Integration of RFID into Wireless Sensor Networks: Architectures ...

brainybootsΚινητά – Ασύρματες Τεχνολογίες

21 Νοε 2013 (πριν από 3 χρόνια και 27 μέρες)

48 εμφανίσεις

Content Based Color Image Retrieval vi
Wavelet Transformations

Information

Retrieval

Class

Presentation

May 2, 2012

Author: Mrs. Y.M. Latha

Presenter: Mahbubur Rahman

Advisor: Prof. Susan Gauch

Mobile Pervasive and Sensor Systems Laboratory

Table of Contents


Introduction


Target Environment


Proposed CBIR


Wavelet Transform


Feature Extraction


Similarity Criteria


Progressive Retrieval Strategy


Experiment Result


Conclusion

1

Mobile Pervasive and Sensor Systems Laboratory

Introduction


Content Based Image Retrieval


Database is huge


Retrieved the desired image from the database

2

Mobile Pervasive and Sensor Systems Laboratory

Introduction


Content Based Image Retrieval


Images have specific features
-
horizontal or vertical lines


Image features are compared to find similar images

3

Query image

Database Image

Feature extract to compare

Mobile Pervasive and Sensor Systems Laboratory

Target Environment


Color Image Retrieval


Based on Object Visual contents of image


Color, Texture and Shape


Multimedia image with audio, text and video are not covered

4

Mobile Pervasive and Sensor Systems Laboratory

Proposed CBIR


Wavelet Based CBIR


Indexing
-
wavelet decomposition then F
-
norm


Searching
-
wavelet decomposition, F
-
norm then similarity matching

5

Searching

Process

Indexing

Process

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

6


Wavelet Transformation


Decompose using rescaling and keeping details of image

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

7


Haar Wavelet Transform


Find out N/2 wavelet values and N/2 coefficients from N data


Upper half is wavelet functions and lower half is coefficient values

N

N/2

N/2

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

8


Haar Wavelet Transform


Average and differentiate values to get wavelets function and coefficients

576

704

1152

1280

1344

1472

1536

1536

704

640

1156

1088

1344

1408

1536

1600

768

832

1216

1472

1472

1536

1600

1600

832

832

960

1344

1536

1600

1536

1536

832

832

960

1216

1536

1600

1536

1536

960

896

896

1088

1600

1600

1600

1536

768

768

832

832

1280

1472

1600

1600

448

768

704

640

1280

1408

1600

1600

640

1216

1408

1536

-
64

-
128

-
128

0

672

1122

1376

1568

32

68

-
64

-
64

800

1344

1504

1600

-
32

-
256

-
64

0

832

1152

1568

1536

0

-
384

-
64

0

832

1088

1568

1536

0

-
256

-
64

0

928

992

1600

1568

32

-
192

0

64

768

832

1376

1600

0

0

-
192

0

608

672

1344

1600

-
160

64

-
128

0

First half is the average

of each pair

second half is the

Difference of each pair

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

9


Haar Wavelet Transform


Average and differentiate values to get wavelets function and coefficients

640

1216

1408

1536

-
64

-
128

-
128

0

672

1122

1376

1568

32

68

-
64

-
64

800

1344

1504

1600

-
32

-
256

-
64

0

832

1152

1568

1536

0

-
384

-
64

0

832

1088

1568

1536

0

-
256

-
64

0

928

992

1600

1568

32

-
192

0

64

768

832

1376

1600

0

0

-
192

0

608

672

1344

1600

-
160

64

-
128

0

First half is the average

of each pair

second half is the

Difference of each pair

656

1169

1392

1552

-
16

-
30

-
96

-
32

816

1248

1536

1568

-
16

-
320

-
64

0

880

1040

1584

1552

16

-
224

-
32

32

688

752

1360

1600

-
80

32

-
160

0

-
16

47

16

-
16

-
48

-
98

-
32

32

-
16

96

-
32

32

-
16

64

0

0

-
48

48

-
16

-
16

-
16

-
32

-
32

-
32

80

80

16

0

80

-
32

-
32

0

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

10


Haar Wavelet Transform


First level decomposition

576

704

1152

1280

1344

1472

1536

1536

704

640

1156

1088

1344

1408

1536

1600

768

832

1216

1472

1472

1536

1600

1600

832

832

960

1344

1536

1600

1536

1536

832

832

960

1216

1536

1600

1536

1536

960

896

896

1088

1600

1600

1600

1536

768

768

832

832

1280

1472

1600

1600

448

768

704

640

1280

1408

1600

1600

656

1169

1392

1552

-
16

-
30

-
96

-
32

816

1248

1536

1568

-
16

-
320

-
64

0

880

1040

1584

1552

16

-
224

-
32

32

688

752

1360

1600

-
80

32

-
160

0

-
16

47

16

-
16

-
48

-
98

-
32

32

-
16

96

-
32

32

-
16

64

0

0

-
48

48

-
16

-
16

-
16

-
32

-
32

-
32

80

80

16

0

80

-
32

-
32

0

LL

LH

HL

HH

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

11


Haar Wavelet Transform


Haar matrix can do these steps in one operation

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

12


D4 Wavelet Transform


Use scaling function


Upper half scaling coefficients and lower half wavelets coefficients

Mobile Pervasive and Sensor Systems Laboratory

Wavelet Transform

13


D4 Wavelet Transform


D4 use four scaling function to transform image

Scaling functions

Wavelet functions

Mobile Pervasive and Sensor Systems Laboratory

Features Extraction

14


Feature Vector


F
-
norm extract the image features from scaled image matrix

Mobile Pervasive and Sensor Systems Laboratory

Features Extraction

15


Feature Vector


F
-
norm extract the image features from scaled image matrix

576

704

1152

1280

1344

1472

1536

1536

704

640

1156

1088

1344

1408

1536

1600

768

832

1216

1472

1472

1536

1600

1600

832

832

960

1344

1536

1600

1536

1536

832

832

960

1216

1536

1600

1536

1536

960

896

896

1088

1600

1600

1600

1536

768

768

832

832

1280

1472

1600

1600

448

768

704

640

1280

1408

1600

1600

||A
0
||
F

||A
0
||
F
=0;


||A
1
||
F
=(576
2
+704
2
+704
2
+640
2
)
1/2


∆A
1
= ||A
1
||
F
-

||A
0
||
F
=1316.29


||A
1
||
F

||A
2
||
F

||A
7
||
F

||A
3
||
F

||A
4
||
F

||A
5
||
F

||A
6
||
F

Feature vector :



V
AF
={∆A
1,

∆A
2,

∆A
3,

∆A
4…….

∆A
n
)


Mobile Pervasive and Sensor Systems Laboratory

Similarity Criteria

16


Image matching criteria


Feature vector is calculate both for query image and indexed image


Extracts similarity criteria from feature vector

Similarity
α
i

of ∆A
i
and

∆B
i

Image A

Image B

Similarity
α
i

of full two images

Mobile Pervasive and Sensor Systems Laboratory

Progressive Retrieval

17


Rough Filtering from LL coefficient


Calculate Standard variances vectors


Query image as(
σ
r
q
,

σ
g
q
,

σ
b
q
) & database image as(
σ
r
d
,

σ
g
d
,

σ
b
d
)


Roughly filter out database image using


F=(
βσ
r
q

<
σ
r
q

<

σ
r
q
/
β
) && (
βσ
g
q

<
σ
g
q

<

σ
g
q
/
β
) && (
βσ
b
q

<
σ
b
q

<

σ
b
q
/
β
)
where
β

ε

(0,1)


If F is false then image is not any kind of similar


Progressive Rough Filtering


Filter considering the high frequency component with LH and HL coefficients


More precise filtering


LL coefficient best reflect the image feature


Apply similarity criteria to LL coefficient


If
α

exceeds certain threshold, discard as mismatch


Iteration


Iterate filtering process for all decomposition level to return precise image


Mobile Pervasive and Sensor Systems Laboratory

Experimental Result

18


Experiment Setup


D4 and Haar wavelet transform to decompose images


Maximal decomposition level =4


F
-
norm apply to extract image feature both for indexing and query image


Total 4 groups of images indexed, each containing 600 images


All images are preprocessed to be 256X256 sizes

Mobile Pervasive and Sensor Systems Laboratory

Experimental Result

19


Query result using Haar Wavelet


Relevant images retrieved using the similarity constants

Mobile Pervasive and Sensor Systems Laboratory

Experimental Result

20


Query result using D4 Wavelet


Relevant images retrieved using the similarity constants

Mobile Pervasive and Sensor Systems Laboratory

Experimental Result

21


Recall Rate Comparison


D4 wavelet recall rete is higher than the haar and existing wavelet histogram


Mobile Pervasive and Sensor Systems Laboratory

Experimental Result

22


Retrieval Speed Comparison


Both D4 and Haar are slower than existing histogram wavelet


Mobile Pervasive and Sensor Systems Laboratory

Conclusion

23


Proposed CBIR applied


Wavelet decomposition of images


F
-
norm to extract images features


Progressive retrieval to get the precise result




Proposed CBIR


Retrieve more accurate result than existing wavelet technique


D4 wavelet ensure greater speed with increase recall rate


Achieved high retrieval performance in real time CBIR systems


Mobile Pervasive and Sensor Systems Laboratory

24

Mobile Pervasive and Sensor Systems Laboratory

25