Cognitive Neuroscience Lecture 9: Face and Scene Recognition

''L9: Face and Scene Recognition ''

The dress: color constancy

Specialization in ventral occipital cortex

-         Sub-regions specialized for particular visual categories

o E.g. faces, places, words, bodies, other ppl’s thoughts

Face vs. Scene Area

-         Fusiform Face Area (FFA): right fusiform gyrus: Faces>scenes

-         Parahippocampal place area (PPA): Scenes>faces

-         PPA is superior and medial to FFA

Fusiform & Parahippocampal gyrus: WHICH OF THESE DO WE NEED TO KNOW?

-         Fusiform: lateralish; right ffa.

-         Parahippocampal: medialish

-

“If neuroimaging is the answer, what is the question?”

-         It’s not good enough to just put a bunch of people thru an fMRI during a task and then retroactively assign meaning to the observed activated parts

-         You should have targeted question about targeted area

Useful types of imaging studies:

-         Characterization of a single neuron’s responses

o What happens with one vs another

-         fMRI study on Face

o Kanwisher FFA paper most successful empirical fMRI paper (not method/survey paper)

o Studied macaque temporal lobe

§ Showed cells that specifically responded to faces, not scrambled faces or other obj

o Humans lose ability to recognize faces but not other objs

§ Prosopagnosia

o Human fusiform gyrus was implicated in face in earlier PET studies

§ They gave it a catchy name

-         Background: fusiform gyrus activated for faces, but control stim weren’t controlled for

-         Question: is processing of faces distinct from processing of obj when other confounds are eliminated?

1.      Face localizer scans

a. Faces vs objs.

b. Block design (faces, fixation, objs, fixation)

c. Higher during face blocks

d. '''Subtraction method''': faces – objs = face specific area

2.      ID ROI in each individual

a. Is there a place in each individual that responds more to faces than obj?

b. Is this place the same in all subjects?

c. yes in 12/15 subjects

3.      Test alternative hyps in ROIs: low level diffs, exemplars, animate?

a. Could be that FFA is for any low-level feature differences (overall contrast, luminance, shape?)

i. If FFA is truly face selective, it should respond more to faces than '''scrambled faces'''

ii. This would mean it’s not responding just to anything with differences in luminance/contrast

iii. Block design of Scrambled vs Intact faces shows it does

b. Could also be that FFA is for anything that is exemplar of a category

i. Faces are exemplars of a face category but there are multiple basic-level category of objects

ii. If FFA is truly face selective, it should respond more to faces than houses even though individual houses, like faces, are exemplars within a category of similar objs

c. Could be that FFA is for anything animate

i. If FFA is truly face selective, it should respond more to faces than hands

ii. It does (block face vs. hands)

iii. What if it’s just for anything that requires more attention?

iv. When 1 back task forced attention to hands, it still didn’t light up for handsà not attention

Conclusions

-         A part of '''right fusiform gyrus''' is preferentially active during face viewing

-         “faces are special”à not just another stim

Why was this paper so successful?

-         Discovered area that became hugely studied

-         Goal was not just localization but a deeper understanding of brain activation patterns

-         Theoretically driven

o “are faces special”

o Relates well to other lit

-         Unambiguous

o Good controls

-         Beginning of controversy

o Category specificity

o Nature vs. nurture

At what level does the brain represent categories?

-         Many more questions…viewpoint, ID vs appearance, facial expression, facial parts, “special”

fMRI adaptation: what counts as the same in the brain?

-         If you show a stim twice in a row, you get a reduced response the second time

-         fMRI adaptation/attenuation

o “different” trial: high response both times

o “same” trial: low response second time

o When you do it a second time, neurons are tired and you get less activity if you do the same thing twice (same neurons activated)

-         Why is adaptation useful?

o Now we can ask what it takes for stim to be considered “Same”

o E.g. viewpoint

-         Repeated individual, diff viewpoint

o Possibility 1: viewpoint specificity (Area codes face as diff when viewpt changes; high response)

o Possibility 2: viewpoint invariance (area codes face as same despite viewpt change)

-         Results

o Identical<Translation=size<rotation<different face

-         But is it ID of faces or physical appearance?

o Test Maggie vs. Marilyn Monroe: facial identity

o Identicalà repetition, fMRI adaptation

o Within: Physical different, identity repeated…will it be treated as identical or between?

§ Physical difference % is same in within and between, (30% in both cases) but in “between” it crosses identity boundary while in “within” it does not

o Within is more similar to identical than to between

§ Suggests that FFA codes for identity

o Within behaves similar to identical in FFA, but within behaves similar to between in OFA

§ OFA doesn’t code for identity; codes for PHYSICAL changes

o '''Identity in rFFA, physical in OFA '''

-         Facial expressions

o Superior temporal sulcus (right below superior temporal gyrus)

o 2 hypotheses: categorical (discrete) and graded (continuous)

o Facial expression & emotion in superior temporal sulcus

§ '''Left STS: graded'''

§ '''Right STS: categorical (clusters)'''

-         Facial Parts

o TMS à virtual lesion, test causality

§ FFA is too medial for TMS

o Occipital face area (OFA)

o Same or different parts?

o If you disrupt right OFA, it gets fewer correct than control

o Facial parts processed early (before 100 ms of presentation) in OFA

§ Possible because TMS has higher temporal res

-         Are faces special, really?

o We are all experts on faces

o Car experts had FFA active for cars

o Bird experts had FFA active for birds

o Can you train people to be experts and see if FFA activates? YES (e.g. greeble)

o However '''face-like is innate''' (infants look longer at faces)

§ Experience also important though

§ Prefer familiar/female/same race (mediated by exposure though)

-         Monkey deprivation study: is the preference for faces innate?

o Infants raised by humans & deprived of face exposure

o Only exposed to obj

o Initial exposure to either human OR monkey face

o After 1 month: both

o Before exposure, monkeys preferred humans&monkeys in photos over objects

§ Goes against expertise

o Suggests '''experience-independent ability for face processing''' (innate)

-         Summary: At what level does brain represent categories?

o   What happens if you change viewpoint?

§  Sensitive to viewpoint changes (rotation)

o   Is it ID or physical appearance?

§  Identity in rFFA

o   How about facial expressions?

§  Facial expressions in STS (left = graded and right = categorical)

o   How about facial parts?

§  Facial parts processed early in OFA

o   Are faces special, really?

§  Yes and no (innate [monkeys] and can be learned [greeble])

'''Category selectivity in Ventral Visual Cortex '''

-         Parahippocampal cortex: Parahippocampal Place Area (PPA)

o Scenes > Faces/ objects

-         PPA: right above cerebellum, right below ventricles

-

-         Parahippocampal gyrus: below the hippocampus, left side

-         Fusiform gyrus: lateral

-         PPA: Empty room, furnished room, lego room, fractured room as long as spatial layout intact, houses&landmark, geometry

o Not furniture

-         Perhaps PPA computes features that describe geometry of scene

o Nope; real scenes give higher % signal change than scrambled scenes, rooms, objs

-         Viewpoint invariance or specificity in PPA? Ask using fMRI adaptation paradigm

o Obj change, viewpoint change, place change

o Identical<obj<<viewpoint change (VIEWPOINT SENSITIVE)

o Viewpoint change treated as entirely different images

-         Scene processing network

o Both for landscapes and cityscapes

-         '''Retrosplenial Cortex: location task, familiar scenes; WIDER WORLD '''

o '''Both needed for location ID task, but RSC more preference for it '''

o Responds to location ID more than category or situation ID

o More active during location and orientation compared to simple familiarity tasks

o More active for familiar scenes compared to unfamiliar scenes

-         PPA vs RSC

o   PPA doesn’t care as long as it’s a scene

§  Just processes visual information of scenes & geometry of space

§  Contextual association, not scenes per se

o   RSC is more for familiar scenes, location tasks, and orientation tasks

§  Recruiting memory representation

§  Navigation/orientation

o RSC damage:

§ Can ID scenes/landmarks

§ CANNOT orient selves in wider world

§ Expansive representations  allow current scene to be placed in larger spatial framework

o PPA damage:

§ CANNOT ID scenes/landmarks

§ PPA represents visuospatial structure of individual scene

-         Contextual objects

o Do scene specific areas only care about 3d layout and location, or do they also care about obj associations and schemas?

o Contextual objects (e.g. beach chair) activate scene-specific areas

o So, PPA is not for scenes specifically…just things that associate with scenes

o Some evidence for object schemas in PPA and RSC

o Areas not just layout; also association

Lateral Occipital Complex

-         Matches damaged region in agnosia patients

-         Object shape and luminance, not dependent on motion patterns

-         SPECIFICALLY for objects, not just textures

-         Represents ‘perceived’ object shape; bars in front don’t matter, texture doesn’t matter

-         '''Size invariant''': sensitive to semantic changes, but not size changes

Extrastriate Body Area: mostly left hem

-         Body parts except faces

-         Weak response to objs

-         Intermediate response to facial parts or mammals

-         '''Bodies>other objs/shape'''

Modular vs distributed code for categories? Still ongoing debate

-         Different categories show different patterns across ventral visual cortex; ffa responds to shoesàDistributed coding of objects throughout entire ventral visual cortex

-         Evidence for category-selective processing (modular view): Triple dissociation with TMS of objects, bodies, and facesà supports modular coding of ventral visual cortex

'''^ Evidence for both. '''

We don’t know of other localized categories yet.

-         Maybe organization too fine for fMRI res

-         Maybe another way to think of obj representation

-         Evolutionary reason for categorization

Shape selectivity in LOC

-         New shape on old texture gives almost as big a spike in LOC as new shape on new texture

-         Shows that it really only cares about shape