Block 2- Object recognition

Question 1

What is the problem with object recognition?

Answer 1

There is an infinite number of retinal images that could correspond to a particular object
So how are our brain able to look at all of these different images, and interpret them as the same thing

Question 2

What is the solution to the problem of object recogintion?

Answer 2

= Our brain uses the visual input and matches this with internal object representations held in object stores ( eg held in memory)

eg if asked to imagine an elephant, you can do that because you have seen an image of an elephant before, you can activate a visual representation of an elephant in memory

(However you cannot do this if you have aphantasia)

So object representation really mean - mental image of what we think an object looks like

So if the visual input matches the representation, then we have object representation

If the visual input does not match the representation, then we do not have object recognition

Question 3

How does the matching process work?

Answer 3

We have possible different theories for this

• Marrs theory of object recognition ( Structural analysis)
• Ellisons anf young theory of object recognition ( built off of structural analysis)

Question 4

What is the main idea of MArrs structural analysis, give analogises

Answer 4

Two objects may differ in their 2-dimensional appearance, but they become equalaivulent when they have a 3 dimensional representation

ie visual inputs must be analysed at different levels
-WHen presented with this 2 dimensional elephant along the bottom, difficult to match with the two dimensional elephant on the top

Its not easy to compare the two images and decide that they are the same
While, if you have a 3-dimensional model of an elephant in front of your inner eye, which you can mentally rotate in all directions, it is not as difficult to decide that this is all of the same elephant

Question 5

What are the 4 sequential levels of structural analysis?

Answer 5

• Raw primal sketch ( Initial representations)
• Full primal sketch ( initial representations)
• 2 1/2 D sketch ( Viewer-centred representation)
• 3 D sketch ( Object centred representation)

Question 6

What is the raw primal stage of structural analysis?

Answer 6

• This is literally a distribution of grey scales
• More or less light that is distributed on your retina
• So some areas of your visual field are darker or lighter

Question 7

What is the full primal stage of structural analysis?

Answer 7

-This is a little more elborate - puts together basic geometrical organisations

eg edges become clear, contours become clearer

Basic shapes may appear

Question 8

What is the 2 1/2 D stage of structural analysis?

Answer 8

So we just perceive whether some objects are closer or further away from us

That is why it is called the 1/2 dimensional sketch, because it includes some depth information

It is viewer centred representation

It doesn’t necessarily help us to recognise an object, but it can ( only if the representation is prototypical)

Question 9

What is the 3D stage of structural analysis?

Answer 9

-In order to fully recognise an objec, we need a 3 dimensional model ( not a sketch)

-Includes depth in that we can rotate the object mentally

So this is an object centred representation , not dependant on our viewpoint

This can then be compared and matched with the representation that we hold in memory

Question 10

What is so good about Marrs hypothesis?

Answer 10

-They have inspired a cognitive model

-This is important because cognitive models make predictions about human behaviour

-This is good because we can then test test predictions, and thus verify or falsify the model

-The more evidence that we find the confirms the models predictions, the more reliable and valid the model is

The cognitive model that this inspired is the Ellsion and Youngs model of object recognition

Question 11

What is Ellion and Youngs overall model look lile - Draw this out

Answer 11

Question 12

Explain how cognitive models work

Answer 12

• Cognitive models always have boxes - these boxes always represent functional processing stages
• Between the boxes you will find arrows -these arrows show the direction of information flow
• So you can see whether a model is a hierarchical model, starts at the top and goes down
• This would mean that whenever a processing stage does not work, this means that all following processing stages are interrupted as well
• Some models are not hierarchical - more like networks

Question 13

What is the initial represenation stage of the model?

Answer 13

We then generate an initial representation of this object (this is what Marr would call the primal sketch)

So this is the levels of brightness on our retina, first definitions of edges and contours ( eg we talked about receptive fields which are specialised for this)

So all about the grouping and segmentations -in terms of gestalt psychologists

Question 14

What is the viewer centred recognition stage of the model?

Answer 14

This primal sketch, is then transformed into a 2 1/2 dimensional sketch

So these edgess and contours are now assembled - so that there is an object representation

However, is it an object representation that is solely based on our viewpoint

Question 15

What is the object centred recognition stage of the model?

Answer 15

. The 2 1/2 D sketch is then transferred into a 3dimensional model

This objected centred representation can be manipulated mentally, it’s the real shape and position of objects in the visual scene

Independent of the observers viewpoint

This is the object representation that we can then use in order to compare to the object representations that we hold in memory

Question 16

What is the object recognition stage of the model?

Answer 16

This is where Ellison and Young expand Marrs original theory)

So we have the memory storages/spaces that holds all of the object representations that we accumulate in a lifetime

Every new object that we meet, will create a new representation in our object representation units

And when we find an object, we will pull out a representation that matches these visual inputs the best, which will hopefully allow us to recognise the object

( however in the model there is also a shortcut from the viewer centred representation to object recognition units)- Recognition shortcut

So this is for unambiguous viewer centred representations ( eg where you can still recognise the object and dont need the 3dimensional model

It is a prototypical representation of an object

Question 17

What is the semantic system stage of the model?

Answer 17

Once the visual input is matched with the object recognition units, that allows us to gain access to our semantic memory system in which we store general information, but also personal facts about objects. This also includes the name of the object

Question 18

What is the name retrial stage of the model?

Answer 18

Accessing these facts generates the name retrieval

Question 19

How can these 7 stages be grouped into 3 sections?

Answer 19

We have the initial representations- which we can refer to as the sensory processing

We have viewer centric and object centric representations - Which can be put together as perceptual processing

And then we have the recognition units - Which we can put together and refer to as recognition

Question 20

What does it mean by testing the cognitive model?

Answer 20

• The way we test cognitive models, is using observations of human behaviour - and try to see whether the model explains this human behaviour
• If it does, we have evidence for the model, if it doesn’t, we have evidence against the model

Question 21

What are the 3 examples we use to test the cognitive model?

Answer 21

• Mr M
  -Righter posterier lesion
  -Patient AB

Question 22

Describe patient Mr M

Answer 22

Who- Reported by Benson and Greenberg

What happened - Suffered carbon monoxide poisoning

What are the neurological effects -
CAN-
-Maintain fixation
-Name colours
-Describe sensations
-Detect small changes in brightness and wavelength
-Detect movement of small objects

CANNOT
-Percive shape or form
-Cannot copy simple figures, letters, digits
-Cannot match simple objects
-Recognise or name objects and pictures of objects

We can seen that he does not have viewer-centred representation, as he has no pereception of basic shapes

Therefore, he does not have the rest of the model either

Question 23

Describe the pateint with the right posterior lesion

Answer 23

Who - Reported by Warrington
What happened- Patients had right poster lesions
Neurological effects:
CAN-
Can match objects for size, colour, brightness
Can recognise objects from pictures of usual views
No impairment with object recognition in real life
Can explain the use of objects

CANNOT
Cannot recognise objects in unusual views

In order to perceive objects in unusual views, we need 3D perception, in order to mentally rotate the object

But despite this, they can still access the object recognition units via the recognition shortcut - for usual objects

Which is why they can correctly explain the use of the object as they can assess the semantic system and they can also name the objects

Question 24

Describe pateint AB

Answer 24

Who - Reported by Warrington
What happened - Cerebral atrophy ( loss of neurons and connections between them)
Neurological affects:
CAN-
Distinguish basic shapes
Identify colours, numbers, letters
Match pictures of different views ( usual and unusual)

CANNOT-
-Cannot Name common objects or famous faces - says they look familiar but cannot name them
-Makes subsitiuons when categorising objects eg a donkey is a horse, a dog is a cat - eg can’t explain what objects do
-Cannot identify meaningful sounds

So, the whole semantic part of the system has been iterrupted

So whether is a problem with the object recognition unit or accessing the semantic system, we dont know for sure

Question 25

What can this model help to explain in real life?

Answer 25

All of the observable behaviour that we have seen so far, can be explained by the model It can predict the behaviour patterns of two different groups of patients with visual agnosia

Question 26

What is visual agnosia?

Answer 26

**Visual agnosia** = Inability to recognise seen objects

Question 27

WHat are the two types of visual agnosia?

Answer 27

Two different groups: - Impaired perceptual stages ( **apperceptive agnosia**) eg Mr S - eg not matching simple shapes, profound impairment in shape perception - Impaired semantic stages ( **associative agnosia )-** eg patient AB- Intact perception, can perceive object but cannot name or identify it. Percept has no meaning

Question 28

Why dont the patients with right posterior lesions have visual agnosia?

Answer 28

These cannot create object-centred representations They therefore have a perceptual categorisation deficit - but dont actually have visual agnosia as can recognise obejcts