Principal component analysis of good continuation cues

1
Principal component analysis of good continuation
cues

• Aaron Clarke

2
Background
3
Background
4
Brunswik and Kamiya (1952)
Proximity
Similarity
Symmetry
Good Continuation
Closure
5
Elder and Goldberg (1998/2002)
6
Orientation
7
Blur Scale
8
Contrast
9
Elder and Goldberg
10
Elder and Goldberg

• Good Continuation
• Parallelism
• Co-circularity

11
Elder and Goldberg

• Good Continuation
• Parallelism
• Co-circularity

12
Parallelism
13
Co-circularity
14
Good Continuation
15
Parallelism and Cocircularity
16
Discretization
17
Enter Aaron
18
Sub-Pixel Localized Edgels
Pixel Localized Edgels Sub-Pixel Localized Edgels
19
The Task
20
The Task
21
The Traced Elephant
22
Individual Edgels
23
Nearest Neighbours
24
8 Edgel Separation
25
64 Edgel Separation
26
Inferential Power
Second Principal Component
First Principal Component
27
Discussion

• The inferential power of the good continuation
  cues is maximal for neighbouring edgels, falling
  steadily as edgel separation increases.
• While for neighbouring edgels the parallelism cue
  is stronger than the cocircularity cue, this
  reverses for edgels separated by 4 edgels or
  more, suggesting that estimation noise limits the
  utility of the cocircularity cue at small
  separations.

28
(No Transcript)
29
Elder and Goldberg
Luminance
Position
30
Elder and Goldberg
Proximity Good Continuation Parallelism
Co-circularity Similarity Brightness
Contrast
31
Elder and Goldberg
Proximity Good Continuation Parallelism
Co-circularity Similarity Brightness
Contrast
32
Principal Components Analysis
33
The Gestalt Psychologists
34
Brunswik and Kamiya (1952)
35
Resulting Distributions
36
Principal Components Analysis
37
Rotation Angles
Parallelism Stronger
Cocircularity Stronger
38
Distribution Parameters