A novel blind color images watermarking based on SVD

Transcription

A novel blind color images watermarking based on SVD
G Model
IJLEO-54174; No. of Pages 7
ARTICLE IN PRESS
Optik xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Optik
journal homepage: www.elsevier.de/ijleo
A novel blind color images watermarking based on SVD
Shao-li Jia ∗
College of Civil Engineering, Lu dong University, Yantai, 264025 Shandong, PR China
a r t i c l e
i n f o
Article history:
Received 12 June 2013
Accepted 16 November 2013
Available online xxx
Keywords:
Color image
Watermark
SVD
U matrix.
a b s t r a c t
Since the color image watermark has more bit information, it is a challenging problem to design a robust
and blind color watermarking scheme for copyright protection. In this paper, a blind watermarking
scheme based on singular value decomposition (SVD) is proposed. By analyzing the orthogonal matrix
U via SVD, it is found that there exists a strong similarity correlation between the second row first column element and the third row first column element. Hence, this paper utilizes this property for image
watermarking. Firstly, the 4 × 4 non-overlapping pixels block of each component in color host image is
processed by SVD. And then, the color watermark is embedded by slightly modifying the value of the second row first column element and the third row first column one of U matrix, and the modified relation
can be utilized to extract watermark. Experimental results, compared with the related existing methods,
show that the proposed color image scheme has stronger robustness against most common attacks such
as image compression, filtering, cropping, noise adding, blurring, scaling and sharpening et al.
© 2014 Elsevier GmbH. All rights reserved.
1. Introduction
Today storing information and data such as documents, images,
video, and audio in digital formats is very common. As is well
known, due to the nature of digital information, it is easy to
make unlimited lossless copies from the original digital source,
to modify the content, and to transfer the copies rapidly over
the Internet. Therefore, the demands of copyright protection,
ownership demonstration, and tampering verification for digital
data are becoming more and more urgent. Among the solutions
for these problems, digital watermarking is the most popular one.
Researchers have given consideration to this in the past decade.
Digital watermarking is to embed the important information
into the digital data (audio, image, video, and text). According
to the processing domain of the host image, the existing techniques on image watermarking may be roughly divided into
two categories, i.e., frequency-domain and spatial-domain methods [1]. Although more information for embedding and better
robustness against the common attacks can be achieved through
frequency-domain method, the computational cost is higher than
the ones based on spatial domain. Embedding the watermark
into the component of the original image in spatial domain is a
straightforward method, which has the advantages of low computational complexity and easy implementation. However, the
watermarking algorithm in spatial domain is generally fragile to
common image processing operations or other attacks. In order to
∗ Tel.: +86 535 6673554.
E-mail address: ldujiaoshaoli@163.com
overcome these shortcomings, the method based on singular value
decomposition (SVD) has been becoming one of the research hot
fields. The SVD-based watermarking method has three advantages
as follows: (1) the size of the matrix from SVD transformation is
not fixed; (2) when a small perturbation is added to an image,
larger variation of its singular values does not occur; (3) singular
values represent intrinsic algebraic image properties [2].
In the last few years, SVD-based watermarking technique and its
variations have been mostly considered, e.g., in [3–6]. Among them,
a vector quantization-and SVD-based image hiding algorithm was
introduced for embedding the secret data into the D matrix of the
SVD in [3]. Chang, et al. [4] discussed a block-based watermarking algorithm, in which the image was divided into several blocks
and then the elements in U matrix in each block were modified to
achieve the watermarking effect. In [5], two notes were proposed
to increase invisibility and capacity of SVD-based watermarking
scheme, in which the elements in column/row vector were modified to attain less visible distortion than modifying the elements
in row/column vector of U matrix after SVD transformation. On the
basis of the method in [5], Fan, et al. [6] further considered modifying the elements in the first column of U matrix and V matrix for
watermarking. Moreover, the V matrix or U matrix to compensate
visible distortion was adopted in [6] when embedding watermark
into the matrix of SVD.
However, it is noted that the SVD-based methods have four disadvantages:
1) Most methods only consider the case that embedding binary
watermark image into the gray-scale image [3–6].
0030-4026/$ – see front matter © 2014 Elsevier GmbH. All rights reserved.
http://dx.doi.org/10.1016/j.ijleo.2014.01.002
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002
G Model
IJLEO-54174; No. of Pages 7
2
ARTICLE IN PRESS
S.-l. Jia / Optik xxx (2014) xxx–xxx
2) The watermarked image has lower invisibility [7]. This is
because all values in a pixel matrix would be changed when
one singular value is modified, that is, if one singular value of a
N × N matrix is modified, then N2 pixels will be modified.
3) The false positive detection problem is existing in most SVDbased watermarking algorithm [8]. The detailed reason for false
positive detection problem is only the singular values of the
watermark W are embedded into the host image, that is, if the
T , only the
singular value decomposition of W is W = UW SW VW
diagonal singular values matrix SW is embedded while the orthogonal matrices UW and VW are not. In extraction procedure, only
the diagonal matrix SW is extracted, but the UW and VW can be
simply provided by the owner without any extraction. However,
the orthogonal matrices UW and VW contain the major information about an image. Thus any one can provide a fake pair of
orthogonal matrices and claim that his watermark is present in
the watermarked image, which causes false positive detection
problem.
4) Most the aforementioned watermarking methods were usually performed in a non-blind manner. For example, in [9], the
singular values of original watermark are required to extract
the embedded singular value, and then, the U and V orthogonal matrices of original watermark were utilized to recover
the watermark. In [10], three matrices U, V and D of SVD for
watermark are used the user’s secret keys to extract watermark, and the original host image is also need to extract
watermark.
2. Singular value decomposition
For a N × N square matrix I with rank r, r ≤ N, its SVD is represented by Eq. (1).
⎡u
1,1
⎢ u2,1
⎢
⎣ ..
I = UDVT = ⎢
.
uN,1
···
u1,N
···
u2,N
..
.
.
.
.
···
⎤⎡ ⎥⎢
⎥⎢
⎥⎢
⎦⎣
uN,N
1
0
···
0
0
2
···
0
.
.
.
.
.
.
..
.
0
0
0
···
N
⎤⎡ v
1,1
⎥ ⎢ v2,1
⎥⎢
⎥⎢ .
⎦⎣ .
···
v1,N
···
v2,N
..
.
.
.
.
.
⎤
⎥
⎥
⎥
⎦
(1)
vN,1 · · · vN,N
where U and V are N × N orthogonal matrices and D is singular,
diagonal matrix with diagonal elements satisfying
1 ≥2 ≥· · ·≥r > r+1 = · · · = N = 0
It is assumed that the 4-by-4 matrix A is one of the blocks of the
input image, whose SVD is given by Eq. (2).
⎡ A1
A2
⎢ A5
⎣ A9
A6
A7
A8
A10
A11
A12
A=⎢
A13
⎡ a1
A14
A3
A4
A15
a2
a3
⎤
⎥
⎥ = UDVT
⎦
A16
a4
⎤ ⎡ 1
⎢ a5
⎣ a9
a6
a7
a10
a11
⎥⎢
⎥⎢
a12 ⎦ ⎣
a13
a14
a15
a16
=⎢
a8
0
0
0
0
2
0
0
0
0
3
0
0
0
⎤ ⎡ c1
c2
c3
c4
⎤T
c6
c7
0
⎥ ⎢ c5
⎥⎢
⎦ ⎣ c9
c10
c11
⎥
⎥
c12 ⎦
4
c13
c14
c15
c16
(2)
c8
Perform the matrix multiplications for UDVT , in which each pixel
Although the watermarking proposed in [11] can attain a blind
color image extraction watermarking, one or more singular values
must be modified to keep the order of singular values such that the
quality of the watermarked image will be seriously affected. That
can be seen from the following simple example. For example, supposed that the singular values 1 − 16 of one pixel block with size
16 × 16 are 3165.613, 457.5041, 31.54169, 9.85382997, 5.796001,
4.991171, 3.688464, 2.544742, 2.064232, 1.691997, 1.130058,
1.074023, 0.819865, 0.448544, 0.37897, 0.101045, respectively.
When a watermark value is 0, according to the method in [11], these
singular values will be changed to 3165.613, 457.5041, 31.54169,
9.85382997, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0. That is, 12 singular values
will be modified such that all pixel values in this pixel block will
be obviously changed. Thus, the visual quality of the watermarked
image may be affected.
Motivated by the above discussion, this paper proposes a blind
SVD-based dual color watermarking scheme for protecting copyrights and overcoming the false positive detection problem. In
this paper, it is firstly found the elements in the second row
first column and the third row first column are the closest elements of U matrix after SVD, which means that the relation
between the both elements can be preserved and further used to
extract the embedded watermark without resorting to the original data. Thus, the blind extraction can be achieved. In order to
keep the similarity between the anterior two elements in first
column of U matrix, it is only required to slightly change the
value of the two elements. The experimental results show that
the proposed method has better performance than some existing
methods.
The rest of this paper is organized as follows: Section 2 gives
a brief description of the SVD principle and points out its key
feature that be used in this paper. Section 3 describes the proposed watermarking method that includes watermark embedding
and watermark extraction. In Section 4, the experimental results
are presented to show the performance of the proposed watermark. Finally, we draw out the conclusions of this paper in
Section 5.
is given by Eq. (3).
A1 = a1 1 c1 + a2 2 c2 + a3 3 c3 + a4 4 c4 ; A2
= a1 1 c5 + a2 2 c6 + a3 3 c7 + a4 4 c8
A3 = a1 1 c9 + a2 2 c10 + a3 3 c11 + a4 4 c12 ; A4
= a1 1 c13 + a2 2 c14 + a3 3 c15 + a4 4 c16
A5 = a5 1 c1 + a6 2 c2 + a7 3 c3 + a8 4 c4 ; A6
= a5 1 c5 + a6 2 c6 + a7 3 c7 + a8 4 c8
A7 = a5 1 c9 + a6 2 c10 + a7 3 c11 + a8 4 c12 ; A8
= a5 1 c13 + a6 2 c14 + a7 3 c15 + a8 4 c16
(3)
A9 = a9 1 c1 + a10 2 c2 + a11 3 c3 + a12 4 c4 ; A10
= a9 1 c5 + a10 2 c6 + a11 3 c7 + a12 4 c8
A11 = a9 1 c9 + a10 2 c10 + a11 3 c11 + a12 4 c12 ; A12
= a9 1 c13 + a10 2 c14 + a11 3 c15 + a12 4 c16
A13 = a13 1 c1 + a14 2 c2 + a15 3 c3 + a16 4 c4 ; A14
= a13 1 c5 + a14 2 c6 + a15 3 c7 + a16 4 c8
A15 = a13 1 c9 + a14 2 c10 + a15 3 c11 + a16 4 c12 ; A16
= a13 1 c13 + a14 2 c14 + a15 3 c15 + a16 4 c16
According to the above formulas, the value of Aj (1 ≤ j ≤ 16)
depends on the singular values i (1 ≤ i ≤ 4). The more the modified number for all i is, the bigger the changed magnitude in pixel
values is, and the worse the invisibility of the watermark is. That is
the main shortcoming of the method in [11].
It is noted that the matrix U has an interesting property, i.e., all
elements in the first column are of same sign and their values are
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002
G Model
ARTICLE IN PRESS
IJLEO-54174; No. of Pages 7
S.-l. Jia / Optik xxx (2014) xxx–xxx
3
Color watermark image
Color host image
Embedding
Watermarked image
Process
Extraction
Extracted watermark image
Process
Key
Key
Fig. 1. Block diagram of the proposed watermarking scheme.
Fig. 2. Original watermark images: (a) PEUGEOT logo, (b) 8-color image.
very close. For example, a sample matrix A obtained from a digital
image is
⎡
252
250
250
251
⎤
⎢ 255 255 255 255 ⎥
⎥
A=⎢
⎣ 254 254 253 253 ⎦
244
(4)
239 234 229
By undergoing SVD to matrix A, the orthogonal matrix U is given
as
⎡
−0.5034
⎢ 0.5119
⎣ 0.5089
U=⎢
−0.4749
−0.2588
0.8221
−0.0616
−0.3288
−0.3641
0.7051
0.8780
−0.0099
0.0595
⎤
⎥
⎥
−0.2325 −0.4376 −0.7039 ⎦
(5)
As can be seen from U matrix, all the elements in the first column of U matrix are of same sign (negative) and difference of first
column elemental is very small. A matrix consisting of m-th row
first column element Um,1 of each SVD decomposed U matrix block
and another one consisting of n-th row first column element Un,1
of each SVD decomposed U matrix block is formed. Normalized
Cross-Correlation (NC) between the two matrices is calculated and
is listed in Table 1. As can be seen from the table that this value
is very close to 1 for most of the images, and the average value of
NC (U2,1 , U3,1 ) is 0.9886 which shown they are the closest elements
than any of other elements. Therefore, it is noted that there exists
a strong correlation between the second row first column element
and the third row first column element of U matrix when SVD is
used. This property can be explored for image watermarking.
3. The proposed watermarking scheme
As mentioned above, one of the features of SVD is that the relation between the elements in the first column vector of the U matrix
could be preserved, while the others are changed when the general
image processing is performed. In this work, we shall make full
use of the property to embed the color watermark image into the
color host image. As shown in Fig. 1, the color watermark image
is embedded into color host image and the final watermark can be
extracted from the watermarked image without the original host
image or watermark image.
Without loss of generality, let the original host image H be 24bit color image with size of M × M and the watermark image be
24-bit color image W with size of N × N. The detailed embedding
procedures are as follows.
Step 1. Pre-processing of the color image watermark. The 3-D original watermark image W is firstly divided into three components R,
G, B by dimension-reduction treatment. And then, the 2-D component watermarks Wi are obtained, where i = {1, 2, 3} presents the R,
G and B component, respectively. In order to enhance the security
and robustness of the watermarking, each component watermark
is permuted by Arnold transformation based on the private key K
and converted from decimal format to binary sequence.
Step 2. Block processing of the host image. The host image H is
represented by R, G, B component images and each component
image is partitioned into 4 × 4 non-overlapping blocks.
Step 3. Performing SVD decomposition on each block as Eq. (1) to
obtain the U matrix of each embedding block.
Step 4. According to the watermark information w to modify the
elements of u2,1 and u3,1 in the U matrix of each embedding block.
The watermark is embedded by changing the relation between the
second (u2,1 ) and the third (u3,1 ) elements in the first column. If
the embedded binary watermark bit is 1, the value of (u2,1 − u3,1 )
should be negative and its magnitude is greater than a threshold T.
If the embedded binary watermark bit is 0, the value of (u2,1 − u3,1 )
should be positive and its magnitude is greater than a threshold T.
When these two conditions are violated, the elements of u2,1 and
u3,1 should be modified as u2,1 and u3,1 , respectively, based on the
following rules in Eqs. (6) and (7).
u2,1 = sign(u2,1 ) × (Uavg + T/2)
if w = 1 &u2,1 − u3,1 < −T, then
if w = 0 &u2,1 − u3,1 < T, then
(6)
u3,1 = sign(u3,1 ) × (Uavg − T/2)
u2,1 = sign(u2,1 ) × (Uavg − T/2)
(7)
u3,1 = sign(u3,1 ) × (Uavg + T/2)
where sign (x) presents the sign of x, Uavg = (u2,1 + u3,1 )/2,
|x| denotes the absolute value of x.
Step 5. Obtain the watermarked component image by performing
inverse SVD to all selected blocks subsequently.
Step 6. Repeat step 3-step 5 until all watermark bits are embedded in the host image. Finally, recombine the watermarked R, G, B
components and the watermarked image H is obtained.
The steps in the watermark extraction procedure are as follows:
Step 1. The watermarked image H is divided into R, G, B component images, which are further divided into watermarked blocks
with size of 4 × 4 pixels, respectively.
Step 2. Apply SVD to watermarked image block and get the U matrix of each block.
Step 3. The relation between the second and the first elements in
the first column of the U matrix is used to extract the watermark
information w , as shown in Eq. (8).
w =
0, if u2,1 > u3,1
1, if u2,1 ≤ u3,1
(8)
Step 4. Repeat step 2–step 3 until all embedded image blocks are
performed. These extracted bit values are converted to decimal
format, then the inverse-Arnold transform based on the private
key K is executed and the extracted watermark of each component
is reconstructed.
Step 5. Reconstruct the final extracted watermark W from the
extracted watermarks of the three components.
4. Experiment results
In this experiment, all 24-bit 512 × 512 color images in the CVGUGR image database are used as the host images [12]. Additionally,
two 24-bit color images with size of 32 × 32, as shown in Fig. 2(a)
and (b), are used as original watermarks.
Generally, the bigger the size of image block is, the smaller the
watermark capacity is, the worse the watermark invisibility is and
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002
G Model
ARTICLE IN PRESS
IJLEO-54174; No. of Pages 7
S.-l. Jia / Optik xxx (2014) xxx–xxx
4
Table 1
NC values of different elements in first column of U matrix after SVD.
Image
NC (U1,1 ,U2,1 )
NC (U1,1 ,U3,1 )
NC (U1,1 ,U4,1 )
NC (U2,1 ,U3,1 )
NC (U2,1 ,U4,1 )
NC (U3,1 ,U4,1 )
Lena
House
Peppers
F16
Baboon
Bear
Kid
Sailboat
Barbara
Couple
Average
0.9934
0.9966
0.9673
0.9921
0.9709
0.9153
0.9942
0.9879
0.9882
0.9323
0.9738
0.9886
0.9942
0.9482
0.9873
0.9589
0.8848
0.9896
0.9798
0.9785
0.9006
0.9610
0.9871
0.9935
0.9444
0.9815
0.9525
0.8839
0.9823
0.9779
0.9728
0.8907
0.9567
0.9969
0.9990
0.9871
0.9972
0.9796
0.9564
0.9962
0.9967
0.9947
0.9818
0.9886
0.9901
0.9949
0.9554
0.9884
0.9579
0.9069
0.9852
0.9796
0.9814
0.9219
0.9662
0.9940
0.9969
0.9692
0.9940
0.9716
0.9341
0.9919
0.9876
0.9913
0.9538
0.9784
the better the watermark robustness is, vice verse. Meanwhile, the
bigger the threshold T is, the worse the watermark invisibility is and
the better the watermark robustness is, vice verse. Considering the
tradeoff between the invisibility and the robustness of watermarking and based on many experiments, let the size of image block be
4 × 4 and the threshold T be 0.040.
For the imperceptible capability, the structural similarity (SSIM)
index as a new method is used to measure the similarity between
the original color image H and the watermarked image H . SSIM
is designed to improve on traditional methods like peak signalto-noise ratio (PSNR) and mean squared error (MSE), which have
proved to be inconsistent with human eye perception [13]. The
detailed SSIM is defined as follows:
SSIM(H, H ) = l(H, H )c(H, H )s(H, H )
(9)
where
⎧
2H H + C1
⎪
l(H, H ) = 2
⎪
⎪
+ 2 + C1
⎪
H
⎪
H
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎩
c(H, H ) =
2H H + C2
H2
+ 2
H
(10)
+ C2
+ C3
s(H, H ) = HH
H H + C3
The first term in Eq. (10) is the luminance comparison function
which measures the closeness of the two images’ mean luminance (H and H ). This factor is maximal and equal to 1 only
if H = H . The second term is the contrast comparison function
which measures the closeness of the contrast of the two images.
Here the contrast is measured by the standard deviation H d H .
This term is maximal and equal to 1 only if H = H . The third
term is the structure comparison function which measures the cor
relation coefficient between the two images H and H . Note that
HH is the covariance between H and H . The positive values of the
SSIM index are in [0,1]. The result 0 means no correlation between
images, and 1 means that H = H . The larger the SSIM value is,
the more imperceptible the embedded watermark is. That is, the
watermarked image is very similar to the original ones. The positive
constants C1 , C2 and C3 are used to avoid a null denominator.
Moreover, Normalized cross-correlation (NC), which calculated
by the color original watermark image and the extracted watermark W in Eq. (11), is used to measure the robustness of the
watermarking [14,15].
Q
P
3 NC =
(W (x, y, i) × W (x, y, i))
i=1 x=1 y=1
3 P Q
3 P Q
2
2
[W (x, y, i)] [W (x, y, i)]
i=1 x=1 y=1
i=1 x=1 y=1
(11)
Fig. 3. Original host images: (a) Lena, (b) Avion, (c) Peppers, (d) TTU.
In which, P, Q respectively denote the width and height of the
watermark image. Generally, the NC can take any value between
0 and 1. If the NC value is closer to 1, the extracted watermark
is getting more similar to the embedded one, which means that
the watermarking has strong robustness. In general, the watermark
may be efficient if the NC is more than or equal to 0.750, conversely
maybe inefficient [16].
4.1. Invisibility test
In order to fairly evaluate and prove the invisibility of watermark, as shown in Fig. 3, three standard 24-bit color images (Lena,
Avion, Peppers) with size of 512 × 512 are selected from the [11]
and one image (TTU) from [14].
Fig. 4 illustrates the watermarked color images and their
SSIM values, and shows the comparison of extracted watermarks
between [11] and the proposed method without any attacks. The
NC values are close to 1, which shown the extracted watermark is
very similar to the original watermark, and the extracted watermark also has proved the matter. In addition, the watermark that
extracted from the watermarked image by the proposed scheme
has a little better visual performance than the method in [11]. The
SSIM values are also equal to 1, which illustrates the method [11]
and the proposed method all have better watermark hiding feature. Relatively, the proposed method is superior to the method
[11], which because the tradeoff between the invisibility and the
robustness is considered in this proposed method, that is, the bigger threshold T is, the worse the invisibility is and the better the
robustness is, and vice versa.
4.2. Robustness test
JPEG compression attack is one of the common attacks that
should be verified in watermarking algorithm. In this experiment,
the watermarked Lena image is lossy-compressed with different
compression factors from 10 to 100.
JPEG 2000 is developed by the JPEG in the aim of improving the
properties of the JPEG standard. The watermarked Lena image is
also performed by JPEG 2000 compression with the compression
ratio from 1 to 10 increasing in steps of 1.
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002
G Model
IJLEO-54174; No. of Pages 7
ARTICLE IN PRESS
S.-l. Jia / Optik xxx (2014) xxx–xxx
5
Fig. 4. The watermarked color images and the extracted watermarks without any attacks.
In addition, a salt & peppers noising scheme is performed by
generating noise to corrupt the watermarked Lena image. The noising scheme generates 2% and 10% noise, respectively, to degrade
the watermarked images. Moreover, the Gaussian noising is also
added to corrupt the watermarked image. The mean of Gaussian
parameters are set to 0.1, 0.3, respectively.
In the procedure of sharpening, the radii are 0.2 and 1.0, respectively. And two scaling operations of 400% and 25% are utilized to
deteriorate the watermarked Lena image. In the blurring attacks,
two cases are simulated here to degrade the two watermarked
images. The radius in the first case is 0.2.
After the above image attacks, the part extracted watermarks
can be list in the Fig. 5.
To highlight the robustness of the proposed method, many other
kinds of image watermarking attacks are also tested with the Lena
image. Table 2 lists the NC results compared with the method in
[11]. From Table 2, it can be seen that the proposed watermarking method has strong robustness against some common attacks,
including salt & pepper noise, Gaussian noise, contrast adjustment, median filtering, scaling, blurring, sharpen and cropping
attacks.
In order to further prove the robustness of the proposed methods, we also compared to the method [14]. In [14], two color images,
‘Peppers’ and ‘TTU’, were used as the host image in Fig. 3(c) and (d),
one color image was viewed as the watermark image in Fig. 2(b)
and all attacks in Table 3 came from the [14]. It can be seen from
Table 3, the robustness of watermark in the proposed method is
superior to the method in [14].
Table 2
NC values comparison between Golea et al. [11] and the proposed scheme under
some common attacks.
Attack types
Parameters
Golea et al. [11]
Proposed method
Salt and pepper noise
0.02
0.04
0.06
0.08
0.1
0.1
0.2
0.3
0.4
0.5
default
1×1
2×2
3×3
4×4
5×5
0.5
1.5
2
2.5
0.1
1
0.1
0.2
10:100,10:100
40
90
10:1
0.5167
0.3350
0.27242
0.2364
0.2150
0.8158
0.8600
0.8327
0.8164
0.7469
0.9913
0.9936
0.6519
0.5074
0.3792
0.3217
0.5698
0.7714
0.8385
0.7892
0.9936
0.2702
0.8481
0.7808
0.8706
0.7257
0.9772
0.8071
0.9970
0.9938
0.9870
0.9782
0.9730
0.9579
0.9537
0.9111
0.8265
0.7431
1.0000
1.0000
0.8950
0.7136
0.7001
0.6708
0.9037
0.9932
0.9921
0.9943
1.0000
0.8804
0.9994
1.0000
0.9297
0.8772
0.9955
0.9587
Gaussian noise
Contrast adjustment
Median filtering
Scaling
Blurring
Sharpening
Cropping
JPEG
JPEG2000
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002
G Model
IJLEO-54174; No. of Pages 7
ARTICLE IN PRESS
S.-l. Jia / Optik xxx (2014) xxx–xxx
6
Table 3
Comparison results of extracted watermarks in terms of visual perception and NC values for Peppers and TTU images.
Host image
Attack type
Chou et al. [14]
NC
Proposed method
Extracted watermark
NC
Peppers
Low-pass filtering
0.539
0.8906
Peppers
Crop50%
0.553
0.9288
Peppers
Scale 1/4
0.536
0.8272
Peppers
Scale 4
0.851
0.9657
Peppers
Rotation 30
–
–
Peppers
JPEG compression ratio of 12
0.439
0.9978
Peppers
JPEG compression ratio of 27.5
0.343
0.9340
TTU
Low-pass filtering
0.423
0.8781
TTU
Gaussian noises addition variance 4
0.982
0.9239
TTU
Gaussian noises additionvariance 25
0.360
0.7535
TTU
Median-filtering
0.170
0.2788
Extracted watermark
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002
G Model
IJLEO-54174; No. of Pages 7
ARTICLE IN PRESS
S.-l. Jia / Optik xxx (2014) xxx–xxx
7
Fig. 5. Extracted watermarks after different attacks: (a) contrast adjustment, (b) cropping, (c) JPEG 70 (d) median filtering 3 × 3, (e) JPEG2000 CR 10:1, (f) scaling 0.5, (g)
scaling 2.5, (h) sharpening 0.2, (i) sharpening 0.1, (j) Gaussian noise 0.1, (k) Gaussian noise 0.5, (l) salt & pepper noise 0.002, (m) salt & pepper noise 0.01, (n) blurring 1.0, (o)
blurring 0.1.
5. Conclusions
An image watermarking scheme SVD-based for protecting copyrights is proposed in this paper. This method has the following
advantages: 1) the embedded watermark image is color image, 2)
the color watermark image is embedded into a color host image
by modifying the relation between the second (u2,1 ) and the third
(u3,1 ) elements of U matrix in each block after SVD, 3) the false
positive detection problem that existing in most SVD-based watermarking algorithm is overcame, and 4) this method belongs to blind
manner. The experimental results show that the proposed scheme
succeeds in making the watermark perceptually invisible and also
robust against various signal processing operations and geometric attacks. Therefore, we conclude that the new proposed scheme
is more suitable for using color image information to protect the
copyright of color images that will be transmitted on the Internet.
As for future work, the artificial intelligent optimization algorithm
will be utilized to improve the watermarking performance.
Acknowledgements
The authors would like to thank anonymous referees for their
valuable comments and suggestions which lead to substantial
improvements of this paper.
References
[1] H.M. Al-Otum, N.A. Samara, A robust blind color image watermarking based on
wavelet-tree bit host difference selection, Signal Process. 90 (2010) 2498–2512.
[2] R. Liu, T. Tan, An SVD-based watermarking scheme for protecting rightful
ownership, IEEE Trans. Multimed. 4 (2002) 121–128.
[3] K.-L. Chung, C.-H. Shen, L.-C. Chang, A novel SVD- and VQ-based image hiding
scheme, Pattern Recog. Lett. 22 (2001) 1051–1058.
[4] C.-C. Chang, P. Tsai, C.-C. Lin, SVD-based digital image watermarking scheme,
Pattern Recog. Lett. 26 (2005) 1577–1586.
[5] K.-L. Chung, W.-N. Yang, Y.-H. Huang, S.-T. Wu, Y.-C. Hsu, On SVD-based watermarking algorithm, Appl. Math. Comput. 188 (2007) 54–57.
[6] M.-Q. Fan, H.-X. Wang, S.-K. Li, Restudy on SVD-based watermarking scheme,
Appl. Math. Comput. 203 (2008) 926–930.
[7] R. Sun, H. Sun, T. Yao, SVD and quantization based semi-fragile watermarking
technique for image authentication, Proc. Internat. Conf. Signal Process 2 (2002)
1952–1955.
[8] A.A. Mohammad, A. Alhaj, S. Shaltaf, An improved SVD-based watermarking
scheme for protecting rightful ownership, Signal Process. 88 (2008) 2158–2180.
[9] C.-Q. Yi, L. Li, A.-Q. Lv, L. Qu, Color image watermarking algorithm based on
DWT-SVD, in: IEEE International Conference on Automation and Logistics,
2007, pp. 2607–2611.
[10] S.-C. Pei, H.-H. Liu, T.-J. Liu, K.-H. Liu, Color image watermarking using SVD,
in: IEEE International Conference on Multimedia and Expo (ICME), 2010, pp.
122–126.
[11] N.E.H. Golea, R. Seghir, R. Benzid, A bind RGB color image watermarking based
on singular value decomposition, in: IEEE/ACS International Conference on
Computer Systems and Applications (AICCSA), 2010, pp. 1–5.
[12] University of Granada, Computer Vision Group. CVG-UGR Image
Database.[2012–10–22]. http://decsai.ugr.es/cvg/dbimagenes/c512.php
[13] Z. Wang, A.C. Bovik, H.R. Sheikh, E.P. Simoncelli, Image quality assessment:
from error visibility to structural similarity, IEEE Trans. Image Process. 13 (4)
(2004) 600–612.
[14] C.-H. Chou, T.-L. Wu, Embedding color watermarks in color images, Eurasip J.
Adv. Signal Process. 1 (2003) 32–40.
[15] A. Al-Gindy, H. Al-Ahmad, R. Qahwaji, A. Tawfik, A high capacity digital watermarking technique for the authentication of colour images, IEEE International
Symposium on Signal Processing and Information Technology (ISSPIT) (2009)
37–42.
[16] L. Li, X. Yuan, Z. Lu, J. Pan, Rotation invariant watermark embedding based on
scale-adapted characteristic regions, Inform. Sci. 180 (15) (2010) 2875–2888.
Please cite this article in press as: S.-l. Jia, A novel blind color images watermarking based on SVD, Optik - Int. J. Light Electron Opt.
(2014), http://dx.doi.org/10.1016/j.ijleo.2014.01.002