Color wavelet covariance(CWC)

pucefakeΤεχνίτη Νοημοσύνη και Ρομποτική

30 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

190 εμφανίσεις

Color wavelet covariance(CWC)

Texture feature

2013
-
06
-
04

Jo
Yeong
-
Jun

Karkanis
, Stavros A., et al.

"
Computer
-
aided tumor detection in endoscopic video using color
wavelet features
."


Information
Technology in Biomedicine, IEEE Transactions on

7.3 (2003): 141
-
152.


Introduction



Haralick

texture features



Color wavelet covariance(CWC) texture feature


DWT
transformation


Statistical measures (from
haralick
)


Covariance between color channels



Experimental Results

C
ontent

2

3

Introduction


What is texture?


Texture


사전적

정의


질감
(
質感
) :
재질의

차이에서

받는

느낌
.



Image texture


Pixel intensities


변화에

따른

반복적

패턴
.


이미지의



배열
, intensities
등의

정보를

나타냄
.








공개

Database


Brodatz

Vistex



Textures
Textures









4

Introduction

Computer
-
aided tumor detection in endoscopic video using color wavelet features

Artificial texture

Natural texture

-
Color wavelet covariance(CWC)
feature


텍스처

분석

feature.


-
2003


CWC


대장

용종

검출에

사용된


,

용종

검출

뿐만

아니라

의료

영상

분석에

많이

사용됨
.


-
1973
년에

제안된

haralick

texture feature[1
]


distinctiveness


있는



개의

measure


이용
.


-
Discrete wavelet transform(DWT)


통해

texture




표현하는

주파수

대역을

이용

[2]
.


-
최종적으로

Color space(RGB)
간의

관계를

feature


정의



[3
].

Introduction

5

Computer
-
aided tumor detection in endoscopic video using color wavelet features

[1]
Haralick
, Robert M.,
Karthikeyan

Shanmugam
, and Its'
Hak

Dinstein
.


"Textural features for image

classification."

Systems, Man and Cybernetics, IEEE Transactions on

6 (1973): 610
-
621.

[2]
Julesz
,
Bela
. "
Texton

gradients: The
texton

theory revisited."

Biological Cybernetics

54.4
-
5 (1986): 245
-
251
.

[3]
Van de
Wouwer
,
Gert
, et al. "Wavelet correlation signatures for color texture characterization."

Pattern

recognition

32.3 (1999): 443
-
451.



Color wavelet covariance feature(CWC)

6

Haralick

Texture features


Assumption


이미지의



픽셀은

주변

픽셀과

공간적


관계가

있음
.



Gray level co
-
occurrence matrix(GLCM)


픽셀간의

공간적

관계를

나타내는

행렬

Haralick

texture features

7

Computer
-
aided tumor detection in endoscopic video using color wavelet features

0˚ 45˚ 90˚ 135˚

0

1

0

2

0

2

1

1

3

1

0

0

0

0

2

3

2

1

3

0

2

1

0

0

0

1

0

1

0

1

0

0

4 gray
-
level image

0 1 2 3

0


1


2


3

1

1

1

1

1

1

1

2

1

0.16

0.08

0.25

0

0.16

0.08

0

0

0

0.08

0

0.08

0

0.08

0

0

GLCM

Normalized GLCM


Extract features from GLCM


제안된

14
가지의

statistical features


GLCM
에서

계산
.


각각의

feature


각기

다른

통계적

성질을

분석하는

measure.

Haralick

texture features

8

Computer
-
aided tumor detection in endoscopic video using color wavelet features

0.16

0.08

0.25

0

0.16

0.08

0

0

0

0.08

0

0.08

0

0.08

0

0

GLCM

Feature

Extraction

0
.
018
3
.
048
0
.
807
1
.
152
0
.
632
1
.
231
0
.
125
0
.
002
5
.
123
0
.
721
2
.
112
0
.
521
0
.
892
12
.
52

Haralick

feature

(14 dimension)


Ex :
두가지

이미지에서

뽑힌

haralick

features

Haralick

texture features

9

Computer
-
aided tumor detection in endoscopic video using color wavelet features

Grassland

Ocean

ASM

Contrast

Correlation

.0128

3.048

.8075

.0080

4.011

.6366

.0077

4.014

.5987

.0064

4.709

.4610

.0087

3.945

.6259

Angle

0

45

90

135

Avg

ASM

Contrast

Correlation

.1016

2.153

.7254

.0771

3.057

.4768

.0762

3.113

.4646

.0741

3.129

.4650

.0822

2.863

.5327

10

Color wavelet covariance

Texture features


Haralick

features


14
가지

feature


가장

분별력





4
가지

feature
조합을

이용
.



Angular second moment


Correlation


Inverse difference moment


Entropy



14
가지



서로

correlation


작은

것들을

선택함
.


Correlation


크면

dependent
하기

때문에

쓸데

없는

정보일



있음
.




Color wavelet covariance(CWC) texture
feature

11

Computer
-
aided tumor detection in endoscopic video using color wavelet features

14
haralick

statistics measures


Discrete wavelet transform(DWT
)

이용

Color wavelet covariance(CWC) texture
feature

12

Computer
-
aided tumor detection in endoscopic video using color wavelet features

H(z)

L(z)





Row
방향

H(z)



H(z)



L(z)



L(z)



HH

HL

LH

LL

Column
방향

LL

LH

HL

HH

3 level DWT

-
이미지

압축

기법인

JPEG2000
에서



.


-
각각의

변환

이미지들은



대역폭에서


밝기

변화를

묘사
.


-
이미지의

패턴분석에

사용
.


3 level DWT




texture


가장



표현하는

level


이용
.



Julesz
,
Bela
.[2]


따르면
,

DWT


Second
-
order
정보


가장

Texture




나타낸다고

알려짐
.


Color wavelet covariance(CWC) texture
feature

13

Computer
-
aided tumor detection in endoscopic video using color wavelet features

[2]
Julesz
,
Bela
. "
Texton

gradients: The
texton

theory revisited."

Biological Cybernetics

54.4
-
5 (1986): 245
-
251.

3 level DWT

R
(
i
=1)

g
(
i
=2)

B
(
i
=3)

(3
-

channel)




,



=
1
,
2
,
3
,


=
4
,
5
,
6

14

R
(
i
=1)

g
(
i
=2)

B
(
i
=3)

(3
-

channel)

Color wavelet covariance(CWC) texture
feature

Computer
-
aided tumor detection in endoscopic video using color wavelet features




,



=
1
,
2
,
3
,


=
4
,
5
,
6


𝑎



,



=
1
,
2
,
3
,

=
4
,
5
,
6

,
𝛼
=
0
°
,
45
°
,
90
°
,
135
°

𝛼
=

Co
-
occurrence matrix

1. Angular
second
moment

2. Correlation

3. Inverse
difference
moment

4. Entropy

𝐹


𝑎



,



=
1
,
2
,
3
,

=
4
,
5
,
6

,



𝛼
=
0
°
,
45
°
,
90
°
,
135
°
,


=
1
,
2
,
3
,
4


=

4
Haralick

features

15

Computer
-
aided tumor detection in endoscopic video using color wavelet features

Color wavelet covariance(CWC) texture
feature

𝐹


𝑎



,



=
1
,
2
,
3
,

=
4
,
5
,
6

,



𝛼
=
0
°
,
45
°
,
90
°
,
135
°
,


=
1
,
2
,
3
,
4

[3] Van de
Wouwer
,
Gert
, et al. "Wavelet correlation signatures for color texture characterization."

Pattern recognition

32.3 (1999): 443
-
451.


마지막으로

color space


간의

covariance


고려
.



Van de
Wouwer
,
Gert
, et al
.[3]


따르면
,

이미지의



color space
들은

서로

밀접한

관계를

가진다
.

이러한

경향성을

이용하면

효과적으로

Texture


정의





있다
.

𝑊


(

,

)
=
𝑜𝑣
𝐹


𝑎



,
𝐹


𝑎






과정

정리

16

Color wavelet covariance(CWC) texture
feature

Computer
-
aided tumor detection in endoscopic video using color wavelet features

Input image

R
(
i
=1)

g
(
i
=2)

B
(
i
=3)

level DWT (3
-

channel)




,



=
1
,
2
,
3
,


=
4
,
5
,
6

Co
-
occurrence matrices


𝑎



,



=
1
,
2
,
3
,

=
4
,
5
,
6

,



𝛼
=
0
°
,
45
°
,
90
°
,
135
°

Statistical feature extraction

𝐹


𝑎



,



=
1
,
2
,
3
,

=
4
,
5
,
6

,



𝛼
=
0
°
,
45°
,
90°
,
135°
,



=
1
,
2
,
3
,
4

1. ASM

2.
Correlation

3. IDM

4. Entropy

𝑊



,

=

𝑜𝑣
𝐹


𝑎



,
𝐹


𝑎







채널에서

추출된

Feature


Covariance
계산

17

Experimental
results

Computer
-
aided tumor detection in endoscopic video using color wavelet features

18

Experimental results

Computer
-
aided tumor detection in endoscopic video using color wavelet features

Specificity = True negative/
전체

Negative

Sensitivity =
True
positive /
전체

Positive

19

Experimental results

Computer
-
aided tumor detection in endoscopic video using color wavelet features

20

Experimental results

[1]
Iakovidis
, D. K., et al. "A comparative study of texture features for the discrimination of gastric
polyps in endoscopic video." Computer
-
Based Medical Systems, 2005. Proceedings. 18th IEEE
Symposium on. IEEE, 2005
.


[2]

Alexandre
,
Luís

A.,
Nuno

Nobre
, and
João

Casteleiro
. "Color and position versus texture features for
endoscopic polyp detection."

BioMedical

Engineering and Informatics, 2008. BMEI 2008. International
Conference on
. Vol. 2. IEEE, 2008.

Thank you