasting but crosses domains)
• Covert recognition in prosopagnosia
• Predicted face-name matching in patient ME
Name retrieval in IAC?
• Burton and Bruce (1992) proposed names like other semantic information but with fewer connections.
Name retrieval in IAC?
• This position, however, has not stood up to empirical test.
• E.g. Bredart et al (1995) showed that you were not slower (actually faster) to name people about whom you knew a lot rather than a little information.
Much better ideas about the kinds of visual
representations that form the core of the ‘FRUS’ or
equivalent
Burton, Bruce & Hancock (1999)
Cognitive Science• IAC model of person
recognition (familiar)• FRUs driven by
distributed reps - PCA• Look at how model
behaves in recognition and priming now using real faces as input.
Data set
• 50 young men• all captured in a neutral expression and
2 or 3 other expressions
In total• 50 neutral faces + 136 expressive faces
Results
Face recognition
Correct PIN identified Neutralfaces (/50)
Expressingfaces (/136)
Shape-free (50-bit) 50 129 (95%)
Raw image (50 bit) 50 113 (83%)
Shape-free plus shape (70 bit) 50 131 (96%)
Distinctiveness
Human subjects rated neutral versions of faces.(1=typical, 15=distinctive)
Correlation between human rating and cycles-to-reach-PIN
= - 0.31
Semantic primingPairs defined as sharing 2 semantic units
Mean cycles to threshold for test faces
Unrelated prime Related prime
Face prime 65 38
Name prime 63 41
Repetition priming
Procedure:1. Present prime face2. Cycle model & Hebb update3. ISI - present lots more faces (c. 100) 4. Present test face (same or different view)
Mean cycles to threshold for test faces
Unprimed Primed with sameimage
Primed withdifferent image
78.6 60.1 64.8
How do we represent familiar faces?
• Just the average of each distinct image we see of them?
• See Burton, A.M., Jenkins, R., Hancock, P.J.B. & White, D. (2005) Robust representations for face recognition: The power of averages. Cognitive Psychology, 51 (3), 256-284
• Jenkins, R. & Burton, A.M. (2008), Science, 319, p.435.
What about Face Space?
• Valentine (1991) and later
• Adaptation studies (Rhodes et al..)
• PCA dimensions can be thought of as forming the dimensions of ‘face space’ (though this is not the only possible model)
Are faces special?
Or, is face recognition special?
• Innateness (congenital prosopagnosia, congenital cataracts suggest sensitive period)
• Localisation (FFA active even in congenital Ps)
• Specificity (still debated...)
Exciting hot topics...Gaze
• Information from dynamic patterns
• Interactions between systems
• Gaze and social cognition: certainly eyes are special..
• But why eyes?
STRUCTURAL ENCODING
FACE RECOGNITION UNITS
PERSON IDENTITY NODES
NAME GENERATION
COGNITIVE SYSTEM
EXPRESSION ANALYSIS
FACIAL SPEECH ANALYSIS
DIRECTEDVISUAL PROCESSING
Bruce & Young (1986)-dynamics-interactions-gaze!
Eyes important for..Social reasons
• We look at other people’s eyes for
• Intimacy
• Control
• Regulating conversational turns etc
Cognitive reasons• We look at other people’s eyes to
– Mind-read (Baron-Cohen)– Establish shared attention– Dogs do this too..(Miklosi et al, 2003)
• Can’t ignore what another person gazes at– Gaze cuing– But sometimes we must look away (gaze
aversion)
• Different gaze patterns in different genetic learning disorders
So, why eyes?
• We need to look at them/use them for other social and cognitive purposes
• They tell us about gaze and also other expressions
• They don’t change when other facial features do.
• Probably explains why representations of familiar faces are weighted to the eyes.