L3 Object Recognition

Question 1

Translational invariance

Answer 1

able to recognise objects when they are moved to a different location

Question 2

Size invariance

Answer 2

able to recognise objects when they have a different retinal image size

Question 3

Colour invariance

Answer 3

CAN STILL RECOGNISE OBJECTS WHEN ANOTHER COLOUR IS LAYERED OVER THE TOP OF THEM

Question 4

Rotation invariance

Answer 4

recognise when objects are rotated

Question 5

occlusion

Answer 5

recognise when objects are occluded

Question 6

variability in visual scenes

Answer 6

recognise objects when they are surrounded by or occluded by other objects

Question 7

intraclass variation

Answer 7

recognise them as belonging to a category despite variance from template (chairs all differ)

Question 8

viewpoint variation

Answer 8

we recognise things from all viewpoints

Question 9

template hypothesis:

Answer 9

2D pattern matching - we store tmeplates of everything we’ve seen which we call on for recognition

1. many many templates for each object
2. one template with a degree of flexibility (barcodes)

Question 10

template hypothesis problems

Answer 10

cant account for degree of flexibility - we’d need objects to be the same size and orientation

Question 11

prototype hypothesis

Answer 11

2D pattern matching – came from the template hypothesis. –we combine all the images of something we’ve seen to form a prototype/prototypical characteristics of an object.

Question 12

prototype evidence

Answer 12

when shows a load of images Ps are sure they’ve seen the prototype before

Question 13

Feature theory

Answer 13

patterns consists of attributes (A=2 lines and a bar) but was too simplistic

Question 14

Structural description theory

Answer 14

2D pattern matching –evolution of the feature theory. – The features and a description of how they are structurally related (T = horizontal and vertical line. Vertical supports and bisects horizontal)

Question 15

Marrs and Nishihara

Answer 15

3D object recognition – objects are all composed of a hierarchy of cylinders (that can vary in size and shape) – the cylinder position is always the same relative to its axis –works much better for biological objects

Question 16

Biedermans Geon hypothesis

Answer 16

(invariant/viewpoint independent)
36 different geons which can be used to describe any shape. consider their relative size, verticality, centring and size of surface at joints. to determine their form and the non accidental properties to map their 3D shape

Question 17

non accidental properties

Answer 17

geon hypothesis – the 2D rep of the NAs will translate directly into the 3D rep to give an idea of how the object looks in space. They include the curvature, parallel, cotermination, symettry and colinearity (points on a straight line) of features.

Question 18

support for geon

Answer 18

deleting parts of the objects where geons meet makes them harder to recognise than deleting curved regions

Question 19

Biederman geon good points:

Answer 19

flexible parsimonious and recognises importance of part recognition

Question 20

Biedermans geon bad points

Answer 20

• why 36?
• de emphasise importance of context
• simplifies contribution of viewpoint dependence
• structural info not always enough (peach vs. nectarine)
• within category discrimination

Question 21

viewpoint dependent theories

Answer 21

these suggest we have stored templates for each view point of the object — they were designed to coexist with independent theories. genereally beleived we use ID for between category and D for within

Question 22

issues to overcome in object recognition

Answer 22

• binding problem – how do other modalities help us to recognise
• when presented with many objects how do we figure out which geon belongs where

Question 23

Beyond recognition original:

Answer 23

structural description (round) –> semantic (crunchy) –> name representation (apple) –> name

Question 24

cascade model:

Answer 24

there is cross talk between each stage of recognition explaining why the similarity leads to higher confusability – it is supported by anecdotal and empirical evidence

Question 25

Apperceptive agnosia

Answer 25

–LATERAL -early process problem –can recognise COMPONENTS of the object but cannot put them together in a meaningful representation of the object –Bad at segmentation, degraded image tasks and have difficulty recognising objects

Question 26

Associative agnosia

Answer 26

-Bi lateral –late process –can structurally describe object/visually process it fully but cannot retrieve semantic representation so cannot name it. perform well on segmentation and copying tasks. no perceptual problems.

Question 27

category specific agnosia

Answer 27

bilateral temporal lobe damage – have issues with animate objects only

Question 28

why category specific agnosia?

Answer 28

1. different ways of recognising (functional vs. sensory)
2. kinaesthetic representations evoked by inanimate objects
3. living things are more similar than non living things (lion line drawings, 20ms)

Question 29

AGNOSICS

Answer 29

fail to feel familiarity with the object