Normalized Cuts and Image Segmentation - Final Project | GEOG 8, Study Guides, Projects, Research of Geography

Material Type: Project; Class: GLOBAL WARMING; Subject: Geography; University: University of California - Santa Barbara; Term: Unknown 2000;

Typology: Study Guides, Projects, Research

Pre 2010

Uploaded on 09/17/2009

koofers-user-yxb-1
koofers-user-yxb-1 🇺🇸

4

(1)

10 documents

1 / 12

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
ECE 278 Final project
Normalized Cuts
Normalized Cuts
and Image Segmentation
and Image Segmentation
Baris Sumengen
Jelena Tešic
Vision Research Lab
Electrical and Computer engineering
pf3
pf4
pf5
pf8
pf9
pfa

Partial preview of the text

Download Normalized Cuts and Image Segmentation - Final Project | GEOG 8 and more Study Guides, Projects, Research Geography in PDF only on Docsity!

ECE 278 Final project

Normalized Cuts Normalized Cuts

andand Image SegmentationImage Segmentation

Baris Sumengen

Jelena Tešic

Vision Research Lab

Electrical and Computer engineering

June 8, 2000

Outline

n Introduction

n Graph Partitioning

n Grouping Algorithm

n Experiments

n Results

n Related Approaches

n Conclusion

June 8, 2000

Graph Partitioning

n Dissimilarity measure between two sets of

graph G=(V, E),

n Minimum cut Optimal bi-partitioning

n Clustering method (Wu, Leathy, PAMI

  1. that favors outliers

n Different cost function proposed:

,

u A v B

cut A B w u v

∈ ∈

= (^) ∑

A ∪ B = V A , ∩ B = ∅

cut A B cut A B

Ncut A B

asso A V asso B V

,

( , ) ( , )

u A t V

asso A V w u t

∈ ∈

= (^) ∑

June 8, 2000

Optimal Partitioning

i (^) j

d = ∑ W i j

n N – number of nodes in the graph V

n W – NxN symmetrical matrix with elements W(i,j)

n D:

n x – indication vector:

n Final result:

with the condition:

D = d i a g ( d 1 , d 2 ..., dN )

xi = { +1, V i ( ) ∈ A −1, V i ( ) ∉ A }

min ( ) min

T

x y T

y D W y

Ncut x

y Dy

i
i

x^ i

i

x i

d

y

d

T

N

d

y

d

M

June 8, 2000

The Grouping Algorithm

n Use spatial proximity term and feature similarity terms to

calculate weighted coefficients W(i,j)

n Summarize information in W and D

n Find an approximate solution for the second smallest

eigenvalue of the system

n Bipartition the graph by finding the best splitting point

n Check the stability of the cut and decide on the current

partition status

n Stop the recursion if Ncut exceeds certain limit

( DW y ) = λDy

June 8, 2000

Experiments

n Different parameter values and feature desriptors

2

( , ) ,

i j i j
I X

F F X X

i j

W i j e e X X r

σ σ

− − − −

= ∗ − <

1

( ), intensity

( ) , sin( ), cos( ),color

* ,..., * , texture n

I i

F i v v s h v s h i

I f I f i

 = 

g g g g

June 8, 2000

5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20 5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20 5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20

5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20

5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20 5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20

5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20 5 10 15 20 25 30

2 4 6 8 10 12 14 16 18 20

June 8, 2000

5 10 15 20 25 30

2

4

6

8

10

12

14

16

5 10 15 20 25 30

2

4

6

8

10

12

14

16