Object recognition

Question 1

why study it?

Answer 1

Key component of consciousness

We don’t simply perceive external reality, we represent it.

Object recognition is a key part of this representation process

How we build structure and meaning into our sensory world

Question 2

features of object recognition

Answer 2

Modular – the object recognition system is built of specialised functional modules
- E.g. Anatomical modularity

Constructive – it builds representations from sensory input and contextual information

Semantic – higher level information about e.g. objects’ functions are built into the representation

Question 3

modularity of the visual system: 2 visual pathways (Milner and Goodale, 1992)

Answer 3

see notes

It was originally hypothesised that the visual system could broadly be divided into a dorsal ‘where’ system for locating objects in space and a ventral ‘what’ system for identifying objects

Question 4

receptive fields

Answer 4

see notes

All neurons have receptive fields, which means they are differentially sensitive to objects appearing in different regions of space.

If you present an object within a neuron’s receptive field it will fire.

If you present an object outside the receptive field it won’t fire.

Question 5

temporal cortex receptive field

Answer 5

Receptive fields always encompass fovea

Better at fine discrimination – process features

Question 6

parietal cortex (dorsal) receptive field

Answer 6

60% of neurons have receptive fields that exclude the fovea – good at processing what is going on in the periphery

Question 7

what v where in monkey visual cortex (Pohl, 1973)

Answer 7

In this task specific pairs of objects predict food reward
- Lesions to inferotemporal (ventral) cortex impair object recognition (‘what’)

In this task the proximity of the cylinder to the foodwell predicts reward
- Lesions to parietal cortex (dorsal) impair spatial recognition (‘where’)

Question 8

what v where: neuropsych evidence

Answer 8

(often found in people with a stroke)

Temporal cortex lesions (ventral)

• Visual agnosia
• Deficit in recognizing objects

Parietal cortex lesions (dorsal)
- Deficits of spatial awareness
– Hemispatial neglect – inattention to objects on the opposite side to the lesion, eat half of food on your plate etc.
- Optic ataxia

Question 9

what v where: neuroimaging evidence (Kohler et al., 1995)

Answer 9

Task 1: Spatial locations: Same or different?

• Spatial task > object task
• Dorsal activation (inferior parietal cortex)

Task 2:
Object identities: Same or different?
- Object task > spatial task
- Ventral activation (primarily fusiform gyrus)

Question 10

effects of occipitotemporal (ventral visual cortex) lesion on vision for action and vision for perception (Karnath et al., 2009)

Answer 10

Task 1 – Perception

• Patient must rotate the disc until the orientation of the two ‘slots’ matches
• Errors greater in controls

Task 2 – Action

• Patient must ‘post’ a rectangular object through the slot
• The patient needs to interact with the materials

They found that the patient was impaired on the perception task but performed normally on the motor task.

This shows that ventral lesions impair vision for perception but not vision for action, suggesting the ventral/dorsal distinction may be more along these lines

Only a single dissociation – slight criticism

see slides

Question 11

.

Answer 11

Modularity is more fine-grained than dorsal vs ventral visual streams

Within the ventral stream, there are separate systems for

• Object constancy: recognising objects from different viewpoints
• Integrating features into whole objects
• Recognising functions of objects

Lesions to ventral cortex: Visual Agnosia

Question 12

apperceptive agnosia

Answer 12

Little difficulty recognising common objects under normal conditions

Problems occur when stimulus information is limited or when objects presented from unusual viewpoints

Question 13

unusual views test (Warrington, 1985)

Answer 13

Patients asked to identify objects when seen from their ‘normal’ viewpoint and when seen from more unusual viewpoints

Hypothesised to be a deficit in ‘object constancy’

Question 14

object constancy

Answer 14

ability to recognise an object as being the same object despite retinal input being completely different.

As an example, we can all recognise that both of these images are the houses of parliament even though they are taken from different perspectives.

essential part of our ability to interact successfully with the world around us

Question 15

fMRI reveals a region in the left fusiform cortex that represents objects in a viewpoint independent manner (Vuilleumier et al., 2002)

Answer 15

see notes

Looked at fMRI adaptation – the extent to which activation decreases with repetitions (when you present the same object or word twice, activation tends to decrease – neurons ‘adapt’ their responses to the object – by varying different properties of stimuli you can assess the extent to which a brain region processes that property).

A simple example would be tones of different frequency.

You might present tones of different length and different frequency.

Neurons that process frequency will adapt their response to tones of the same frequency even if the length of the tone differs.

Found reduced activation in left fusiform cortex to the same object from a different viewpoint relative to when different objects were presented.

Thus, this region treats the top two pictures the same even though they’re presented from a different angle, indicating this region may play a role in object constancy – enabling us to recognise an object under multiple different contexts, viewpoints etc.

Question 16

integrative agnosia

Answer 16

Impairment in the ability to integrate features of objects into a coherent whole

see notes

Clearly, this patient doesn’t see the 3 separate shapes, all he sees are individual lines. And he’s unable to integrate these lines into coherent shapes.

Question 17

fMRI shows there’s a brain region - Lateraling Occipital Complex - that combines features into shapes (Kourtzi and Kanwisher, 2001)

Answer 17

In this study, subjects viewed three different types of object – famililar objects, novel objects and non-objects.

Only two of these types of objects required the integration of features into shapes.

The other objects were just collections of disjointed features.

Activation in the lateral occipital complex, part of the ventral processing stream, was higher for the familiar and novel objects than the scrambled non-objects.

This region plays a role in integrating features into whole shapes.

Question 18

associative agnosia

Answer 18

Matching by functions task

• Patients required to match the two objects most closely related by function
• Patients with associative agnosia will choose the two most visually similar items indicating that they are unable to retrieve the functions associated with the objects.

Question 19

fMRI reveals a region in the media temporal lobe that represents objects by their function (Yee et al., 2010)

Answer 19

Words presented that had similar shape, function or ‘manipulation’

Investigated fMRI adaptation effects

Here, several brain regions showed adaptation to function (blue regions) especially in the medial temporal lobe (ventral stream) suggesting that these neurons actually represent the function of an object – essentially, objects can be represented in these regions by their functions, explaining how loss of these neurons can result in associative agnosia.

see slides

Question 20

what does agnosia tell us about object recognition?

Answer 20

Object recognition is a constructive process

• The brain constructs representations of objects based on contextual information.
• These representations, not simply retinal input, are what we are consciously aware of.

Object recognition is a semantic process
- Information about the meaning of an object is automatically processed when we see it, e.g. its function

Object recognition is not actually a single process

• It’s a collection of processes carried out in different brain regions, each one dedicated to solving a different computational problem – object recognition is modular
• Functional modularity vs Anatomical modularity?

Question 21

further evidence for modularity from another type of agnosia: Prosopagnosia

Answer 21

Selective deficit in face recognition

No problem recognising common objects

Can recognise people from their voices

Prosopagnosia can be acquired or developmental

Question 22

faces are processed holistically (Tanaka and Farah, 1993)

Answer 22

Face task: Learn to associate faces with names

House task: Learn to associate
houses with names

After learning, subjects asked to identify a) individual features or b) whole faces/houses

Results showed subjects were better at processing whole faces than face parts but there was no such difference between whole houses and house parts.

Suggests we encode individual faces by encoding the spatial relations between features whereas other objects may involve simply coding of individual features.

see notes

Question 23

the FFA: a face processing ‘module’? - evidence for anatomical modularity (Kanwisher et al., 1997)

Answer 23

There is also fMRI evidence for a separate face processing module

Activation in Fusiform Face Area

One of most reproducible fMRI experiements

see slides

Question 24

trained subjects to recognise novel objects (‘Greebles’) and found activation in FFA in experts but not in novices

Answer 24

Just happen to be experts at processing faces

The greebles look like faces – confound

Activation also more distributed across brain regions - limitation

see slides

Question 25

FFA: faces/expertise

Answer 25

Showed bird experts and car experts pictures of birds, cars and faces

Stronger FFA activation to birds in bird experts and to cars in car experts

However, activation to expertise categories extended over a much wider area of ventral cortex than FFA.

see slides

Question 26

evaluating the expertise hypothesis

Answer 26

Evidence for increased FFA activation for ‘expertise’ is weak and inconsistent – increases are small and several studies have failed to replicate findings

Studies that have found increased FFA activation for expertise have also found increased activation outside this region

Prosopagnosics can become experts at identifying other objects – e.g. case of a prosopagnosic sheep farmer who could recognise individual sheep

Part/whole behavioural effects are observed for faces but not for other ‘expertise’ objects e.g. dog experts

Question 27

fMRI evidence for other category selective processing modules in ventral visual cortex

Answer 27

The parahippocampal place area (PPA)

• A region in ventral visual cortex that activates selectively to scenes (places) – anywhere you can imagine yourself being
• (Epstein et al., 1999)

The extrastriate body area (EBA)
- A region in ventral visual cortex that activates selectively to pictures of human bodies
(Downing et al., 2001)

No other category of objects shows a selective pattern of activation in a circumscribed cortical region

Only biologically important stimuli have dedicated processing modules

Question 28

challenging the anatomical modularity of the ventral visual object recognition system - evidence from multivariate fMRI (Haxby et al., 2001)

Answer 28

Univariate fMRI looks for ‘peaks’ of activation

Multivariate fMRI looks for patterns of activation

Crucial distinction as multivariate fMRI can ask what information is represented in patterns of activation across a brain region

Ratio of within-category to between-category correlation provides a measure of how much category-related information is carried in the voxelwise pattern of activation

For the brain region they were interested in, they converted it into a vector of voxel values and looked at the correlations between vectors.

Across runs, they looked at the within-category correlation and also the between category correlation. Then they computed a ratio of the within:between category correlation.

The logic of this is that if the within category correlation is higher than the between category correlation, the brain region must be encoding category-related information

Within-category correlations consistently higher than between category correlations

Even when they removed voxels that showed higher activation to faces

Suggests a more distributed architecture of ventral cortex than previously suspected