A NEW SCHEME FOR DUAL WATERMARKING USING
DWT-PCA TECHNIQUE
Marzieh Amini
1
, Hamidreza Sadreazami
2
and Khashayar Yaghmaie
1
1
Department of Electrical Engineering, University of Semnan, Semnan, Iran
2
Department of Electrical Engineering, Shahid Beheshti University, Tehran, Iran
Keywords: Dual Watermarking, Discrete Wavelet Transform, Principal Component Analysis.
Abstract: In this paper, we propose a new scheme of dual watermarking by applying discrete wavelet transform and
principal component analysis. For the embedding, we identify two best sub bands of wavelet transform by
considering the intensity variance of each sub band. Two watermarks are embedded within the selected sub
bands with respect to their principal components. Using dual watermark improves robustness and
imperceptibility of algorithm and also takes advantage of using efficient bandwidth by embedding two
watermarks into host image. Interleaving technique is used to improve imperceptibility of the algorithm.
Experimental results show no visible difference between host and watermarked images. The proposed
watermarking method is also robust to various attacks such as JPEG compression, cropping, histogram
modification and gamma correction. This robustness is more noticeable when cropping or gamma correction
applies to the watermark image.
1 INTRODUCTION
In recent years, digital watermarking technology has
attracted great research interests. Watermarking is a
way of embedding a key into the original data in
order to increase security and copyright protection.
Several spatial domain and transform domain digital
watermarking algorithms have been proposed
(Mehul, 2003), (Dugad, 1998) Watermarking
schemes of transform domain have more advantages
than those in spatial domain. Discrete wavelet
transform (DWT) has some unique characteristics in
order to compatibility with human visual system
(HVS). Several works have been proposed to
combine DWT with other techniques in order to
increase robustness and imperceptibility
(Loukhaoukha, 2009), (Reddy ,2005). Principal
component analysis (PCA) is one of transformations
which have been used in watermarking (Hien, 2008),
(Kang, 2008), (Mostafa, 2009). The main
characteristic of this transform is its high energy
concentration and complete decorrelation. To
improve the robustness and protection, dual
watermarking has been employed (Bhatnagar, 2008).
In dual watermarking the primary and secondary
watermarks are embedded into host image in a way
that the primary watermark is a copyright symbol
and the secondary watermark is a fusion image. Also
we can take advantage of using efficient bandwidth
by embedding two watermarks into host image.
In this paper, after employing DWT on the host
image, we select two significant sub bands for
embedding two watermarks not only by using HVS
characteristics but also based on intensity variance
of each sub bands (Mansouri, 2009). In this way we
consider both robustness against attack and the best
quality of watermarked image. Then PCA is applied
to both sub bands in order to concentrate the energy
of coefficients and distributes the watermark energy
over embedding sub bands. As a result, we achieve
better robustness, perceptual transparency, and good
localization. The rest of the paper is organized as
follows: Section 2 provides a quick review on PCA.
In Section 3, proposed embedding and extracting
algorithms are explained. Experimental results are
presented in section 4. Finally, section 5 concludes
the paper.
43
Amini M., sadreazami H. and Yaghmaie K.
A NEW SCHEME FOR DUAL WATERMARKING USING DWT-PCA TECHNIQUE.
DOI: 10.5220/0002848500430046
In Proceedings of the International Conference on Imaging Theory and Applications and International Conference on Information Visualization Theory and Applications (VISIGRAPP 2010),
page
ISBN: 978-989-674-027-6
Copyright
c
 2010 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

2 PRINCIPAL COMPONENT
ANALYSIS
Principal component analysis is used to project a
large number of variables to lower dimension of
their linear combinations that adequately describe
the system. In this way, it re-expresses data as a
linear combination of its basic vectors. The basic
idea behind using the PCA is to distribute the
watermark energy over all sub bands of wavelet
transform, resulting watermark robustness by
representing an excellent domain for inserting the
watermark. To obtain the PCA components of
matrix
X
, firstly calculate the covariance matrix as
is defined by:
{
}
T
x
mXmXEC )()( −×−=
(1)
Where
E
, m and
T
denote expectation operation,
mean of matrix X and matrix transpose, respectively.
The principal components of X are the eigenvectors
of
x
C
which can be derived by:
Φ=Φ λ
x
C
(2)
In which
Φ
and
λ
are the matrix of eigenvecors
and matrix of eigenvalues defined as
()
n
eeee ,...,,,
321
=Φ and ()
n
λλλλλ ,...,,,
321
= The
matrix
Φ is an orthogonal matrix called basis
function of PCA. PCA transform the correlated
image into uncorrelated coefficients by taking the
inner product of the Image with basic function
Φ
.
X
Y
T
Φ=
(3)
Where
Y
is the PC matrix which represents the
principle component of matrix
X
.
3 WATERMARKING SCHEME
In this section a new dual watermarking is presented.
We have used DWT and PCA in order to develop
the algorithm. The proposed watermark algorithm
can be decomposed into two parts: embedding and
extracting.
Figure 1: The binary watermark and interleaved logos.
3.1 Embedding Algorithm
We assume that the grayscale host image
X
of size
NN
×
and two binary image watermark of size
P
P
×
, shown in Figure 1. The embedding algorithm
can be divided into the following steps:
1. Permute the watermark bits randomly using an
interleaver with a secret seed in order to disperse
the spatial relationship and to increase the
invisibility.
2. Perform 1-level wavelet transform on host
image. To choose the appropriate sub bands for
embedding watermark, the intensity variance of
each sub-band is calculated, and then two sub-
bands with the mid values are selected as the best
ones.
3. Divide the two sub bands into
n blocks which
PPn
×
=
,
is the number of watermark bits.
4. Perform PCA transform on all non overlapping
blocks of both sub bands.
5. First watermark bits are embedded in an additive
way into bits of each block based on their energy
order. For example the first bit of watermark is
added with a bit of the first block which is in
position with the highest energy or the first
principal component of each block (
Mostafa,
2009). Embedding equation is as follow:
WaYY
ow
.+=
(4)
Where
o
Y and
w
Y are, the first principal
component of each block before and after adding
the watermark bits and
a
is the embedding
intensity factor. This procedure is done in the
same way for the second watermark bits.
6. Apply inverse PCA on two sub bands and then
inverse wavelet transform to get the watermarked
image.
3.2 Extraction Algorithm
As our proposed watermarking scheme is a non-
blind watermarking, the host image is required in
watermarking extraction. The extracted process is as
follows:
1. Decompose the watermarked image (with or
without considering attack) into four sub bands.
2. Divide two selected sub bands into
n blocks and
apply PCA to each block.
3. Extract every bit embedded into the first
principal component of each block by the
following equation:
IMAGAPP 2010 - International Conference on Imaging Theory and Applications
44

a
YY
W
ow
extract
)( −
=
(5)
4. Each watermark logo is recovered by
deinterleaver and using its secret seed.
4 EXPRIMENTAL RESULTS
Several experiments are conducted to demonstrate
the imperceptibility and robustness of the proposed
method. In all of these experiments, the original
image is 512×512 grayscale image of Lena and the
watermark are selected as logos with 32×32 size.
Peak signal to noise ratio (PSNR) between the
original and watermarked images is computed in
order to evaluate the imperceptibility of the
algorithm.
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
),(
)max(
log10),(
2
10
w
w
XXMSE
X
XXPSNR
(6)
2
11
2
)),(),((
),(
N
jiXjiX
XXMSE
N
i
N
j
w
w
∑∑
−
=
==
(7)
Also normalized correlation (NC) is used for
evaluating the similarity of the watermark and
extracted logos.
∑∑
∑∑
=
==
==
P
i
P
j
P
i
P
j
extracted
jiW
jiWjiW
NC
11
2
11
),(
),().,(
(8)
Host Watermarked
PSNR=40.67
Watermark logos Extracted logos NC=1
Figure 2: The host image, its watermarked, watermark
logos and their extracted. PSNR=40.67 and for both logos
NC=1.
Figure 2 indicates that the applied method
created no perceptible artefact on the host image.
Also both watermark logos have been perfectly
extracted. Figure 3 shows the effect of some
common attacks on process of extracting. PSNR and
NC related to each attack have also been computed.
As shown in Figure 3, the proposed watermarking
scheme has good robustness when faced with attacks
like cropping, histogram modification and gamma
correlation. NC for each extracted logo implies a
satisfactory level of watermark robustness.
Figure 3: Extracted logos with their corresponding NC and
PSNR values after various attacks on watermarked Lena
image.
JPEG compressed
QF=100
Extracted
logo1
Extracted
logo2
PSNR =35.9628 NC =0.9420 NC =0.9754
JPEG compressed
QF=80
Extracted
logo1
Extracted
logo2
PSNR =32.0405 NC =0.9049 NC =0.9381
Cropping 25% Extracted
logo1
Extracted
logo2
PSNR =11.8377 NC =0.8956 NC =0.9440
Cropping 50% Extracted
logo1
Extracted
logo2
PSNR = 8.7675 NC = 0.8307 NC =
0.9393
A NEW SCHEME FOR DUAL WATERMARKING USING DWT-PCA TECHNIQUE
45

Figure 3: Extracted logos with their corresponding NC and
PSNR values after various attacks on watermarked Lena
image. (cont.)
5 CONCLUSIONS
A new scheme of dual watermarking based on PCA
and DWT is presented in which both watermark
logos are visually meaningful binary images.
Choosing the most significant sub bands is in
accordance with intensity variance of each sub band.
Dividing each sub band to
n blocks, PCA is applied
on each single block in order to concentrate the
energy of block coefficient. First principle
component of each block is selected for embedding
watermark. Also interleaving technique is used to
improve imperceptibility of the algorithm.
Experimental results show no visible difference
between the host and watermarked images. It is also
demonstrated that proposed scheme is considerably
robust against attacks such as cropping, histogram
modification and gamma correction.
REFERENCES
Mehul, R. & Priti, R., 2003. Discrete Wavelet Transform
Based Multiple Watermarking Scheme, Proceeding of
IEEE, Technical Conference on Convergent
Technologies for the Asia-Pacific Bangalore, pages
14-17.
Dugad, R., Ratakonda & K., Ahuja, N., 1998. A new
wavelet-based scheme for watermarking images.
Image Processing , 2:419-423.
Loukhaoukha, K. & Chouinard, J. Y., 2009. A NEW
IMAGE WATERMARKING ALGORITHM BASED
ON WAVELET TRANSFORM, CCECE, 229-234.
Reddy, A. A. & Chatterji, B. N. , 2005, A new wavelet
based logo-watermarking scheme, Pattern Recognition
Letters , 26:1019-1027.
Hien, T. D. & et al., 2004. Robust digital watermarking
based on principle component analysis, International
Journal of Computational Intelligence and
Applications, 4(2):183-192.
Kang, X., Zeng, W. & Huang, J., 2008. A Multi-band
Wavelet Watermarking Scheme, International Journal
of Network Security, 6(2):121-126.
Mostafa, S. A. K. & et al., 2009. Video Watermarking
Scheme Based on Principle Component Analysis and
Wavelet Transform, IJCSNS International Journal of
Computer Science and Network Security, 9(8):45-52.
Bhatnagar, G., Raman, B., 2008. Dual watermarking
scheme via sub-sampling in WPT-SVD domain, First
International Conference on Emerging Trends in
Engineering and Technology, 2(3):850-855.
Mansouri, A., Aznaveh, A. A. & Torkamani Azar, F.
2009. SVD-based digital image watermarking using
complex wavelet transform, Sadhana, 34(3):393–406.
Histogram
Modification
Extracted
logo1
Extracted
logo2
PSNR=19.4186 NC =0.8535 NC =0.9301
Gaussian Filtering
mask 7×7
Extracted
logo1
Extracted
logo2
PSNR =40.1165 NC =0.8507 NC =0.9209
Gamma correction
Gamma=1.5
Extracted
logo1
Extracted
logo2
PSNR =27.7601 NC =0.9158
NC
= 0.9490
Gamma correction
Gamma=0.7
Extracted
logo1
Extracted
logo2
PSNR 19.2352 NC =0.9159 NC =0.9747
IMAGAPP 2010 - International Conference on Imaging Theory and Applications
46