International Journal of Security and Its Applications

Vol.8, No.5 (2014), pp.349-360

http://dx.doi.org/10.14257/ijsia.2014.8.5.31

A robust Optimal Zero-Watermarking Technique for Secret

Watermark Sharing

Jumana Waleed1, Huang Dong Jun1 and Saad Hameed2

1
School of information Science and Engineering, Central South University,
Changsha, 410083, China
jumana_waleed@yahoo.com, djhuang@csu.edu.cn
2
College of Information Science and Engineering, Hunan University, Changsha,
China
saad@hnu.edu.cn

Abstract
In this paper, a robust optimal zero-watermarking technique based on genetic algorithm
for secret watermark sharing is proposed for the purpose of copyright protection. It is
implemented in discrete cosine transform (DCT) for gray scale images in which the visual
secret sharing is used to generate unexpanded master and secret shares for the watermark.
The GA based zero watermarking is used to select the perfect positions to extract the robust
features. The experimental results indicate that the proposed scheme is highly robust even
after different types of attacks being applied, where NC values of each selected attack are
considered in the fitness function of the genetic optimization algorithm.

Keywords: Optimal Zero-Watermarking, Genetic Algorithm, Visual Secret Sharing,

Discrete Cosine Transform.

1. Introduction
Digital watermarking is a powerful technique for copyright protection and digital
data ownership. Until the present day, many numbers of watermarking techniques
depends of manipulating both spatial and frequency domains which will render the
digital data with high distortion, as distortion, as commonly known there is a tradeoff
between robustness and imperceptibility. On the other hand, an emerging lossless
technique which known as zero watermarking is proposed, this technique relies on
extracting specific set of features that could identify the digital data uniquely and could
contribute with the construction of the watermark, then registering these features into a
third party database for safe keeping. This database is an intellectual properties rights
for digital media. Many researchers have gone through this field for the purpose of
copyright protection. In [1] two zero watermarks from its host image were generated
together, to improve the robustness, where the first is obtained from low-frequency
coefficients in the (DWT), and the other is obtained from DWT coefficients of the log-
polar mapping to the host image. While [2] constructs the watermark from the low
frequency area in BOR multi-wavelet domain to produce a simple while robust zero-
watermarking. The scheme creates a relationship inside the sub image with low average,
where the center value considered with the mean of the four neighboring coefficients
resulting in a strong technique against many attacks. A zero-bit watermarking algorithm

ISSN: 1738-9976 IJSIA

Copyright ⓒ 2014 SERSC
International Journal of Security and Its Applications
Vol.8, No.5 (2014)

based on the optimized support vector regression (SVR) which presented in [3] to
improve the performance of resisting geometric attacks in digital watermarking. [4 ]
Presents a copyright protection zero-watermark scheme based on discrete cosine
transformation, in which a normalization process commenced on both embedding and
extraction phases to increase the robustness to geometric distortions. Another scheme
where presented which fully exploits the characteristics DWT as described in [5]. This
novel image zero-watermarking scheme based on DWT-SVD, that achieves multi-
resolution decomposition and the SVD, which in turn efficiently represents intrinsic
algebraic properties of an image. In [6] the method takes the highly stable invariant
centroid as the reference point for geometric correction based on the combination of the
image invariant centroid and SIFT feature point. Later it finds out the farthest SIFT
point from the invariant centroid, and then finally estimates out geometric parameters
by changes of these points’ locations.
The visual secret sharing scheme is a visual cryptographic protocol which has the ability
for sharing images in secure way, and restores it without the use of computations [7]. In its
basic model, a binary image is encrypted into n separated images, which reveals the original
image when overlaid. If the user does not have the complete n images then no information
about the encrypted image can be retrieved. Some researchers used the concept of visual
secret sharing of two participants with the zero-watermarking technique to protect the
copyrights of digital images. [8] Embeds a binary watermark in a gray-level image using the
concept of expanded visual secret sharing. While [9] and [10] proposed a scheme aims at
reducing the size of shares (unexpanded shares).
The key of zero-watermarking is to build a watermark to adapt the feature data of the
carrier with the ability to hold a data capable of fully representing the carrier data in addition
to its uniqueness, this is done by considering The transform domain of the carrier data as the
main structural position selected by most existed zero-watermarking technique. It is always
preferable improve the robustness of zero watermarking by exploiting artificial intelligence
techniques and Genetic Algorithms (GAs) as optimization algorithms. The zero watermarking
problems can be viewed as an optimization problem. Therefore, in this paper the attempt is to
provide a solution using genetic algorithms (GAs).
In this paper, an optimal zero-watermarking technique was proposed using the visual
secret sharing to generate unexpanded secret and master shares for the watermark, this
technique extract the feature bits by utilizing the most important parts in the DCT of the gray
image based on genetic algorithm (GA) to get perfect results and obtained a highly robust
watermark. The rest of this paper is organized as follows: the visual secret sharing is given in
the subsequent section. Section 3 is the explanation of the proposed optimal zero-
watermarking technique based on genetic algorithm. In Section 4, we show the presentation
of experiments to demonstrate the robustness of the proposed technique. In Section 5 we have
the conclusions of our technique and the findings obtained.

2. Visual secret sharing scheme

With the emerging of visual cryptography scheme (VCS) or visual secret sharing
(VSS) proposed by Naor and Shamir [7], a new non-intrusive watermarking techniques
guideline have been introduced as in [11] and [12]. In the original visual secret sharing,
any black and white printed material such as text or images etc. are considered as
images and can be split into n different share images. By stacking out a set of qualified
images k where (k ϵ n) the image can be obtained from this k set of images, knowing

350 Copyright ⓒ 2014 SERSC

 International Journal of Security and Its Applications
Vol.8, No.5 (2014)

that even k-1 would not reveal the image at all, not even close to the watermark image.
No computation is involved in obtaining the watermark image from the k shares.
Recently, 2-out-of-2 visual secret sharing schemes are widely adapted in protecting the
copyrights of high precision digital images. In the traditional OR-based (2,2) visual secret
sharing scheme, a secret image is encrypted in two random looking images (transparencies),
each pixel in the secret image is expanded into four sub-pixels (2×2 binary code-words) in
each transparency. The content of the secret image becomes visible by stacking these two
transparencies over each other. Knowing that, there is another model of OR-based (2,2) visual
secret sharing in which each pixel in the secret image is replaced by two sub-pixels in each
share, the width of the decoded image is twice that of the original image as shown in Figure
1. These models increase the size of each share and result in loss of contrast in the recovered
image.
In this paper, XOR-based (2,2) visual secret sharing is used without expansion in the
secret image pixels. The secret watermark image is split into two unexpanded master and
secret shares.

(a) (b)

(c) (d)

Figure 1. An Example of (2,2) Visual Secret Sharing Scheme, for (a) Secret
Image, (b) Share1, (c) Share2, (d) Overlapped Expanded Image

3. The Proposed Optimal Zero-Watermarking Technique based on GA

In this section, we propose a copyright protection technique for improving zero-
watermarking using visual secret sharing applied to a binary watermark and optimized
using genetic algorithm. The proposed optimal zero watermarking technique can be
characterized as follows:

3.1. Embedding Process

To generate the secret share for the watermark, the DCT is performed to the gray image
blocks. The DCT block is consisted of several frequency bands; The single direct current
(DC) coefficient, the low frequency coefficients of the block (BL), the height frequency band
(BH) and the middle frequency coefficients of the block (BM). The most important part (DC)

Copyright ⓒ 2014 SERSC 351

International Journal of Security and Its Applications
Vol.8, No.5 (2014)

coefficients are chosen for generating the feature matrix bits. The embedding process can be
described as follows; Figure 2 is the representation of the embedding process.

Watermark W Arnold Scrubby

Transform Watermark Ws

Rules of X-
Original DCT DC Extract the OR Based Secret
Gray Feature Visual Secret Share
Image Matrix Bits Sharing

GA
Optimization
Secret
Sequence

Figure 2. Embedding Process

The proposed embedding process takes place in the following steps, describing in details
what the implications in each step:
1. Read the watermark image of size n×n. To reduce the relationship of pixel space in the
watermark image, the watermark W is spread out evenly using Arnold transform with
scrambling time K. Empirical value of K is 20 which applied on the watermark W to obtain
spread out watermark Ws as in Equation (1).

(1)

2. Divide the original gray scale image with size m×m into 8×8 non overlapping blocks. To
extract the feature matrix bits; these blocks are converted into the frequency domain using the
two dimensional DCT, B= {B1, B2, …, Bt}, where t is the total number of blocks. Since the
number of the watermark bits (n×n) are very less as compare to total number of blocks (t),
2(n×n) blocks need to be selected from B. A random matrix R having dimension 2(n×n) is
generated, R= {v1, v2, …,v2(n×n)}, where v lies in between 1 and t. The blocks whose block
number matches with v are separated from B.
3. For the randomly selected blocks, the (DC) coefficients are selected to obtain the two (DC)
matrices D1 and D2, each of size (n×n). Then, each element in D1 will be compared with the
corresponding element in D2. The feature matrix F is constructed by using Equation (2).

352 Copyright ⓒ 2014 SERSC

 International Journal of Security and Its Applications
Vol.8, No.5 (2014)

(2)

4. Generate the secret share by using the rules given in Table 1. The secret share and the
secret sequence are then registered into a third party database for safe keeping. This database
is an intellectual properties rights for digital media.

Table 1. Construction Rules of X-OR Based Visual Secret Sharing

Watermark Pixel White Black

Bit in Feature Matrix F 0 1 0 1

Master Share Pixel

Secret Share Pixel

Master Share X-OR Secret

Share

3.2. Extraction process

The inputs to the extraction process are a controversial image and the secret share image.
The output is a watermark image W'. To extract the features of controversial gray image, the
owner uses the same process which is used in the embedding process. Then, the master share
is constructed by using the rules given in Table 1. The X-OR logic function is then applied
between the secret and master shares to reveal the Scrubby watermark W's. Finally, Arnold
transform is applied K times to W's to obtain W' which is used to verify the copyright. Note
that, the size of the shares and the extracted watermark is exactly same since there is no pixel
expansion. Figure 3 represents the extraction process.

Secret
Sequence

Rules of X-
Controversial DCT DC Extract the OR Based Master X- Secret
Color Image Feature visual secret Share OR Share
Matrix Bits sharing

Watermark Arnold Scrubby

W' Transform Watermark
W's

K
Figure 3. Extraction process

Copyright ⓒ 2014 SERSC 353

International Journal of Security and Its Applications
Vol.8, No.5 (2014)

In order to measure the robustness of the zero-watermarking system, the normalized

correlation coefficient (NC) of the extracted watermark is applied in conjunction to the
original one ; the maximum value of this measure is 1 which determines the best
robustness of the watermarking process.

(3)

Where: N is the size of the original and extracted watermark bits. After obtaining the post
attack NC values, we use them to evaluate the fitness of the initial population for the genetic
algorithm.

3.3. Genetic algorithm (GA) for optimal positions selection

GA is a nature inspired optimization technique based on evolution and natural selection of
individuals. To solve the problem of selecting the best positions to extract a robust features
matrix bits, the GA is used considering the following components:
1. Initial population: The chromosomes are randomly generated in the initial population.
The length of chromosomes is equal to the double of the size of the watermark i.e.
2(n×n). The initial population size contain of 30 chromosomes. The number of
generations was considered as 200 times. The best gene is considered as the locations
of the best features extraction.
2. Appling attacks: three types of attacks must be applied, JPEG compression with quality
factor equal to 10, 15% cropping and adding 5% salt and pepper noise.
3. Fitness function (FF): the FF can be defined as the function of robustness. For z kinds
of attack to the zero watermarked image, the FF is quantified as follows:

(4)

The objective is to maximize the fitness function to achieve the optimal robustness
for digital image zero-watermarking technique.
4. Perform the crossover and mutation on the selected population to generate the new
population. The probability of crossover is set to 0.95, and mutation rate is 0.5.
5. The GA cycle is repeated until the maximum number of generation is reached. The
population with the highest fitness value in the final generation is the optimal selected
positions.

4. Experimental results
To test the performance of the proposed optimal zero-watermarking technique, the results
of the experiments are obtained using MATLAB. The proposed Technique is tested on gray
scale images of size 512×512 (Lena, Peppers, Barbara and Boat). The gray image Lena is
shown in Figure 4 (a). The binary watermark image of size 32×32 is shown in Figure 4 (b).
The resultant Master and Secret Shares for the proposed technique are shown in Figure 4 (c),
and (d) respectively. The recovered watermark from an unaltered image is shown in Figure 4
(e), note that the size of all these images is same as the original watermark and looks like a
random scatter of black and white pixels.

354 Copyright ⓒ 2014 SERSC

 International Journal of Security and Its Applications
Vol.8, No.5 (2014)

(a) (b)

(c) (d) (e)

Figure 4. (a) The original gray scale image (Lena), (b) watermark image (32× 32),
the resultant images of the zero watermarking technique (c) Master Share, (d)
Secret Share and (e) Reconstructed watermark

To check the robustness of the proposed technique, some common image processing
attacks were performed on the original images. Noise addition to the image is obtained by
adding 1% and 5% salt and pepper noise to the original image. Gaussian noise is obtained
with mean=0 and variance=0.0005. Median filtering is performed with window size 3×3. And,
Gaussian low pass filtering of the image was done with a window of size 2×2. Gamma
Correction 1.5, Intensity Adjustment ([0 0.8], [0 1]), and histogram equalization are applied to
the original image. The JPEG compression attack is performed by compressing the image
with quality factors 15 and 10. The scaling of an image is done by first reducing the original
host image size from 512×512 pixels to 256×256 pixels, and then zoomed to its original size
by means of pixel replication. The cropped image is obtained by cropping 15% of the original
image.
To evaluate the proposed scheme fidelity, we measure the similarity between the original
and attacked images by using Peak Signal to Noise Ratio.

(5)

Where, the MSE is the mean squared error (MSE) between the original and the distorted
one, is defined as Equation (6).

(6)

Where: m×m is the size of the image, and I(i, j), I'(i, j) are the pixel values of the host and
the attacked images. Figure 5 demonstrates the values of PSNR for the gray scale images.
Figure 6 shows the NC values of the reconstructed watermark with random positions

Copyright ⓒ 2014 SERSC 355

International Journal of Security and Its Applications
Vol.8, No.5 (2014)

selection for the proposed zero-watermarking (without optimization) under different types of
attack, while Figure 7 shows the NC values with the proposed optimization technique.

40.5
36.5
32.5
Lena
28.5
24.5 Peppers
20.5
Barbara
Boat

Figure 5. The effect of attacks on PSNR values

1
0.99
0.98
0.97
0.96 Lena
0.95 Peppers
0.94
Barbara
Boat

Figure 6. The effect of attacks on NC values without optimization

1
0.9 9
0.9 8
0.9 7 Lena
0.9 6 Peppers
0.9 5
Barbara
Boat

Figure 7. The effect of attacks on NC values with proposed optimization

technique

356 Copyright ⓒ 2014 SERSC

 International Journal of Security and Its Applications
Vol.8, No.5 (2014)

Table 2 lists comparison results of NC values between the optimized and non optimized
zero-watermarking techniques. Table 2 shows the values of all NC obtained by each attack
through the passing generations of the GA algorithm in contrast to NC values of a zero
watermarking technique lacking the GA optimization ability.
As in Figure 8, the values of NC are shown for all 4 test images suffering from three
different attacks and showing how these values are improved across the progression of
generations in the GA algorithm, on the other hand, Figure 9 shows the impact of the
mentioned NC values improvement on the calculated fitness function, resulting a fitness
values showing a clear climbing curve towards optimum results. The exact digital values of
maximum fitness for the test images are 2.9733, 2.9775, 2.9806 and 2.9489 for Lena, Peppers,
Barbara and Boat respectively.

Table 2. Performance comparison between the proposed non-optimized and

optimized zero-watermarking techniques
Without optimization With optimization based on GA
Attack No.
types Lena Peppers Barbara Boat of Lena Peppers Barbara Boat
gen.
10 0.9833 0.9857 0.9845 0.9725
Salt & 50 0.9923 0.9947 0.9953 0.9851
Pepper 0.9815 0.9833 0.9821 0.9694 100 0.9923 0.9927 0.9957 0.9867
Noise 5% 150 0.9929 0.9941 0.9959 0.9851
200 0.9923 0.9947 0.9953 0.9851
10 0.9893 0.9845 0.9887 0.9797
JPEG 50 0.9935 0.9915 0.9951 0.9825
Comp. 0.9785 0.9791 0.9742 0.9541 100 0.9937 0.9941 0.9947 0.9893
QF=10 150 0.9943 0.9937 0.9943 0.9901
200 0.9931 0.9943 0.9955 0.9889
10 0.9701 0.9737 0.9803 0.9608
50 0.9879 0.9885 0.9927 0.9749
Cropping
0.9633 0.9591 0.9737 0.9491 100 0.9843 0.9889 0.9903 0.9789
15%
150 0.9861 0.9849 0.9869 0.9724
200 0.9879 0.9885 0.9898 0.9749

(a) (b)

Copyright ⓒ 2014 SERSC 357

International Journal of Security and Its Applications
Vol.8, No.5 (2014)

(c) (d)
Figure 8. The NC values for JPEG compression with quality factor equal to 10,
15% cropping and adding 5% salt and pepper noise versus GA generations for
(a) Lena , (b) Peppers, (c) Barbara and (d) Boat test images.

(a) (b)

(c) (d)
Figure 9. The Fitness Function values versus GA generations for (a) Lena , (b)
Peppers, (c) Barbara and (d) Boat test images

358 Copyright ⓒ 2014 SERSC

 International Journal of Security and Its Applications
Vol.8, No.5 (2014)

5. Conclusions
In this paper, we have proposed an optimal zero-watermarking technique that extracts the
feature bits from the most important parts in the host gray image instead of embedding
watermark into that image to protect the copyright information in a robust way. Its high
robustness is due to the exploitation of GA for the evaluated selection of the positions to
extract the feature bits. The feature bits are used to split the watermark into unexpanded
master and secret shares. The proposed technique generates a secret share to the watermark
from host image during the embedding procedure, which is constructed from that the DC
coefficient in discrete cosine transform domain of the host image. The secret share and the
optimal secret sequence are then registered in a third-party database for copyright protection.
The master share is constructed from the controversial gray image during the extracting
procedure. According to the obtained results of NC values that shows the proposed technique
is robust against different kinds of attack, such as: noise addition, Median filtering, Gaussian
low pass filtering, intensity adjustment, gamma correction, histogram equalization, scaling,
JPEG compression, and cropping.

Acknowledgements
This work was supposed by the project of National Science Fund of China
(No. 60873188).

References
[1] L. Jing and F. Liu, "Double Zero-Watermarks Scheme Utilizing Scale Invariant Feature Transform and Log-
Polar Mapping", IEEE International Conf. on Multimedia and Expo, Beijing, China, (2007) July 2-5, pp.
2118-2121.
[2] X. Tang, J. Wang, C. Zhang, H. Zhu and Y. Fu, "A Fast and Low Complexity Zero-Watermarking Based on
Average Subimage in Multiwavelet Domain", In Proc. of IEEE 2nd International Conf. on Future Computer
and Communication (ICFCC), Wuhan, China, vol. 2, (2010) May 21-24, pp. 178-182.
[3] G. G. Yong and J. G. Ping, "Zero-bit watermarking resisting geometric attacks based on composite-chaos
optimized SVR model", The Journal of China Universities of Posts and Telecommunications, vol. 18, no. 2,
(2011), pp. 94-101.
[4] M. Shakeri and M. Jamzad, "A Robust Copyright Protection Scheme Based on Discrete Cosine Transform
And Image Normalization", In Proc. of IEEE 7th Iranian Machine Vision and Image Processing (MVIP),
Tehran, Iran, (2011) November 16-17, pp. 1-6.
[5] Y. Zhou and W. Jin, "A Novel Image Zero-watermarking Scheme Based on DWT-SVD", In Proc. of IEEE
International Conf. on Multimedia Technology (ICMT), Hangzhou, China, (2011) July 26-28, pp. 2873-2876.
[6] L. Peili and T. Yuehui, "Robust Zero-watermarking Algorithm Based on Invariant Centroid", IEEE Fifth
International Conference on Computational and Information Sciences (ICCIS), Shiyang, China, (2013) June
21-23, pp. 758-761.
[7] M. Naor and A. Shamir, "Visual Cryptography", Advances in Cryptography Eurocrypt'94, Lecture Notes in
Computer Science, Springer-Verlag, Berlin, vol. 950, (1995), pp. 1-12.
[8] S. Punitha, S. Thompson and N. S. R. Lingam, "Binary Watermarking Technique based on Visual
Cryptography", IEEE International Conference on Communication Control and Computing Technologies
(ICCCCT), Ramanathapuram, India, (2010) October 7-9, pp. 232-235.
[9] F. Liu and C.-K. Wu, "Robust visual cryptography-based watermarking scheme for multiple images and
multiple owners", Information Security, IET, vol.5, no 2, (2011), pp. 121-128.
[10] B. Surekha and G. Swamy, "A Semi-blind Image Watermarking based on Discrete Wavelet Transform and
Secret Sharing", IEEE International conf. on Communication, Information & Computing Technology
(ICCICT), Mumbai, India, (2012) October 19-20, pp. 1-5.
[11] B. Surekha and D. G. Swamy, "A Spatial Domain Public Image Watermarking", International Journal of
Security and Its Applications, vol. 5, no. 1, (2011) January, pp. 1-12.
[12] P. V. Jithi and A. T. Nair, "Progressive visual cryptography with watermarking for meaningful shares",
International Multi-Conference on Automation, Computing, Communication, Control and Compressed
Sensing, Kottayam, India, (2013) March 22-23, pp. 394-401.

Copyright ⓒ 2014 SERSC 359

International Journal of Security and Its Applications
Vol.8, No.5 (2014)

Authors

Jumana Waleed is a Ph.D. student in the School of information

science and Engineering at Central South University, Changsha,
China. Her research activity focuses on image processing, and
information security working on digital watermarking. She received
the B.S. degree in computers sciences from the Al-Yarmouk
University College, Iraq, in 2004, and the M.S. degree in Computer
Science/Data Security from the University of Technology, Baghdad,
Iraq, in 2009.

Dr. Huang Dong Jun is a professor at Central South University.

He received his PHD degree in computer science and technology
from the Central South University, China, in 2004. He worked as a
visiting academic to University of Glasgow, UK, from 2007 to 2008.
Currently, he is the director of the Department of Computer
Engineering, Central South University and the member of the IOT
Education Expert Group of the Ministry of Education, China. His
research interests include computer networks, multimedia
technology, image and video processing, video conferencing system
and video surveillance techniques.

Saad Hameed was born in Baghdad – Iraq in 7 June 1979, he

received his B.Sc. Degree in computer science at AL-Mansour
University College 2001, and Masters Degree in computer Sciences,
Iraqi committee for computer and informatics in 2004, he continued
working in academic teaching in AL-Mansour University College
for 11 years, through that time all of his research was self-funded
and concentrated in automation and control, he has been promoted
from assistant Lecturer to Lecturer in 2011, he is now a Ph.D. degree
student at Hunan University in P.R. China.

360 Copyright ⓒ 2014 SERSC