Robust Digital Image Watermarking based on Hybrid DWT

and GWO Optimization Technique

Snehlata Maloo1 , Mahendra Kumar2 , Prof. N.Laxmi3 , Dr. N. K. Pareek4

1,4
Department of Computer Science, MLSU Udaipur (Raj.)
3
Department of Physics, MLSU Udaipur(Raj.)
2
Department of Electrical Engineering, IIT Roorkee, India
sneha.maloo21@gmail.com mireearchlab@gmail.com
nlakshmi@mlsu.ac.in npareek@yahoo.com

Abstract

This paper offers a novel optimization method for digital watermarking

in the Wavelet domain (DWT). Digital image watermarking has deter-
mined its proficiency in ensuring illegal confirmation of media. The
visuality factor of the watermark is the colossal parameter that helps in
enhancing the perceptual transparency and robustness of the water-
marking system. The trade-off between the openness and robustness is
considered as an optimization problem and is explained by applying
GWO Optimization technique. The PSNR, Mean Square Error (MSE),
and computational time are evaluated for an arrangement of images
using the MATLAB R2014b software.
Keywords: Computational Time, DWT, GWO, MSE, PSNR, Robustness,
Transparency,

1. INTRODUCTION

Digital watermarking is one of the several protection processes embed secret infor-
mation. The necessity for digital image copyright protection methods has turned into
an essential substance in multimedia applications because of the fast development of
illegal access and propagation of original digital objects like sound, video, and imag-
es. Embedding information in an agile and reliable way has lead to the application of
frequency domain techniques like discrete cosine or the discrete wavelet transforms.
Robust watermarking in spatial domain can be obtained at the cost of explicitly
demonstrating the local image characteristics. However, these highlights can be
achieved with much ease in the frequency domain.
The two main properties – robustness and imperceptibility are funda-
mental in saving the security of images from unauthorized access. The capacity to
detect the watermark of the image after use of standard signal processing distortions is
known as robustness. The decision of amplification factors can be shown as an opti-
2

mization problem and deal with using Genetic Algorithm. M. Ketcham et al., [9] have
proposed an original DWT watermarking scheme given Genetic Algorithms for audio
signs. The optimal localization and intensity were obtained using GA and the strategy
was discovered robust against editing, low pass filter and additive noise. Ali Al-Haj et
al. [11]
To illustrate an imperceptible and robust digital image watermarking
scheme based on a combination of DWT and DCT. Thus, Franco et al.[5], gave a
DWT based method for evaluation of fidelity and robustness. These algorithms were
equipped for separating the watermark. However, experienced the issues of unsuitable
values of loyalty and robustness to different attacks concentrated in these papers.
Zhicheng Wei et. al. [10] proposed an algorithm that yielded a watermark that is faint
to human eyes and robust to different image controls, and the outcomes demonstrated
that closed some particular areas were the best choice for embedding the watermark.
The authors applied GA to train the frequency set for inserting the watermark and
compared their approach and the Cox's method [8] to demonstrate robustness. The
analysis of GA was limited to JPEG compression attack in this technique. In [9], offer
a scheme that does not require the original image because the neural network memo-
rized the data from the shape particular points of the original image. This system ap-
plies the shape specific point's technique and highlights point matching method by
genetic algorithm for opposing geometric attacks. In [7] proposed another flexible and
efficient evaluation instrument based on genetic algorithms to test the robustness of
digital image watermarking techniques. Given an arrangement of possible attacks, the
method finds the ideal un-watermarked image regarding Weighted Peak Signal to
Noise Ratio (WPSNR). [11] Proposed an inventive watermarking scheme based on
genetic algorithms (GA) in the transform domain considering the watermarked image
quality.
In this paper, GWO is utilized to adaptively optimize the watermark vis-
ibility factor at each chosen DWT sub-band that will enhance the intangibility and
robustness of the watermark. The proposed technique uses the standardized correla-
tion of the cover image and the watermarked images as the reason for evaluating the
fitness function. The fitness function fills in as the objective function that will be
streamlined and looks through the population comprising of appropriate embedding
locations of the watermark within the cover image.

2. DIGITAL IMAGE WATERMARKING

The innovation of the DWT algorithm lives in the manner the robustness and the im-
perceptibility is enhanced the watermark image [5]. The significant goal of the wave-
let transform is to deteriorate the input image hierarchically into a progression of suc-
cessive low-frequency sub bands and their related definite sub bands.
The low-frequency sub band and the point by point sub-bands contain
the data required to remake the low-frequency approximation at the following higher
resolution level [9]. Wavelet techniques give such sort of a superb space and frequen-
 3

cy energy compaction and thus DWT has received a mind-boggling enthusiasm for a
several signal and image processing applications.
The watermark amplification factor is modulated based on the local im-
age attributes, in a pixel by pixel way. A significant portion of the DWT based wa-
termarking concepts concentrates on the sub-bands or block-based techniques, while,
here the watermark amplification factor is balanced pixel-wise. As an outcome, the
grey- level sensibility, iso-frequency masking, non-isofrequency masking, noise sen-
sibility and so on, are considered [5]. Because of the superb spatial-frequency locali-
zation property of DWT, it is easier to recognize the image areas in which, an disturb
can be hidden more probable [2]. Rather than the DFT/DCT watermarking tech-
niques, if a DWT coefficient is adjusted, only the area of the image corresponding to
that factor will be modified.

2.1 WATERMARK EMBEDDING

Let the image to be watermarked, initially decomposed through DWT into two levels.
Where 𝐵𝑙𝑥 denote the sub-band at level = 0,1,2,3 and the orientation x  {0, 1, 2,3} as
shown in Fig.1.

Fig.1. Decomposition of an image into two levels through DWT technique

The watermark is inserted into the three detail bands at level 0 by modifying the
wavelet coefficients. The choice of adding the watermark into this standard is based
on experimental tests such that the robustness and invisibility are compromised. The
result of insertion is weak, resulting in low robustness, but given the partial visibility
of disturbs added, a higher watermark amplification factor is allowed thus compensat-
ing for the high fragility.
The watermark information of dimension M x N is converted into uni-
dimensional antipodal sequenced(i,j)  {+1,-1}, where M and N represent the num-
ber of rows and columns. The image is partitioned into 2-levels and all the acquired
wavelet coefficients at the chosen subband are divided into n segments such that n =
MN.
I={H,V,D}
Here H, V and D refer to the horizontal, vertical and diagonal wavelets respective-
ly[12]. The DWT of a function of size M X N is then
4

M 1 N 1
1
 ( j0 , m, n) 
MN
  f ( x, y)
x 0 y 0
j0 , m , n ( x, y )

M 1 N 1
1
i ( j, m, n) 
MN
  f ( x, y)
x 0 y 0
i
j ,m,n ( x, y )
(1)
W ( j0 , m, n) is the approximation in starting scale j0 . W ( j, m, n) Means
i
The
the horizontal, vertical and diagonal details at scales
j  j0 . Where j0  0 ,
j  0,1, 2...... j  1 and m  n  0,1, 2,...2 j  1 . By Using. Eq.(1) and (2), the
f ( x, y ) obtained by the inverse DWT
1
f ( x, y ) 
MN m n
W ( j0 , m, n) j0 ,m,n ( x, y)

1
   
MN i  H ,V , D j  j0 m n
Wi ( j , m, n) ij ,m ,n ( x, y )
(2)
When a 1-level DWT is applied, the image is decomposed into four parts of high,
middle and low frequencies-LL1, HL1, LH1 and HH1 subbands. The subbands la-
beled HL1, LH1 and HH1 represent the more exceptional scale wavelet coefficients.
The LL1 subband contains the approximate image.

Fig 2: (a) Watermark Embedding Process (b) Watermark Extraction Process

2.2 WATERMARK DETECTION

The DWT approach is a blind process, and further does not require the original image
for watermark detection. DWT is utilized on the watermarked image, and the sub-
band to which the watermark was embedded is chosen. The correlation between the
cover and the extracted watermark is then processed as
N

 I .I '
 i 1

N N

 I .2
i 1
 .I '
i 1
2
 5

Where I denote the original and I’ stands for the extracted water-
marks. Each computed correlation value is compared with the average correlation. If
it happens that the calculated value is more prominent than the mean, then the ex-
tracted watermark bit is considered as 0, else if the computed value is lesser than
way, it is taken as 1 [11]. In the end, the watermark image is restructured using the
extracted bits, and the likeness between the original and the watermarked image is
described. Innovative multimedia and watermark are the inputs to the algorithm.

3. Grey Wolf Optimizer (GWO)

A. About Grey Wolf: Grey wolf optimization is a new Meta heuristic algorithm
proposed for solving many multi-model functions. Grey wolves inspire it. Four
types of grey wolves such as α, β, δ, and ω are employed to derive the leadership
of hierarchy of grey wolves. The main steps are hunting, searching for prey, en-
circling prey and attacking prey.

B. Wolf behavior in nature:

Social behavior: Hierarchy exists in the pack. α is the decision maker and the
leader. β and δ assist α in decision making. Rest of the wolves(ω) are as fol-
lows.
Encircling prey: As referred above, grey wolves encircle prey during the hunt.
To encircling behavior, the following equations are proposed.
Where: t- shows the current iteration,
 
A & C - are coefficient vectors,

X P - is the position vector of the prey,

X - represents the position vector of a grey wolf.
 
The vectors A and C are computed as follows:
  
A  2a  r1  a
 
C  2  r2 ……. (3)

a
where components of are linearly decreased from 2 to 0 over the course of itera-
tions and
r1 , r2 are random vectors in [0,1]

Hunting behavior: Group hunting behavior is of equal interest in studying optimization.

1. Tracking, chasing and approaching the prey.
2. Pursuing, encircling, and harassing the prey until it stops moving.
3. Attacking the prey.

Grey wolves can acknowledge the location of prey and encircle them. The alpha usu-
ally guides the hunt. The beta and delta might also participate in hunting occasionally.
However, in an abstract search space, we have no idea about the location of the opti-
mum (prey). To mathematically simulate the hunting behavior of grey wolves, we
6

suppose that the alpha, (best candidate solution) beta and Delta have better knowledge
about the potential location of prey. Therefore, we save the first three sound solutions
obtained so far and oblige the other search agents (including the omegas) to update
their positions according to the location of the best search operator. The undermen-
tioned formulas are given in this regard.

   
D  C1  X   X
   
D  C 2  X   X
…….(4)
   
D  C3  X   X
   
X 1  X   A1  ( D )
   
X 2  X   A2  ( D )
…….(5)
   
X 3  X   A3  (D )
  
 X1  X 2  X 3
X (t  1) 
3 …….(6)

4. IMPLEMENTATION

In Digital Watermarking, the populace is instated by choosing an arrangement of

random positions in the host image and embeddings the watermark image into the
selected area. The ideal answers for digital watermarking using DWT are acquired
based on two key factors: 1) The DWT sub-band and 2) The coefficient of the water-
mark amplification factor [11]. The GA algorithm searches its population for the ideal
preparation with all possible combinations of the DWT sub-bands and watermark
amplification factors. The genetic algorithm methodology will attempt to locate the
particular sub-band that will give concurrent perceptual transparency and robustness.
To enlarge the robustness of the algorithm against attacks, the watermark strength or
the amplification factor α ought to be improved, yet this factor differs on each sub-
band.
The objective function is also known as the fitness function is a combination of
the Peak Signal to Noise Ratio (PSNR) and the correlation factor ρ (α * NC) and is
given as,Fitness function = PSNR + 100 *  , Where PSNR is computed as,
 MAX i2 
 
 MSE 
PSNR=10log  
Where, MSE between the original and watermarked image and MAXi =
the maximum pixel value of the image which is 255 in the experiment since pixels
were represented using 8 bits per sample.
 7

Here, the correlation factor (CF) is the product of Normal Correlation

(NC) and the watermark strength factor α. The fitness function increments propor-
tionately with the PSNR. However, NC is the key factor adding to the robustness and
eventually, the fitness value increments with the robustness measure. The CF ρ has
been increased by 100 since its normal values fall in the range 0 ~ 1, whereas PSNR
may achieve the value of 100. The flow-chart of the methodology is additionally illus-
trated in Fig.4.

Watermark image

Watermark PSNR Compu-

Original Image Embedding tation
using DWT

Compute Fitness Evalua-

correlation tion

Watermark Selection of best visi-

Extraction bility factor
using DWT

Fig 4: Watermark embedding and extraction Process using DWT

5. Simulation results:
To test the watermarking method, several conventional signal processing techniques and
geometric distortions were applied to these images to evaluate if the detector can reveal the
presence of the image owner’s watermark. This way, we can measure the algorithm robust-
ness to various kinds of attacks. In this section, we are displaying experimental results ob-
tained on the “bike” and “cherry” images as in Original image is bike.jpg as shown fig. 5.
And Embedding watermark image cherry.jpg as shown fig. 6.

Fig.5: Original image Fig.6: Watermark image

8

Table 1: Proposed watermarking Technique

S.No. Algorithm MSE PSNR Alpha

1. GA-3 level DWT 0.0332 34.2261 0.2042
2. GWO-2 level DWT 0.0024 61.1483 0.2086
Table 2: Proposed watermarking Technique with attacks

S.No. Attacks MSE PSNR Alpha

1. Without 0.0024 61.1483 0.2086
2. Compression with q=90 0.0174 41.1765 0.2086
3. Compression with q=80 0.0195 40.0417 0.2086
4. Compression with q=50 0.0247 37.7024 0.2086
5. Compression with q=30 0.0292 36.0357 0.2086
6. Salt and pepper (0.001) 0.0218 38.9543 0.2086
7. Salt and pepper (0.005) 0.0514 30.3668 0.2086

6. CONCLUSION

This work gives Digital image watermarking, which is based on 2 Level Discrete
Wavelet Transform (DWT) with the GWO, Table1 demonstrates the effectiveness of
proposed technique as PSNR, MSE, Fitness value at generation=10 with visibility
factor=0.2042 for watermarked image and extract the watermark.

REFERENCES
1. Sikander B., Ishtiaq M., Jaffar M.A., Tariq M., Mirza A.M(2010): Adaptive Digital Wa-
termarking of Images Using Genetic Algorithm. IEEE International Conference on Infor-
mation Science and Applications, pp. 1-8.
2. Jagadeesh, B., Kumar, S. S., & Rajeswari, K. R. (2009, December). A genetic algorithm
based oblivious image watermarking scheme using singular value decomposition (SVD).
In Networks and Communications, 2009. NETCOM'09. First International Conference
on (pp. 224-229). IEEE.
3. Tang, C. W., & Hang, H. M. (2003). A feature-based robust digital image watermarking
scheme. IEEE transactions on signal processing, 51(4), 950-959.
 9

4. RGPV, B. Application of Genetic Algorithm to Optimize Robustness and Fidelity of Wa-

termarked Images (A conceptual approach).International Journal on Computer Science and
Engineering, Vol. 1, No. 3, pp. 239-242, (2009).
5. Ketcham, M., & Vongpradhip, S. (2007). Intelligent audio watermarking using genetic al-
gorithm in DWT domain. International Journal Of Intelligent Technology, 2(2), 135-140.
6. Cox I., Miller M., Bloom J. : Digital Watermarking: Principles and Practice. San Mateo,
CA: Morgan Kaufmann, (2001).
7. Al-Haj, A., & Abu-Errub, A. (2008). Performance Optimization of Discrete Wavelets
Transform Based Image Watermarking Using Genetic Algorithms 1.
8. Mahendra Kumar et. al.: Digital image watermarking: A survey. International Journal of
Engineering and research applications (IJERA), Jul-Aug, (2013).
9. Mahendra Kumar et.al.: Digital Image Watermarking using Fractional Fourier transform
via image compression. In : IEEE International Conference on Computational Intelligence
and Computing Research(2013 IEEE ICCIC), 26-28 Dec., (2013).
10. Mahendra Kumar et. al.: Digital Image Watermarking using Fractional Fourier Transform
with Different Attacks. International Journal of Scientific Engineering and Technology,
Volume No.3 Issue No.8, pp : 1008-1011. Aug. (2014), ((ISSN : 2277-1581))
11. Surekha, P., & Sumathi, S. (2011). Implementation of genetic algorithm for a dwt based
image watermarking scheme. ICTACT Journal on Soft Computing: Special Issue on Fuzzy
in Industrial and Process Automation, 2(1), 244-52.
12. Shahid, A., Badsha, S., Kabeer, M., Ahsan, J., & Mahmud, M. (2012). A Non-Blind Wa-
termarking Scheme for Gray Scale Images in Discrete Wavelet Transform Domain using
Two Subbands. arXiv preprint arXiv:1211.2699.