5. Theories of Vision

Question 1

What are the three levels of Marr’s theory

Answer 1

Theory, algorithm and implementation.

Question 2

Describe Marr’s computational level

Answer 2

Sets out the goals and process of how a task can be conducted.

Question 3

Describe Marr’s algorithm level

Answer 3

The processes it uses to carry out the task

Question 4

Describe Marr’s implementation level

Answer 4

The physical method used to carry out the algorithm in the case of vision this is the neural structure and activity.

Question 5

What are Marr’s three levels of representations?

Answer 5

Primal sketch, 2 and 1/2 D sketch, 3D sketch

Question 6

What is a primal sketch and what are the four primatives?

Answer 6

A map of the important parts of an image.

• Blobs: Areas bounded by a change in intensity
• Edge-segments: Sudden changes in intensity
• Bar: Two parallel edge segments
• Termination: Sudden discontinuity.

Question 7

What is the difference between a raw and full primal sketch

Answer 7

Raw primal sketch contains just important parts of the image whilst a full primal sketch has place tokens where things are grouped together and there is information on the location, shape and texture.

Question 8

What is the difference between a raw and full primal sketch

Answer 8

Raw primal sketch contains just important parts of the image whilst a full primal sketch has place tokens where things are grouped together and there is information on the location, shape and texture.

Question 9

What is the difference between 2 1/2 D sketch and a primal sketch?

Answer 9

The 2 1/2 D sketch has information about surface slant and distance at each point for depth relations, distance from object and observer and how far an object extends.

Question 10

How is the 2 1/2 D composed (what cues are used)

Answer 10

• Motion cues
• Texture cues
• Disparity in retinal images from two eyes

Question 11

What is the difference between 2 1/2 D and 3D sketches?

Answer 11

A 2 1/2 D sketch is viewer centred as it contains no information not present in the retinal field. Whereas the 3D sketch is more abstract from the image and is object centred (independent of eye movements).

Question 12

Describe the 3D sketch

Answer 12

Simple primatives are grouped together in a hierarchy.

Question 13

Describe Gibson’s visual theory?

Answer 13

Emphasised vision’s role in carrying out actions rather than forming representations. Task-specific heuristics are used to interpret the images to allow an action to be carried out.

Question 14

What is the texture gradient heuristic?

Answer 14

The density of elements increases with distance and the width and height can reveal shape and orientation of a surface.

Question 15

What is the horizontal ratio relation?

Answer 15

Compare the height of things by seeing if they align on the horizon.

Question 16

What is the ambient optical array and what does outflow and inflow represent?

Answer 16

The optic array (convergence of light) changes as you move.

In-flow specifies retreat and outflow specifies approach.

Question 17

What is the equation for calculating the time until contact (tau)

Answer 17

Tau = initial retinal size / rate of expansion

Question 18

What is the Bayesian inference?

Answer 18

We compare the likelihood of different interpretations using the image and our prior knowledge. When there is poor sensory information we go with the prior interpretation.