Segmentation and Perceptual Grouping. The image of this cube contradicts the optical image.
-
date post
21-Dec-2015 -
Category
Documents
-
view
216 -
download
1
Transcript of Segmentation and Perceptual Grouping. The image of this cube contradicts the optical image.
Segmentation and Perceptual Grouping
The image of this cube contradicts the optical image
Stochastic Completion Fields
Stochastic Completion Fields
Stochastic Completion Fields
Hough Transform
Hough Transform
Shortest Path
Curve Evolution
Curve Evolution
Thresholding
Histogram
0 50 100 150 200 250
0
200
400
600
800
1000
1200
Thresholding
Thresholding
125
15699
Local Uncertainty
Global Certainty
Local Uncertainty
Global Certainty
Minimum Cut
Texture Examples
Coarse Measurements for Texture
Hilbert Transform
-100 -50 0 50 100-0.01
-0.005
0
0.005
0.01
-100 -50 0 50 1000
0.004
0.008
0.012
0.016
Gaussian Hilbert Transform
1 ( )[ ( )]
f x dxH f x
x y
Filter Bank
Textons
image textons
textonassignment
Normalized Cuts
The Pixel Graph
ij w
Couplings
reflect intensity similarity
{}ij w
Hierarchical Graph
Normalized-Cut Measure
Si
Siui 0
1
Normalized-Cut Measure2( ) ( )ij i j
i j
E S w u u
Si
Siui 0
1
( ) ij i jN S w u u
( )( )
( )
E SS
N S
Minimize:
Recursive Coarsening
iuju
Recursive Coarsening
iuju
lU
kU
Representative subset
1 2( , ,..., )NU U U U
Recursive Coarsening
iuju
2
11
2
.
..
Nn
u
U
P
u
Uu
U
For a salient segment :
P ( )n N ,sparse interpolation matrix
kU
lU
Weighted Aggregation
ijwi
jjlp
aggregate k aggregate l
ijk ik jli j
l p pwW
klWikp
Segment Detection
A Chicken and Egg Problem
Problem:Coarse measurements mix neighboring statistics
Solution: support of measurements is determined as the segmentation process proceed
Adaptive vs Rigid Support
AveragingSWA Geometric
Adaptive vs Rigid Support
InterpolationSWA Geometric
AggregateShape
Boundary Integrity
Sharpen the Aggregates
Top-down Sharpening:• Expand core• Sharpen boundaries
Experiments
• Our SWA algorithm (CVPR’00 + CVPR’01) run-time: 5-10 seconds.
• Normalized cuts (Shi and Malik, PAMI’00; Malik et al., IJCV’01) run-time: about 10-15 minutes.
Software courtesy of Doron Tal, UC Berkeley.
images on a pentium III 1000MHz PC:200 200
Use of Multiscale Variance Vector
Use of Multiscale Variance Vector
Variance: Avoid Mixing
aggregation Moving window
Texture Aggregation
Fine (homogeneous) Coarse (heterogeneous)
Squirrel
Leopard
Texture Composition
WithVariance only
With all measures
Tiger
Butterfly
Bird
Owl
Lion Cub
Polar Bear
Segmentation with Ncuts(Malik et al).
Complexity
• Every level contains about half the nodes of the previous level:
• Total #nodes double #pixels
• All connections are local, cleaning small weights
• Top-down sharpening: constant number of levels
• Linear complexity
• Implementation: 5 seconds for 400x400