Testing face space models
Terms
undefined, object
copy deck
- Johnston & Ellis 95
-
Uniform (same as adult so uniform density
Differential model kids - fewer dimensions adult so same density - ?
-
5 yrs
D fx 4 classifcation
no fx 4 recog- consistent w norm-uniform or exemplar differential - Brennan 85
- Caracature generator
- Rhodes 87
- caracatures donÂ’t always enhance performance - decrease in exemp density doesn't work sometimes
- Benson & Perrett 91
-
Add texture
best recog 4.4% - Stevenage 95
-
-c = increae vector length (angle same)
more D so easier to recog - Tanaka and simon 96
-
build neural net model C fx
average from 3 vectors = norm
exaggerate to get C
c= better recog
supports norm based model - Carey (2002)
-
lateral caricatures
norm account
incorrect generate laterals - Rhodes 98
-
lateral = btw anti-c and veridical
non- consistent norm act - Lewis & Johnson 98
- Lateral C no -ive fx inconsisten euclidean distance for recog
- Exemplar : Rhodes 93
-
Absolute coding
makes image more Diff
more competing reps but more diff stored rep
balance= c- better veridical - Lewis and Johnston 99
- Veronoi model
- Boyatt & Rhodes 98
-
Other race faces and caricatures
C advantage when exaggerate away other race norm
consistent - Lee & Perrett 97
- brief presentations caracture adv at 40% exaggerated = brief exposure fx
- Probs w voronoi
- identity regions- fills whole veroni, therefore false positive for unfamiliar face
- face space r
- similar to voroni - comparision btw exemplars
- Nosofsky 86
- generalized context model
- Valentine ferrara 91
- generalized context model linked to face space
- Generalized context model
- uses summed similarity rule: probability of response = sum of probe to exemplars devided by sum of similarity of probe to all exemplars
- Probs w Gen. Context model
- highly distinctive - high caracature recognition & unable to account for caracature fx
- Zaki & Nosofsky 2001
- P's learned new faces, response to morph face greater than parent faces of morph
- Lee Byatt & Rhodes
- GCM- explains caracature fx : 16 famous faces - recog carac, veridicals, anticarac
- GCM- hig recog though extremely caracatured
- - wrong as face stops looking like face after while
- Lewis 99- familiarity
- people seen a lot are faster than less salient
- Valentine2001
- new face = new encounter ed : overcomes false positive rct in voronoi
- computational
- input randomly generated normally distributed data ; add parameters
- Shephard 77
- multidimensional scaling on pattern of clustering, 100 faces 32 most important faces rated, 10/11 factors account for variance
- lewis -face space-r
- 15-22 dimensions
- Turk & Pentland
- similar to no of eigenfaces employed to generate recognisable face
- Light 79
- Mean dist from centre 4 typical = shorter than for distinctive
- Burton, Bruce & Dench 94
- distinctiveness= dist btw face representation and av face
- Burton & Vokey
- Typicality paradox
- typical
- densely populated
- distinctive
- sparse
- lewis
- larger no of dimensions the smaller the skew
- too large dimensions
- decreases variability of distinctiveness
- skew
- larger skew in ratings may b due to fewer dimensions used to process 3/4 faces that frontal
- Benson & Prett 94
- line drawn faces - caracature fx better for typical than distinctive relationship is curvilinear - sim to -ive power function w looking at distinctive and caracature for best likeness
- Bothwell brigham and malpass 89
- other race harder to recog than own
- own race faces
- high exposure fx
- valentine 91 - face space and race fx
- other race faces clustered far away
- norm based race faces
- hard to disting btw norm vectors and other race faces
- Valentine & endo 92 -exemplar based w races
- have less contact therefore less representations
- destinctions btw races
- blue eyes etc
- Byatt & Rodes 99
- agreed w absolute -coding - caracature advantage great for other race faces when away from norm
- Valentine & Bruce : face categorisation
- : typicality advantage
- Valentine & Bruce 86
- rctn time for familiarity slower than categ
- sum of activations of exemplars
- threshold dependent on non -face probes (scrambled)
- Lewis 98
- found typicality fx for non-prob cat faces vs human faces
- unfamiliar : lewis & Johnston 99
- veronoi model cant deal w unfamiliar faces ; probe faces recog as familiar person but at diff speeds (depending on cell distance)
- Valentine 2001
- threshold activation
- Lewis & Johnson 97
- face learning memory task
- Face learning memory task
- accounts for false positive advantage
- Valentine 91
- Absolute encoding & race faces : learn to expand face space and decrease exemplar density
- Busey 98
- location of face space det factors like distinct
- Bartlett & hurry 80
- typical face: familiarity, feeling of "oldness"
- Typicality : Vokey & Read 92
- attractiveness, familiarity, likeability. Memorability,
- context and typicality
- context induced- prior exposure
- Context free
- structurally induced- memory not indexed
- Bartlett 84
- distinctive have high familiarity- prior exposure, easy to encode , context-free
- O'toole, Deffenbacher 94
- digitised pics, put into neural netwrk, white vs asian, memorisability small, local features used, familiarity global aspects used
- Uttal Barauch & Allen 95
- high spatial freq- used for discrim : global shape info underlies familiarity
- Race bias: brigham malpass 82
- field studies
- Lab studies of race bias
- Messiner & brigham 2001
- Goldstien & chance
- race fx : tested white children and adults recog white japenese adults - scores increase w age, white better than japenese recognised
- MacLin and Malpass 2001
- cross race fx sim from 5 yrs to adulthood
- Wright boyd & Tredoux 2003
- own rave bias- people better at recog own face : south africa vs uk : shown pics black and white, had seen faces in deck?; P's give q abt interacial contact; confidence accuracy best w own race
- Malpass& Kravitz 69
- prob w data: low recog performance for other race faces- differential experience
- O'toole 95 perceptual expertise can help
- Anthony 92
- consistency across racial groups : 2.5 x the variance of effect in white vs blacks
- Absolute encoding
- distinctiveness and change in exemplar density
- caractures & absolute encoding
- caractures high sep from other points (potential distractors) = distinctive
- anticaracatures
- decrease exemplar density
- Rhodes & Mclean 90
- high contact norm shifts to rep all faces in space
- Absolute encoding & race faces
- Dense clustering around feature (asian/ white) therefore harder to recog
- Byatt &Rhodes 97
- absolute encoding & race faces supported best
- See pics
- Veridical
- undistorted
- exemplar
- absolute encoding model
- Shepherd, Ellis davies 77 distinctive faces
- multidimensional scaling ; faceshape, hairlength; sig dimens 3/4 dimensions encoding face
- Johnstone & Milne
- use multidimensional model and describe unidimensional scale - typical vs distinctive
- Lewis 2004
- towards a unified account of face recog 15-22 dimensions
- Face space additional dimensions
- for facial featurs ie eyebrows
- Tredoux 2003
- own race bias
- Connors 61
- DNA vindicated prisoner for nearly decade, cell partner convicted for rape
- Wright 2001
- cognitive specialising ? Tested non students in shopping centre: Black - better at black than white
- Kassin 2001
- 90 experts agreed to ORB- reliable for scientific testimony
- Slone 2000
- conflicts over orb
- Messinger & brigham 2001
- small fx for orb - across data and across diff methods and conditions
- Chiriro & Valentine
- results: Black high contact : no own race bias, low white and high white- both good at white , black low good at black
- Wright, Boyd &tredoux 2003
- both white and black accurate at white faces
- Chance & Goldsein 96
- majority of studies find fx
- Kassin 89
- important that surveys are reliable
- Chance & Goldstein 96
- studies - replicatability
- Bothwell, Brigham & malpass 89
- 80% samples ORB fx : replicatability
- Malpass 80
- generalisability : use standard recognition paradigm across studies
- Fallshore 95
- generalisability : used matching task
- Lewis 96
- also other race classification advantage: comes from facilitated classification process
- Tanaka & Taylor
- ORCA sim fx to novices in basic level categorisation fx
- Metanalysis- fx ORB
- own race: mirror fx, high hit and low false alarms, 2 discrimination accuracy 3 whites more likely ORB 4 study time = increases ORB 5 date of study (cohort fx)