Prediction-Based Prediction-Based Reversible Data Hiding Reversible Data Hiding Source: Information Sciences, vol. 179, no. 14, pp. 2460-2469, 2009. Authors: Hsien-Wen Tseng and Chi-Pin Hsieh Speaker: Chia-Chun Wu ( 吳吳吳 ) Data: 2010/01/08 1
Jan 03, 2016
Prediction-Based Prediction-Based Reversible Data HidingReversible Data Hiding
Source: Information Sciences, vol. 179, no. 14, pp. 2460-2469, 2009.Authors: Hsien-Wen Tseng and Chi-Pin HsiehSpeaker: Chia-Chun Wu (吳佳駿 )Data: 2010/01/08
1
OutlineOutlineIntroductionThe Proposed SchemeExperimental ResultsConclusions
2
IntroductionIntroductionReversible data hiding
3
Cover image
Secret data: 0101…
Public channel(Ex: Internet)
Stego-image
SenderSender InterceptorInterceptor
ReceiverReceiver
Secret data: 0101…
Cover image
The Proposed Scheme The Proposed Scheme (1/5)(1/5)
Embedding Algorithm◦ Side-match
prediction
2
)1,(),1(),(ˆ
yxpyxpyxp
),(),(ˆ yxpyxpd
4
24 36ˆ (2,2) 30
2
30 33 3
p
d
3336
2431(2,2) 33p
Predicted value
Pixel value
Prediction error
Stego pixel value
The Proposed Scheme The Proposed Scheme (2/5)(2/5)
Embedding Algorithm◦ Predefined threshold T
2/:3
2/:2
2/:1
TTdCase
TTdTCase
TdTCase
2/T T
Embedding secret bits
Pixel shifting
2/TT
Case 1 Case 2 Case 3
T2
TdCase
TdTCase
TdTCase
2':3
'2/:2
2':1
Extracting secret bits & recovery
Pixel shifting5
The Proposed Scheme The Proposed Scheme (3/5)(3/5)
Example of embedding procedure (Case 1)
5
ˆ( , ) 33, ( , ) 30
30 33 3, 1
T
p x y p x y
d W
32/ T 5T / 2 7T T 102 T
[3,4] [5,6] [7,255]
d=3
ˆ ˆ( , ) 2 , if ( , ) ( , )'( , )
ˆ ˆ( , ) 2 , if ( , ) ( , )
p x y d W p x y p x yp x y
p x y d W p x y p x y
'( , ) 30 7 37
' 2 7
p x y
d d W
d'=7
6
33362431
The Proposed Scheme The Proposed Scheme (4/5)(4/5)
Example of shifting procedure (Case 2)
5
ˆ( , ) 38, ( , ) 32
32 38 6
T
p x y p x y
d
32/ T 5T / 2 7T T 102 T
[3,4] [5,6]
d=6
ˆ( , ) / 2 , if ( , ) ( , )'( , )
ˆ( , ) / 2 , if ( , ) ( , )
p x y T p x y p x yp x y
p x y T p x y p x y
'( , ) 38 2 36
' / 2 4
p x y
d d T
d'=4
7
[7,255]
38362831
The Proposed Scheme The Proposed Scheme (5/5)(5/5)
Example of shifting procedure (Case 3)
5
ˆ( , ) 38, ( , ) 30
30 38 8
T
p x y p x y
d
32/ T 5T / 2 7T T 102 T
[3,4] [5,6]
d=8
ˆ( , ) / 2 , if ( , ) ( , )'( , )
ˆ( , ) / 2 , if ( , ) ( , )
p x y T p x y p x yp x y
p x y T p x y p x y
'( , ) 38 3 41
' / 2 11
p x y
d d T
d'=11
8
[7,255]
38362431
Overflow and UnderflowOverflow and Underflow
91 1001007
K=1 0+1=1 7-1=6
The Experimental Results The Experimental Results (1/5)(1/5)
Three test images
10
1. Side Match
2. Horizontal Predictor
3. Vertical Predictor
4. Causal Weighted Average
5. Causal SVF
The Experimental Results The Experimental Results (2/5)(2/5)
11
ˆ1. Horizontal Predictor: ( , ) ( 1, )
ˆ2. Vertical Predictor: ( , ) ( , 1)
3. Causal Weighted Average:
ˆ ( , ) [2 ( 1, ) 2 ( , 1) ( 1, 1) ( 1, 1)] 6
4. Causal SVF:
( , ) [ ( 1, ) ( ,
p x y p x y
p x y p x y
p x y p x y p x y p x y p x y
Avg x y p x y p x y
1
11
1
1) ( 1, 1) ( 1, 1)] 4
( 1, ) ( 1, ) ( , 1) ( , 1)ˆ ( , )
( 1, ) ( , 1)
( , ) exp( ( ( , ) ( , ) ),
where is the controlling factor.
n
n
p x y p x y
p x y n w x y n p x y w x yp x y
w x y n w x y
w x i y j abs p x i y j Avg x y f
f
The Experimental Results The Experimental Results (3/5)(3/5)
12
Table 6. Experimental results of predictors.
The Experimental Results The Experimental Results (4/5)(4/5)T=2
13
[17] S. K. Yip, O. C. Au, H. M. Wong, C. W. Ho, “Generalized lossless data hidingby multiple predictors,” ISCAS 2006, pp. 21-24, May 2006.
Table 7. Comparisons between the proposed method and Yip et al.’s scheme.
The Experimental The Experimental Results Results (5/5)(5/5)Compare the proposed scheme with that of
Tian [14] and Celik [2] using grayscale image Lena◦ The propsoed scheme
T=2 Capacity 0.220 bpp PSNR 47.31dB
◦ Tian’s scheme Capacity 0.151 bpp PSNR 44.20dB
◦ Celik et al’s scheme Capacity 0.180 bpp PSNR 40.5dB
14
ConclusionsConclusionsA high capacity and low distortion
reversible data hiding technique using prediction error expansion is proposed.
The correlation between the pixel and its neighboring pixels is considered to perform the difference expansion and the secret bit is embedded in difference.
15