chinese university of hong kong department of information engineering a capacity estimate technique...
Post on 19-Dec-2015
215 views
TRANSCRIPT
Chinese University of Hong KongDepartment of Information Engineering
A Capacity Estimate Technique for JPEG-to-A Capacity Estimate Technique for JPEG-to-JPEG Image WatermarkingJPEG Image Watermarking
Peter Hon Wah WongPeter Hon Wah Wong
Department of Information EngineeringDepartment of Information Engineering
Chinese University of Hong KongChinese University of Hong Kong
Chinese University of Hong KongDepartment of Information Engineering
OutlineOutline
JPEG-to-JPEG (J2J) image watermarkingJPEG-to-JPEG (J2J) image watermarking
Background of capacity estimation for image Background of capacity estimation for image
watermarkingwatermarking
Proposed J2J capacity estimate techniqueProposed J2J capacity estimate technique
Necessary condition to achieve maximum capacityNecessary condition to achieve maximum capacity
Experimental resultsExperimental results
ConclusionsConclusions
Chinese University of Hong KongDepartment of Information Engineering
Background of capacity estimation for image Background of capacity estimation for image watermarkingwatermarking
There are some existing methods to estimate There are some existing methods to estimate the data hiding capacity of digital images the data hiding capacity of digital images
Ramkumar [1] et al. focused on comparing the Ramkumar [1] et al. focused on comparing the capacity among different transformscapacity among different transforms
Moulin et al. [2] estimated the capacity under Moulin et al. [2] estimated the capacity under different kind and degree of attacksdifferent kind and degree of attacks
Voloshynovskiy et al. [3] considered security Voloshynovskiy et al. [3] considered security issue of watermark to estimate the capacityissue of watermark to estimate the capacity
Chinese University of Hong KongDepartment of Information Engineering
Background of capacity estimation for image Background of capacity estimation for image watermarkingwatermarking
Shannon [4] showed Shannon [4] showed that for Gaussian that for Gaussian channelchannel
For watermarking, For watermarking, image is considered as image is considered as noisenoise
+X
N
Y
2
2
2 1log2
1Capacity
N
Y
+X
N
Y+
I
22
2
2 1log2
1Capacity
NI
Y
Chinese University of Hong KongDepartment of Information Engineering
Background of capacity estimation for image Background of capacity estimation for image watermarkingwatermarking
Costa [5] showed that if the distribution of the Costa [5] showed that if the distribution of the image is only known at the watermark image is only known at the watermark embedderembedder
2
2
2 1log2
1Capacity
N
Y
Chinese University of Hong KongDepartment of Information Engineering
Most of the digital images are stored in JPEG format Most of the digital images are stored in JPEG format Both the input and output images need to be JPEG Both the input and output images need to be JPEG
compatiblecompatible It is called JPEG-to-JPEG (J2J) watermarkingIt is called JPEG-to-JPEG (J2J) watermarking All DCT coefficients need to be re-quantized after All DCT coefficients need to be re-quantized after
watermark insertionwatermark insertion
JPEG-to-JPEG (J2J) Image WatermarkingJPEG-to-JPEG (J2J) Image Watermarking
Chinese University of Hong KongDepartment of Information Engineering
JPEG-to-JPEG Watermarking (J2J)JPEG-to-JPEG Watermarking (J2J)
JPEG File
Quantizationtable Qo
WatermarkEmbedding
QuantizedDCT
coefficients Dq JPEGEncoder
Quantizationtable Qn
WatermarkedJPEG File
Chinese University of Hong KongDepartment of Information Engineering
Proposed J2J Capacity Estimate TechniqueProposed J2J Capacity Estimate Technique
Capacity of watermarking is defined as the maximum Capacity of watermarking is defined as the maximum number of bits that can be embedded in the image and number of bits that can be embedded in the image and invisible invisible
There are two assumptionsThere are two assumptions– the watermarked images will be JPEG-compressed using after the watermarked images will be JPEG-compressed using after
watermark insertionwatermark insertion– the dimensions of the images are not changed in the watermark the dimensions of the images are not changed in the watermark
embedding embedding Human Visual System (HVS) model is used to estimate Human Visual System (HVS) model is used to estimate
the Just Noticeable Difference (JND) of DCT coefficients the Just Noticeable Difference (JND) of DCT coefficients Does not assume any specific watermarking method and Does not assume any specific watermarking method and
embedding domain embedding domain
Chinese University of Hong KongDepartment of Information Engineering
J2J - Capacity EstimationJ2J - Capacity Estimation
Denote the Denote the ijijthth quantized DCT coefficient of the quantized DCT coefficient of the kkthth block block as as DDqq((ii,,jj,,kk))
The dequantized DCT coefficient The dequantized DCT coefficient DDoo((ii,,jj,,kk) is given by) is given by
The quantized watermarked DCT coefficient The quantized watermarked DCT coefficient DDnn((ii,,jj,,kk) ) should satisfiesshould satisfies
Guarantees the invisibility of the watermarkGuarantees the invisibility of the watermark
81 jikjiDjiQkjiD qoo ,,,,,,
kjiJkjiDkjiD oon ,,,,,,
Chinese University of Hong KongDepartment of Information Engineering
J2J - Capacity EstimationJ2J - Capacity Estimation
Only finite possible values of Only finite possible values of DDnn((ii,,jj,,kk) is allowed) is allowed
The no. of possible value isThe no. of possible value is
J2J data hiding capacity of the image is given J2J data hiding capacity of the image is given approximately by approximately by CCww
KK is the number of blocks in the image is the number of blocks in the image
K
k j iww kjiNC
1
8
1
8
12 ,,log
1
,
,,,,
,
,,,,,,
jiQ
kjiJkjiD
jiQ
kjiJkjiDkjiN
n
oo
n
oow
Chinese University of Hong KongDepartment of Information Engineering
J2J - Capacity EstimationJ2J - Capacity Estimation
),,( kjiDo
),,( kjiJo),,( kjiJo
),( jiQn
NNww((ii,,jj,,k) = 5k) = 5 Capacity = logCapacity = log22(5) bits(5) bits
Chinese University of Hong KongDepartment of Information Engineering
J2J - Capacity EstimationJ2J - Capacity Estimation
Agrees somewhat with the works of Costa Agrees somewhat with the works of Costa WW is assumed to be uniformly distributed is assumed to be uniformly distributed
power of the watermark ispower of the watermark is assuming the quantization noise is uniformly assuming the quantization noise is uniformly
distributed in distributed in the power of quantization noise is the power of quantization noise is
),,(),,,( kjiJkjiJ oo
2
w 3/,, 2kjiJ o
2/),(,2/),( jiQjiQ nn2
n 12/, 2jiQn
Chinese University of Hong KongDepartment of Information Engineering
J2J - Capacity EstimationJ2J - Capacity Estimation
2
2
2
2
2
2
2
2
22
2
2
2
22
2
1log2
1
12/,
3/,,1log
2
1
,
,,41log
2
1
,
,,4log
2
1
,
,,2log
2
1
,
,,2log
1,
,,,,
,
,,,,log,,
n
w
n
o
n
o
n
o
n
o
n
o
n
oo
n
oo
jiQ
kjiJ
jiQ
kjiJ
jiQ
kjiJ
jiQ
kjiJ
jiQ
kjiJ
jiQ
kjiJkjiD
jiQ
kjiJkjiDkjiC
Chinese University of Hong KongDepartment of Information Engineering
Necessary Conditions to Achieve CapacityNecessary Conditions to Achieve Capacity
Consider a DCT coefficientConsider a DCT coefficient The capacity of a communication channel with The capacity of a communication channel with
side information is given by side information is given by
where where UU is a finite alphabet auxiliary random is a finite alphabet auxiliary random variable, by setting variable, by setting UU = = DDnn
onW DUIDUIC ,,max
on
onnnonnW
DDH
DDHDHDHDDIDHC
|max
|max,max
Chinese University of Hong KongDepartment of Information Engineering
Necessary Conditions to Achieve Necessary Conditions to Achieve CapacityCapacity
The maximum of is achieved when is equally The maximum of is achieved when is equally probable in when is known probable in when is known
the corresponding capacity is the corresponding capacity is
nD
oooo JDJD ,oD
otherwise ,0
,1
oonoo
wn
JDDJDNDp
w
QNm
QNm
Qm
Qm
Qm
Qm Ndwwfdwwfdwwf
nw
nw
n
n
n
n
1)(...)()(
2/1
2/3
2/3
2/1
2/1
2/1
wW NC 2log
Chinese University of Hong KongDepartment of Information Engineering
Experimental ResultsExperimental Results
Two common test images 512x512 Two common test images 512x512 gray-scale images are estimated and gray-scale images are estimated and reportedreported
Only the luminance component is used Only the luminance component is used Scaling factor (SF) is used to scale the Scaling factor (SF) is used to scale the
default quantization table in JPEG to default quantization table in JPEG to control the compression ratiocontrol the compression ratio
Chinese University of Hong KongDepartment of Information Engineering
Experimental ResultsExperimental Results
Lena Pepper
Chinese University of Hong KongDepartment of Information Engineering
Experimental ResultsExperimental Results
Lena Pepper
Chinese University of Hong KongDepartment of Information Engineering
Experimental Results – Block CapacityExperimental Results – Block Capacity
Lena,SFbefore=1, SFafter=0.5,
Est. capacity=3507 bits
Pepper,SFbefore=1, SFafter=0.5,
Est. capacity=3663 bits
Chinese University of Hong KongDepartment of Information Engineering
Experimental Results - HistogramExperimental Results - Histogram
Lena Pepper
Chinese University of Hong KongDepartment of Information Engineering
ConclusionsConclusions
A method to estimate the watermarking A method to estimate the watermarking capacity of digital images in the JPEG-to-JPEG capacity of digital images in the JPEG-to-JPEG watermarking framework is proposed watermarking framework is proposed
The estimation does not assume any specific The estimation does not assume any specific watermarking method and embedding domain, watermarking method and embedding domain, it should apply to any watermarking method in it should apply to any watermarking method in the J2J frameworkthe J2J framework
More details can be founded from:More details can be founded from:P.H.W. Wong, Oscar C. Au, "P.H.W. Wong, Oscar C. Au, "A Capacity Estimation Technique for JPEG-To-JPEG Image WateA Capacity Estimation Technique for JPEG-To-JPEG Image Watermarkingrmarking," ," IEEE Transactions on Circuits and Systems for Video IEEE Transactions on Circuits and Systems for Video Technology: Special Issue on Authentication, Copyright Technology: Special Issue on Authentication, Copyright Protection and Information HidingProtection and Information Hiding, pp. 746-752, Aug. 2003. , pp. 746-752, Aug. 2003.
Chinese University of Hong KongDepartment of Information Engineering
ReferencesReferences
1.1. M. Ramkumar and A.N. Akansu, “Capacity Estimates for Data M. Ramkumar and A.N. Akansu, “Capacity Estimates for Data Hiding in Compressed Images,” Hiding in Compressed Images,” IEEE Trans. Image IEEE Trans. Image ProcessingProcessing, vol. 10, no. 8, pp. 1252-1263, Aug. 2001., vol. 10, no. 8, pp. 1252-1263, Aug. 2001.
2.2. P. Moulin and M.K. Mıhçak, “A Framework for Evaluating the P. Moulin and M.K. Mıhçak, “A Framework for Evaluating the Data-Hiding Capacity of Image Sources”, Data-Hiding Capacity of Image Sources”, IEEE Trans. Image IEEE Trans. Image ProcessingProcessing, vol. 9, no. 8, pp. 1450-1455, Aug. 2000. , vol. 9, no. 8, pp. 1450-1455, Aug. 2000.
3.3. S. Voloshynovskiy and T. Pun, “Capacity-Security Analysis of S. Voloshynovskiy and T. Pun, “Capacity-Security Analysis of Data Hiding Technologies,” in Data Hiding Technologies,” in Proc. of IEEE Int. Conf. on Proc. of IEEE Int. Conf. on Multimedia and ExpoMultimedia and Expo, vol. 2, pp. 477-480, Aug. 2002., vol. 2, pp. 477-480, Aug. 2002.
4.4. C.E. Shannon, “A Mathematical Theory of Communication,” C.E. Shannon, “A Mathematical Theory of Communication,” Bell Systems Technical JournalBell Systems Technical Journal, vol. 27, pp.373-423, 623-656, , vol. 27, pp.373-423, 623-656, 1948.1948.
5.5. M. Costa, “Writing on Dirty Paper,” M. Costa, “Writing on Dirty Paper,” IEEE Trans. Inform. IEEE Trans. Inform. TheoryTheory, vol. 29, no. 3, pp. 439-441, May 1983., vol. 29, no. 3, pp. 439-441, May 1983.