Cognitive Neuroscience Lecture 8: Object Recognition

Recording info that wasn’t finished last class: ERP, EEG, and MEG are very good temporal resolution but low spatial resolution. Useful when you care about timecourse.

Connections beyond V1: we’ve studied what happens in V1 so far

There’s more than just patches and bars that we understand.

EyeàLGNàv1àsplits to dorsal and ventral path

Simplified to two pathways: further processing in dorsal/ventral pathways

-         dorsal: “where is the object”: spatial

-         ventral:  “what is the object”: identification of object

Simplified because there are other things in between; not always dichotomous

Monkey experiments are good because they have similar visual systems to us

Also good bc can directly test function/role of area by artificially lesioning brain region

What vs. where experiment

-         Task 1: What—Object discrimination

o Food was hidden under one of them

-         Task 2: Where—Landmark discrimination

o Same shape but distance between obj with food and landmark was different than distance from non-food and landmark

-         Performance good for both tasks in a healthy monkey

-         Performance bad in landmark task for monkey with '''dorsal '''lesion

-         Performance bad in object discrimination task for monkey with ventral lesion

-         This is double dissociation of object and landmark

-         Ventral àwhat, dorsalà where/spatial locations is shown by this experiment

Double dissociation in Humans

-         Can’t do artificial lesion, but using neuropsych, studied people with preexisting lesions

-         Showed selective impairment of tasks

-         Ventral lesion

o CO poisoning

o Perception problem: “what” deficit

o '''Visual-form agnosia/object agnosia: '''impaired object recognition NOTECARD

o '''Prosopagnosia: '''specific to faces

-         Dorsal lesion

o Stroke

o Action problem: “where” deficit

o '''Optic Ataxia: '''had trouble reaching to a spatial location NOTECARD

-         Did similar double dissociation experiment

-         Task 1: Perception—are these identical?

-         Task 2: Action—reach out and grab this

-         Patient with ventral lesion was unable to do Task 1 (perception task), but was able to do Task 2 (action task)

-         Patient with dorsal lesion was unable to do Task 2 (action task), but unable to do Task 1 (perception task)

-         Basically, '''dorsal important for action (since orientation is required) and ventral important for perception'''

-         Ventral lesion patient: object agnosia—difficulty recognizing objects; but how do we know it’s this?

o Alternative hypotheses: It could be a problem with memory of word, difficulty articulating, vision problem, she just can’t name them

o We know it’s truly a visual object recognition problem because she was able to tactilely recognize them when they were in her hands

o This shows that she knows what they are but just cannot VISUALLY recognize them

o Further evidence:

§ Could draw pictures from memory

§ Could not recognize or copy pictures given a model picture

§ Could '''not '''recognize her own drawings later on

o Matching orientation task in ventral lesion pt

§ '''Explicit matching task (perceptual orientation matching): Couldn’t match something she was holding to something she was looking''' at

§ '''Action task (visuomotor “posting”): '''But when performing the action, she could align the object to the slot

-         Dorsal lesion patient: optic ataxia (opposite of ventral lesion patient)

o Fine with recognizing and matching

o Difficulty with visually-guided reaching

o Make inappropriate grasping movements with hands (does not align hand with slot)

-         This was a double dissociation between the two patients

-         '''DORSAL/ACTION/WHERE: spatial action '''

-         VENTRAL/PERCEPTION/WHAT

-         '''Main point: '''visual processing involves several distinct pathways

Further processing in dorsal/ventral pathways

“Where”/action: Dorsal stream NOTECARD

-         Mostly parietal

-         V1àSuperior longitudinal fasciculusàposteroparietal cortex

 “What”/perception: Ventral stream NOTECARD

-         Mostly temporal

-         V1àInferior longitudinal fasciculusàinferior temporal cortex

Oversimplification—connections between them; feeds forward, back to V1, and all directions

Despite this, fMRI shows that distinction is somewhat true

'''Evidence for dorsal/ventral pathways''' NOTECARD

1.       Animal lesion studies (monkeys)

2.       Human neuropsychological cases (DF&RV)

3.       Single unit recordings

-         What: temporal neurons fire for color, shape, faces

o   E.g. will fire for shape of hand regardless of orientation, but will fire less when given a less detailed hand shape

-         Where: parietal neurons fire for change in direction of motion, velocity, control of movements of attention

o   E.g. change if you move your attention to left side to right side

o   To know that something is moving is to know that it has changed location (where)

4.       Human neuroimaging studies with intact subjects

-         Remembering what vs. where

o   Healthy normal adults

o   Position task: “are the objects in the same locations”

o   Object task: “are these the same objects”

o   Same stim, diff task: location vs. identity

o   Object task lights up ventral/temporal area

o   Position task lights up inferior parietal area

o   '''Position/object tasks light up different areas even in neurologically intact patients when they were looking at same stimuli'''

o   Results of PET study

§  Dot-location matching used parietal more

§  Face matching used ventral/temporal more

-         What/where: color/motion

o   Color: compare brain response of color vs. black&white sq

§  ventral

o   Motion: compare brain response to stationary vs. moving black and white regions

§  Dorsal 

o Results: Color is medial (V4), motion is lateral (MT)

'''Lateral occipital complex: object recognition'''

-         Lesions in people who cannot do obj recognition are same as location of LOC in intact people

-         Selectively responds to object information (e.g. not scrambled objects)

Further processing in dorsal/ventral

Computational problems in object recognition:

-         '''Object constancy (car from different views, diff background) '''

o We can recognize objects across changes in illumination, size, occlusion, and viewing position

o diff image on retinaàhow do you still recognize?

o Does take more time to recognize noncanonical

-         '''Grouping of features and parts of objects (Gestalt principle)'''

è Help determine what bits of an image belong together

o   Proximity: closer things get grouped as 1  object

o   Similarity: similar get grouped

o   Closure

o   Good continuation: 2 lines, not scattered

o   Good form (e.g., regularity, symmetry)

-         Use of stored object knowledge

o We can still identify objects even when there aren’t local cues for grouping

o '''Top-down processing''' to ID obj

§ If you already know it’s a dog, you can tell it’s a dog

How do we achieve object recognition? Some theories:

-         Template matching: exemplars 

o Compare input to stored template of an object; yes or no

§ Then respond whether a match or not

§ E.g. stored template ‘B’ would match ‘B’ but not ‘A’

o What’s the problem with this?

§ Doesn’t work for novel objs

§ You could have something that is still a capital B but is a different font

o You can recognize it as something different in a different context

o What’s the template for a chair? You recognize many things as 1 type of obj

o Also you can categorize things that aren’t evolutionarily coded in a template

o '''Problems in template matching '''

§ Variability of exemplars

§ New objects

§ Changing viewpoints

§ The idea of a grandmother cell rejected

·       Grandma changes over time, but you can still recognize her; must not have 1 cell for 1 exact face, then.

-         Recognition by components: Geons 

o “Alphabet” of vision

§ Sphere, block, prisms, etc.

§ Geons: basic building blocks of all object recognition

§ '''Objects = Geons + Spatial relationships '''

o What geons can’t do:

§ Some things can’t be reduced to geons easily

§ Fine with general categories, but not individuals: all would break down into same set of geons, so how do we tell difference between dogs

Deficits in object recognition:

-         Visual agnosia (DF)

o Different from memory loss; tactile enables recognition in agnosia, meaning that they ''know the object but just can’t recognize it visually''

-         Stroke patient: could see visual details and make appropriate hand gestures

o Could not describe obj functions though; used particular parts to infer it was something else like phone

§ Could not put parts together into whole

o '''Apperceptive Agnosia '''

§ Right lateralized (found in people with right damage)

§ Posteriorà earlier in visual processing stream

§ Failure in basic perceptual processing

·       We can tell things from fragmented line drawings, but they cannot

o '''Associative Agnosia '''

§ Left hemisphere

§ Anterior à Higher level processing failure

§ Can perceive obj but cannot assign meaning

§ Fail at matching-by-function

·       Only can match by form

§ Category specificity shown by HSV patient

·       He got common obj 90% correct

·       but live things only 6%

§ What are implications of category specific associative agnosia?

·       We interact more with inanimate things (sensorimotor regions also used, not just vision)

o G.S. motor activation to manmade obj

o Left premotor, sensory becomes active when viewing tools

o   Regions:

§  Extrastriate Body Area (bodies)

§  Fusiform face area (faces)

§  Middle Temporal (Motion processing)

§  Lateral occipital complex (Objects)

§  Parahippocampal place area (places)

§  Superior temporal sulcus-face area (faces)

·       Visual areas are segregated by living vs. inanimate obj

o Specificity in ventral visual stream

o Large scale segregation in ventral stream for living vs nonliving

·       Summary of category-specific agnosia: 2 explanations

o Motor system engaged when viewing inanimate obj à agnosics ID these better than animate

o Ventral visual system is segregated by animate vs inanimate shown in fMRI

§ Vasculature implies that animate regions more prone to damage in encephalitis/stroke

§ Animate is usually one lost; we’ve never seen only inanimate lost

o '''Integrative Agnosia '''

§ Right/left hemisphere

§ Cannot put features into parts, or integrate parts as whole

-         Category-specific defecits

o Apperceptive agnosia

o Integrative agnosia – object recognition

o Associative agnosia

o Alexia—written word recognition (reading) deficit

o Prosopagnosia—face recognition deficit

§ Sometimes following damage to FFA

§ Sometimes congenital

§ Can recognize that it is a face, but not identity of face

§ To ID, use non-facial info like hair/clothes

o What’s special about faces?

§ More accurate at identifying whole face than parts

·       Not true for other obj; just faces

§ Holistic processing of faces: upside down is hard bc holistic effect disappears

§ Evidence from object agnosia

·       Obj recognition difficulties

·       Normal objects are confusing, but agnosia pts see face

·       Obj recognition is diff from face recognition

§ Evidence from single cell recording

·       There are cells that respond more than others to faces

§ Fusiform gyrus: Right fusiform, Fusiform Face Area

Faces activate it more than scenes