An adaptive watermarking scheme for e-government

document images
Shi-Jinn Horng & Didi Rosiyadi & Pingzhi Fan &
Xian Wang & Muhammad Khurram Khan
Published online: 20 July 2013
#
Springer Science+Business Media New York 2013
Abstract This paper proposes an adaptive watermarking scheme for e-government document
images. The adaptive scheme combines the discrete cosine transform (DCT) and the singular
value decomposition (SVD) using luminance masking. As a core of masking model in the human
visual system (HVS), luminance masking is implemented to improve noise sensitivity. Genetic
algorithm (GA), subsequently, is employed for the optimization of the scaling factor of the
masking. Involving a number of steps, the scheme proposed through this study begins by
calculating the mask of the host image using luminance masking. It is then continued by
transforming the mask on each area into all frequencies domain. The watermark image, following
Multimed Tools Appl (2014) 72:30853103
DOI 10.1007/s11042-013-1579-5
This work was supported in part by the National Science Council under contract number NSC-99-2916-I-011-
002-A1, and it was also partially supported by the 111 Project under the grant No. 111-2-14 and One Hundred
Person Project 2012, Sichuan Province.
S.<J. Horng
:
P. Fan
:
X. Wang
Institute of Mobile Communications, Southwest Jiaotong University, 610031 Chengdu, China
P. Fan
e-mail: pingzhifan@gmail.com
X. Wang
e-mail: drwangxian@gmail.com
S.<J. Horng
:
D. Rosiyadi
Department of Computer Science and Information Engineering, National Taiwan University of Science
and Technology, Taipei 106, Taiwan
D. Rosiyadi
e-mail: didi.rosiyadi@gmail.com
M. K. Khan
Center of Excellence in Information Assurance, King Saud University, Riyadh, Kingdom of Saudi Arabia
e-mail: mkhurram@ksu.edu.sa
D. Rosiyadi
Research Center for Informatics, Indonesian Institute of Sciences (LIPI), Jakarta, Indonesia
e-mail: rosiyadi@informatika.lipi.go.id
S.<J. Horng (*)
43, Sec. 4, Kee-Lung Rd., Taipei, Taiwan 106, Republic of China
e-mail: horngsj@yahoo.com.tw
this, is embedded by modifying the singular values of DCT-transformed host image with singular
values of mask coefficient of host image and the control parameter of DCT-transformed water-
mark image using Genetic Algorithm (GA). The use of both the singular values and the control
parameter respectively, in this case, is not only to improve the sensitivity of the watermark
performance but also to avoid the false positive problem. The watermark image, afterwards, is
extracted from the distorted images. The experiment results show the improved adaptive perfor-
mance of the proposed scheme is in resistant to several types of attacks in comparison with the
previous schemes; the adaptive performance refers to the adaptive parameter of the luminance
masking functioned to improve the performance or robustness of an image from any attacks.
Keywords Adaptive
.
Watermarking
.
Luminance masking
.
Human visual system
.
E-government
1 Introduction
Implemented to several multimedia elements such as image, audio, and video; digital
watermarking has been widely functioned as access control, copyright information, authenti-
cation, protection of the intellectual property rights, fingerprint, tamper proofing, and annota-
tion. Watermarking based on the type of the embedded information can be classified into three
categories - non-blind, semi-blind and blind. The non-blind schemes require both the original
image and the secret key(s) for watermark embedding; a semi-blind scheme requires both the
secret key(s) and the watermark sequences, and the blind one can only use the secret key(s).
Furthermore, the adjustment of watermarking strength is to obtain a minimum perceptual
distortion that is by employing the human visual system (HVS) model. For this, robustness,
imperceptibility, and security are three requirements necessary to be fulfilled by an adaptive
digital watermark. For the robustness, the watermark should remain in the host image
regardless of the possible processing that the image might be undergone. For the
imperceptibility, the watermark, meanwhile, should be invisible under certain typical or
specified viewing conditions - reflecting the incapability of human eyes in distinguishing the
watermarked media from the original one. At last, security means the incapability of the
unauthorized parties to recover the watermark even if for the known data embedding or
extraction techniques. In most of systems, this technique could be achieved by using secret
keys known to the authorized user. In the previous researches, the secret key, however, has
been used to scramble the watermark. In addition, a digital watermark should also be robust
against any image disturbance such as noise that could be a serious problem in a multimedia
element (video, audio, and image). Hence, this topic is widely studied [4, 6, 1215, 24, 25, 27, 38].
There have been many schemes proposed in this field. Verma and Tapaswi [31] proposed a
noise sensitive region based watermarking embedding technique that is to embed the watermark
in the given region of image. Kuo and Johnston [20], meanwhile, employed a spatial noise
shaping (SNS) method with human visual sensitivity to exploit the capability of a frequency
domain linear prediction for the spatial envelope retrieval. This algorithm shapes or hides the
noise of an image well in areas and is insensitive to human vision - leading it to be more pleasant
for human eyes. Furthermore, Xue and Teng [35] implemented a human visual sensitivity to
determine the weight for the masking. In their method, the embedding procedure has been
operated in an intermediate domain in which the noise sensitivity can affect the coefficients
value. Li et al. [22] presented a hybrid DWT-SVDdomain-watermarking scheme by considering
human visual properties. Here, the host image is in the decomposition of four sub-bands. SVD,
3086 Multimed Tools Appl (2014) 72:30853103
in this case, was applied to each sub-band and embedding the singular values of the watermark
with human visual model., Bao and Ma [3] proposed an image authentication scheme based on
DWT-SVD by applying a simple quantization-index-modulation process. To improve the
fidelity and the perceptual quality of the watermarked image and to enhance the security of
watermarking, Here, a bit embedded on the singular value (luminance) of the blocks within the
wavelet sub-band of the original image and model, the adaptive quantization parameters based
on the statistics of blocks within sub-band has been used.
Voloshynovskiy [32] described about a number of attacks on digital watermarking
consisting of classification, estimation-based attacks and benchmarks. Here, benchmarks are
to evaluate the performance of different watermarking algorithms. In [9], Cui and Li proposed
an adaptive scheme based on a discrete multiwavelet including a multi bit, multiplicative,
spread spectrum, unbalanced and balanced multiwavelet. This scheme combines various
masking effects of human visual perception by considering the eyes sensitivity to noise
changes depending on spatial frequency, luminance and texture of all the image sub-bands.
Mairgiotis et al. [25] presented a new family of watermark detectors for additive watermarks in
digital images. These detectors are based on a recently proposed hierarchical and two-level
image model found to be beneficial for some problems in image recovery. They also proposed
an estimation method to determine the necessary parameters for these detectors. In [36], Yan et
al. proposed a block-based adaptive image-watermarking scheme using Just Noticeable Dif-
ference (JND). The scheme has been performed in both pixel domain and DCT domain. It
begins by dividing the host image into 88 blocks and, afterwards, calculating the capacity of
each block to embed information without being perceptible. Here, the capacity is determined by
JND threshold with one feature of Human Visual System (HVS) model.
He and Chang [12] proposed an adaptive scheme with the characteristic of Human Visual
System (HVS) and the association memory ability of neural network. The watermarking signal
embedment is through a higher frequency in the lower frequency of an original image by DWT
joined with DCT. Himanshu [13] presented an adaptive scheme based on DCT-SVDby the human
visual model to mask the original image. For watermark embedding mask of the original image is
computed using HVS model and original image is transformed into frequency domain using DCT.
And then embed watermark by modifying the singular values of DCT coefficient, by singular
values of mask coefficient of original image and singular values of watermark logo. Zhu and Liu
[39] presented an adaptive model with the Human Visual System (HVS) characteristics based on
DWT-DCT. In this model, the original visually meaningful binary watermark image has been
preprocessed by using the extended Arnold transform. The secret watermark bits are embedded on
the singular values of the blocks within a lowfrequency of sub-band in host image. Amiri et al. [1]
introduced an adaptive multi-resolution watermarking approach in DWT domain. Here, a multi-
resolution representation of a binary watermark has been inserted into the selected coefficients of
the wavelet-decomposed image - representing the high activity regions of the image. The co-
efficients with maximum local variance in the lowest frequency sub-band and that of with
maximummagnitude in the higher frequency sub-bands have been selected here. The watermarked
image is transparent according to the characteristics of human vision system and robust against
progressive wavelet image coding - even at very low bit rates. The watermark extraction process is
in the completely blind and multiple spatial resolution levels of the watermark.
In [18], the watermarking channel has been modeled as a generalized channel with fading
and nonzero mean additive noise. To improve the watermark robustness against the gener-
alized channel, an optimized watermark extraction scheme has been presented using an
adaptive receiver for the quantization-based watermarking. Hence, for extraction scheme it is
by adaptively estimating the decision zone of the binary data bits and the quantization step.
Barni et al. [4] presented an adaptive scheme based on a wavelet domain through a new
Multimed Tools Appl (2014) 72:30853103 3087
approach to mask the watermark based on the characteristics of the human visual system
(HVS). Lai et al. [21] had proposed a novel image watermarking scheme using singular
value decomposition (SVD) and micro-genetic algorithm (micro-GA). The singular values
of the cover image are modified by multiple scaling factors to embed the watermark image.
The proper values of scaling factors are optimized and obtained efficiently by means of the
micro-GA. Carballeira [7] analyzed the performance of the geometric distortion incurred
when coding depth maps in 3D Video. Several geometric distortion techniques consist of
capture the geometric distortion caused by the depth coding error and the pixel-mapping
precision in view synthesis. Naiss [26] had proposed computable bounds for rate distortion
with Feed forward for stationary and Ergodic sources, which derive a sequence of achievable
and computable rates that converge to the feed-forward rate distortion.
Through a comprehensive review towards a number of previous researches, it is found that
there is no any research using the GA based hybrid DCT-SVD technique in this case by adding
an adaptive parameter in order to improve the performance of e-government document images
towards any attacks. In particular, in the comparison with the reference [28], the proposed
scheme has a novelty in the matter of adding an adaptive parameter of luminance masking.
This paper, in turn, aims to propose an adaptive watermarking scheme for e-government
document images. The proposed scheme combines the discrete cosine transform (DCT) and
singular value decomposition (SVD) using luminance masking in accordance with the charac-
teristics of the human visual system (HVS) on the improved noise sensitivity. Furthermore, an
optimization process of the scaling factor is conducted based on the genetic algorithm (GA).
The rest of this paper is organized as follows. Section 2 is to present the preliminaries of
the Human Visual System (HVS), discrete cosine transform (DCT), and singular value
decomposition (SVD), followed by Section 3 to explain about the proposed scheme.
Furthermore, Section 4 and Section 5 present the experiment results and the conclusion of
this paper, respectively.
2 Preliminaries
This section provides an overview about Human Visual System (HVS), discrete cosine
transform (DCT), and singular value decomposition (SVD).
2.1 Human visual system (HVS)
The goal of image applications is to aid a human observer to perceive the visual information
in an image. Hence, understanding human visual system (HVS) comes to be important. As
one of the most complex biological systems, HVS includes three stages; encoding, repre-
sentation, and interpretation and mainly consists of the eye (image sensor or camera), optic
nerve (transmission path), and brain (image information processing unit or computer) [4, 13, 23].
Based on the features of the image in HVS, the masking model is composed of
luminance, texture, and edge masking [4, 13, 30, 35]
1. The luminance masking is the measure of the darkness and brightness of the image. The
visibility of luminance threshold in the spatial domain depends on two factors: average
background luminance and non-uniformity of the background luminance.
2. The texture masking shows a relationship between an individual pixel value and the
mean value of the group pixels inside the sliding window [9, 13]. This is a key element
to represent whether the texture of the image is high activity or uniform.
3088 Multimed Tools Appl (2014) 72:30853103
3. Edge masking models represent the regions that have a number of abrupt changes in the
pixel values. Edges carry information content in the image. In image processing, edge
detection refers to a process or an operation that can distinguish the edge regions from
the non-edge ones.
2.2 Discrete cosine transform (DCT)
The discrete cosine transform (DCT) is a technique for converting a signal into elementary
frequency components. Discrete cosine transform (DCT) turns over the image edge to make the
image be transformed into the form of even function. It is one of the most common linear
transformations in digital signal process technology. The two-dimensional DCT transformation is
the development of one-dimensional DCT transformation. This transformation is used to convert
two-dimensional coordinates into frequency components. The formula can be seen as follows:
c r; s r : s

x;y0
N1

f x; y :cos
2x 1 r
2N
_ _
:cos
2y 1 s
2N
_ _ _ _
1
The formula of Inverse DCT is
f x; y

x;y0
N1

r : s :c r; s :cos
2x 1 r
2N
_ _
:cos
2y 1 s
2N
_ _ _ _
2
The output of the DCT function is the value of a particular frequency component, and the
output of this function is determined by two parameters, namely r and s. The two dimensional
DCT can not only concentrate the main information of original image into the smallest low-
frequency coefficient, but it also can cause the image blocking effect being the smallest, which
can realize a good compromise between the information centralizing and the computing
complication. DCT obtains the widespread application in the compression coding [8, 16].
Input and output from the function of DCT are the matrix of size NxN. f(x,y) is the value
of pixel at the coordinate (x,y), index is started from 0. (r) and (s) are the functions to take
back the value per square root if r and s=0 and take back value 1 if r and s is not equal to
zero. For the calculation of cosine, it is done in the modus of radian (not degree). Mean-
while, variable N refers to variable stating the size of matrix. In the formula of inverse DCT,
c(r, s) is from the result of formula DCT. The output of the function of inverse DCT is f(x,y),
the value of pixel in the coordinates (x, y).
2.3 Singular value decomposition (SVD)
In linear algebra, the singular value decomposition (SVD) is an important factorization of a
rectangular real or complex matrix, with many applications in signal processing and statistics
[21, 35]. We will diagonalize it, but not by S
1
AS. Its row space is r-dimensional (inside R
n
).
Its column space is also r-dimensional (insideR
m
). We are going to choose special ortho-
normal bases v
1
,......., v
r
for the row space and u
1
,......., u
r
for the column space. For those
bases, we want each Av
1
to be in the direction of u
1
. In matrix form, these equations :
Av
1

i
u
i
3
become
AV U 4
Multimed Tools Appl (2014) 72:30853103 3089
or
A UV
T
5
Where U and Vare unitary (orthogonal), is the singular value of A. =diag(
1
,
2
,....,
m
).
m is the rank of matrix A, so A = U
1
.
1
.V
1
+........+U
m
.
m
.V
m
SVD is a powerful numerical analysis tool, having several applications in scientific
computing, especially in signal and image processing areas. The prominent advantages of
SVD in image processing applications can be seen in [21, 35]:
& The singular values (SVs) of an image are stable, i.e. when small disturbances are added
to an image, the singular values still remain intact.
& The SVs of an image have algebraic image properties which are intrinsic and not visual.
2.4 E-government information security
Until now, the internet is still one of the communication media necessary for daily life,
including in the field of governance. The internet is not only a source of information but also
a tool in implementing the duties of state administration. Thus, the use of internet or e-
government can initiate the transition from a traditional government, which still uses manual
methods, to a modern government with electronic tools.
The objectives of e-government development are the forming of an information network
and the transactions of the public services that have quality and scope which can meet the
needs of society and can be accessed from the whole regions at any time and at a reasonable
cost. Additionally, essential is the forming of interactive links with corporations to increase
the growth of the national economy and strengthen the ability to face change, and the
emulation of international trade, forming the mechanism and communication channels with
state institutions [19, 29, 33, 34].
To safeguard important information in the government requires e-government informa-
tion security. Security is critical in their successful implementation for e-government and
transaction-based services. Some of the security issues in e-government are discussed below
[37]:
& Confidential information should not be accessible to unauthorized users.
& Authenticity, when information is received, it should be verified by a person or a project
claiming to be originator and vice versa.
& Integrity on retrieval or received at the other end of a communication network, the
information should appear exactly as it was stored or sent.
& Non-repudiation after sending/authorizing a message, the sender should be unable to
deny at the time of having done so.
It is compulsory to protect unauthorized copy, misappropriation and misrepresentation of
digital information in an e-government system. There are various criteria of security
assessment of ICT based networks. One of the most important criteria is the European
Information Technology Security Evaluation Criteria (ITSEC), which assists the security of
information communication and technology based network [38]. This criterion covers three
basic threats:
& Confidentiality (revealing unauthorized information)
& Integrity (unauthorized data modification)
& Availability (unauthorized withholding of information or resources)
3090 Multimed Tools Appl (2014) 72:30853103
Therefore, to prevent and counter these threats, there must be a new technology that will
protect the integrity of digital information and safeguard intellectual property rights. This
new technology is called digital watermarking. Digital watermarking is required in e-
government to track print document sources, tamper proofing and assessment, copy control,
finger printing, etc.
3 Proposed scheme
This research presents a DCT-SVD scheme using luminance masking in accordance with the
characteristics of the human visual system (HVS) on improved noise sensitivity. An opti-
mization process of the scaling factor, additionally, is conducted based on genetic algorithm
(GA). An advantage of DCT-SVD schemes is not only for being resistant to various attacks
but also for being able to improve a document images performance, security, and robustness
[28]. Furthermore, in this research the performance for the noise sensitivity is conducted
using the luminance method and for the optimization process it is using a GA to find the
optimization scaling factor of the watermark image for the DCT-SVD technique. Luminance
masking is a core of masking model in HVS. Noted here, this component has been already
used in a number of previous researches [9, 20, 22, 31].
The proposed scheme consists of several steps of watermarking embedding and extrac-
tion. In watermark embedding it begins by calculating the mask of the host image by
luminance masking and transforming it in all frequencies on each area a1, a2, a3 and a4,
respectively. The transformed host image is shown in Fig. 1. The DCT-transformed host
image consists of 2561024 coefficients, which are divided into four rectangular areas a1,
a2, a3, and a4, respectively. Each area has 65,536 (256256) coefficients. The areas are
decomposed into16 blocks, each with 4,096 (64 64) coefficients.
Fig. 1 The mapping of the transformed host image
Multimed Tools Appl (2014) 72:30853103 3091
The watermark image later is embedded by modifying the singular values of DCT-transformed
host image with singular values of mask coefficient of host image and the control parameter of
DCT-transformed watermark image using Genetic Algorithm (GA). In turn, both the singular
values and the control parameter are respectively used to improve the sensitivity of watermark
performance and to avoid the false positive problem. The next step is to extract the watermark
image from the distorted images. It is found from the experiment that the improved adaptive
performance of the proposed scheme is in resistant to several types of attacks as compared to the
previous existing schemes. Figure 2 below shows a block diagram of the proposed scheme.
Figure 2 refers to the improvement from reference [28] in which the parts that have been
revised come to the addition of the calculation of luminance masking in accordance with
Eq. 6. Through this additional calculation, the computation time complexity is relatively
fixed, compared to the reference [28] since in lm it only requires one subtractive calculation,
one divisional calculation and two multiplicative calculations
The time complexity of the proposed scheme for the GA will highly rely on several
factors including genetic operators, their implementation (which may have a very significant
effect on overall complexity), the representation of the individuals and the population, and
the fitness function. In the proposed scheme, the common choices consists of one point
crossover, point mutation, and roulette wheel selection for the Genetic Algorithm; the time
complexity is O(NGen(NPopNumGen+NPopNumGen+NPop)), where NGen is the number
of generations, NPop is the population size and NumGen is the size of the individuals.
Therefore, the total time complexity is then O (NGenNPopNumGen)).
Document Header
Image
Inverse map and
IDCT
Watermark Image
DCT
GA-based
optimization
process
Extract The
singular value
Embedding
(Singular Values)
SVD
DCT and map
DCT into 4 sub
blocks
Comparison
Watermark
DCT and map
DCT into 4 sub
blocks
SVD
Inverse map and
IDCT
Referenced
Watermark
(a) (b)
Calculate the lm
function
Watermarked
Image
Extracted
watermark image
Fig. 2 The block diagram of proposed watermarking technique: a Watermark embedding, b Watermark extraction
3092 Multimed Tools Appl (2014) 72:30853103
3.1 The luminance masking based on HVS
In the scheme proposed in this research, the luminance masking is applied to grayscale
digital image. The luminance masking is defined as follows.
lm
I x; y 128
128

; 6
where lm is the luminance masking, I(x,y) is the background luminance of the pixel (x, y).
The concept of luminance masking shows that the human eye is less sensitive to distortion
dependent on regions either with high brightness or with high darkness compared to mid-grey
regions with the most noticeable distortion. An 8-bit grayscale value is composed of 0 and 255.
In this case, the grayscale value of 0 is the total darkness or the lowest luminance, whereas the
maximum 8-bit grayscale value of 255 is bright white or the highest luminance [4, 5, 13, 35].
3.2 Watermark embedding process
The first step of the watermark-embedding process, reading the host image I and watermark
W, and then calculating the lm function corresponding to I. Assume the size of visual
watermark is n x n, and the size of the cover image is 2n x 4n. In step two, apply DCT to
host image I entirely; I
dct
= f
dct
(I), where I
dct
is a function of the DCT-transformed host
image I and consists of r rows and c columns.
In step three, scan the DCT coefficients into 4 areas a1, a2, a3 and a4 respectively in the
zigzag manner:
z f
zigzag
I
dct
7
Let l be the length of DCT- transformed host image in each area, where l
rc
4
. Let I
dct
a
be
an array to store all DCT coefficients of DCT-transformed host image in each area, where
I
dct
a
= f
convert_two
(z(l(a1)+1:(al))), z is obtained from Eq. (7), convert_two is a function to
convert the one dimensional array z into a two dimension array of size r
c
4
, and a=1, 2, 3, 4.
In step four, perform SVD operation on the host image I
dct
a
in each area of a1, a2, a3 and
a4, respectively.
I
a
dct
U
a
dct
S
a
dct
V
aT
dct
8
where U
dct
a
S
dct
a
V
dct
aT
= f
svd
(I
dct
a
) and a=1, 2, 3, 4.
In step five, perform SVD operation on the watermark image f
dct
(W) to obtain a DCT-
transformed watermark image.
W
dct
U
w
dct
S
w
dct
V
T
w
dct
9
where U
w
dct
S
w
dct
V
T
w
dct
f
svd
f
dct
W .
In step six, the algorithm in [23] is improved by multiplying the control parameter of the
watermark to get the principal component of the watermark as defined:
I
w
dct
U
w
dct
S
w
dct
10
In step seven, the algorithm in [28] is improved by modifying the singular values of DCT-
transformed host image with singular values of mask coefficient of host image and the
control parameter of DCT-transformed watermark image.
S
0
a
dct
lm
a
I
w
dct
S
a
dct
11
Multimed Tools Appl (2014) 72:30853103 3093
In step eight, for each area, decide the optimal value of the scaling factor key using
Genetic Algorithm.

u
i
l
i

w
; l
i
; u
i
_ _
12
where
w
average
w
is the average or mean of the singular values of S
w
dct
, l
i
is the
minimum scaling factor key and u
i
is the maximum scaling factor key.
In step nine, obtain the modified DCT coefficients in each area.
I
0
a
dct
U
a
dct
S
0
a
dct
V
aT
dct
13
In step ten, modify the coefficient back to their original positions and performing IDCT.
I
0 0
a
dct
f
dezigzag
I
0
a
dct
_ _
14
3.3 Watermark extraction process
The first step of the watermarking extraction, apply DCT to the watermarked image I,
I
dct

=f
dct
(I), where I
dct

consists of r rows and c columns. In step two, scan the DCT

coefficients into 4 areas a1, a2, a3 and a4 respectively in the zigzag manner.
z f
zigzag
I
0
dct
_ _
15
Let l

be the length of DCT- transformed host image in each area, where l

0

r
0
c
0

4
. Let I
dct
a
be
an array to store all DCT coefficients of DCT-transformed host image in each area, where
I
dct
a
= f
convert_two
(z(l(a1)+1:(al))), z is obtained from Eq. (15), convert_two is a function to
convert the one dimensional array z into a two-dimension array of size r
0

c
0
4
, and a = 1, 2, 3, 4.
In step three, perform SVD operation on the DCT-transformed host image.
I
0
a
dct
U
a
dct
S
0
a
dct
V
aT
dct
; a 1; 2; 3; 4 16
In step four, the algorithm [13] is improved as follows: Subtract the possibly distorted
watermarked image in each area with the original transformed areas, I
dct
a
= A
dct
a
I
dct
a
, Obtain
the distorted principal component for each area, I
0
a
w
dct

1
_

a
U
z
I
a
dct
V
_ _
.
In step five, construct the DCT coefficients for 4-visual watermarks and apply IDCT.
W
0
a
dct
I
a
w
dct
V
aT
w
dct
17
3.4 Optimization process by GA
In this paper, continuing from the previous research [2, 10, 11, 13, 17, 28], Genetic
Algorithm (GA) is implemented as described in Eq. 12 in order to find an optimum scaling
factor. The GA process is described in the following.
a. A function
u
i
l
i

w
; l
i
; u
i
_ _
as shown in Eq. 12 is created to find the scaling factor .
3094 Multimed Tools Appl (2014) 72:30853103
b. The initialization process is to create a population of chromosomes and to initiate the
vectors randomly for chromosomes in each block.
c. The Evaluation step is to evaluate the fitness of all chromosomes in population. The
fitness value f
i
: f
i
Chromosome i Avgw
2

1
1Chromosome i
d. The crossover stage is to mate two chromosomes to produce the next generation.
Multimed Tools Appl (2014) 72:30853103 3095
e. The mutation step is to prevent premature convergence to local optima by randomly
sampling the number of genes. Alter one or more genes with the mutation probability.
3.5 The objective function of proposed scheme
The selection of the tuning parameters consists of population size, crossover rate and
mutation rate. The optimum values for these parameters are 30, 0.8 and 0.01 respectively.
Meanwhile, generation size is set from 1 to 50 as an initial input and each number has
different evaluation function values.
4 Experiment results
This section provides the experiment setup, experiment results of the proposed scheme, and
the comparison with other techniques.
4.1 Experiment setup
Figures 3a, b, c, and d illustrate the host image of the e-government document image of size
2561,024, the host image of Lena, the host image of the education document image, the
watermark of the logo image of Indonesian Institute of Sciences of size 256256, used in the
experiment of this research, respectively.
To measure the quality of the watermarked image, the criteria of peak signal to noise ratio
(PSNR) is used. The minimum square error (MSE) is used to measure the objective function
values as well.
PSNR 10log
10
255
2
_
MSE
_ _
18
MSE
1
MN

x1
M

y1
N
I x; y I
0
x; y
2
19
3096 Multimed Tools Appl (2014) 72:30853103
For a double (M) x double (N) blocks monochrome images, I (x, y) and I(x, y), on the
other hand, represent the gray-values for the host image and watermarked image. To verify
the existence of the watermark, a correlation coefficient between the host image and the
extracted watermark can be calculated as follows:
V; W

J1
k
V
j
Vmean; k
_
: W
j
Wmean; k
_

j1
k

V
j
Vmean; k
_
: W
j
Wmean; k
_ _
_ _ _
2
_ 20
Fig. 3 a the host image of the e-government document image, b the host image of Lena, c the host image of
the education document image, d the watermark image
Multimed Tools Appl (2014) 72:30853103 3097
where V refers to the original watermark, W is the extracted watermark, Vmean is the mean
of original watermark, Wmean is the mean of extracted watermark and k is (M, N).
4.2 Experiment results of the proposed scheme
In this experiment, the robustness of the proposed scheme in resistant to several types of attacks
such as Gaussian noise, Cropping, JPEG compression, Median filtering, Salt and pepper noise
and Average filtering is investigated. Under those various attacks, the parameter value indicat-
ing the robustness, namely the great robust NC (Normalized Correlation Coefficient) value of
the extracted watermark fromeach area with different attacks and scales (e.g. in Gaussian noise
attack, the used scale are 1, 1.5, 2, 2.5, 3, 4.5, and 5) is tabulated in Table 1.
4.3 Comparison with other techniques
Here, the proposed scheme is compared with some previous schemes (e.g. the schemes
schematically being reported by Himansu et al. [13], Rosiyadi et al. [28], Xue and Teng [35],
Carballeira et al. [7], and Naiss et al. [26]). In such comparison, several types of attacks are
applied on the proposed scheme and other previous schemes with the host image of the e-
government document image of size 2561,024 and the watermark of the logo image of
Table 1 Experiment results under six types of attacks
a
EGDI is E-Government Document Image
b
EdDI is Education Document Image
3098 Multimed Tools Appl (2014) 72:30853103
Indonesian Institute of Sciences of size 256256. Figure 4 compares the performance of the
proposed scheme with those reported by [7, 13, 26, 28, 35] in which the types of attacks applied
are JPEG compression 50, Scaling 1, Gaussian noise 1, Rotation 2, and Median filtering (22).
Based on Fig. 4, the proposed scheme has the highest value of NC for the types of attacks,
Gaussian Noise 1 and Scaling 1. This indicates that the proposed scheme is more robust
towards both types of attacks and the performance of the proposed scheme is relatively better
compared to the previous researches [7, 13, 26, 28, 35].
Regarding the processing time of the proposed scheme, we repeat for 150 times using the
host image of the e-government document image of size 2561,024 and the watermark of
the logo image of Indonesian Institute of Sciences of size 256256 to calculate the
processing time of the proposed scheme and the previous scheme reported by Rosiyadi et
al. [28]. The average processing time over 150 experiments is then counted. The average
processing times of the proposed scheme and the previous scheme reported by Rosiyadi et
al. [28] are 66.5045 s and 65.7910 s, respectively.
It is found that the processing time of the proposed scheme is slightly higher about
713.5ms than the previous scheme reported by Rosiyadi et al. [28]. This result is due to the
proposed scheme including more processes in which the luminance masking process has
been used to improve noise sensitivity, as shown in Table 2.
Fig. 4 The comparison results of NC value between the proposed scheme and other schemes
Table 2 the comparison of the watermarking scheme process between the proposed scheme and the previous
scheme reported by Rosiyadi et al. [28]
Proposed scheme Rosiyadi et al. [28]
Hybrid DCT-SVD Hybrid DCT-SVD
The optimization Process by Genetic Algorithm The optimization Process by Genetic Algorithm
The process of luminance masking model in the
human visual system (HVS).

Multimed Tools Appl (2014) 72:30853103 3099

5 Conclusion
This study presents an adaptive watermarking scheme for e-government document images. In this
scheme, the mask of the host image with luminance masking based on features of the image in
HVS and transforming is calculated for all frequencies on each area. This is followed
by modifying the singular values of DCT-transformed host image with singular values
of mask coefficient of host image and the control parameter used to avoid the false
positive problem of DCT-transformed watermark image using GA. It is found that the
proposed scheme is more robust and sensitive against several types of attacks;
particularly, Gaussian Noise 1 and Scaling 1. Further research could be done in the
adaptive watermarking scheme based on texture and edge masking.
Acknowledgments We would like to thank the Department of Computer Science and Information Engineering,
National Taiwan University of Science &Technology for giving the permission to commence this thesis in the first
instance, to do the necessary research work and to use departmental data.
References
1. Amiri MD, Danyali H, Zahir-Azami B (2010) An adaptive robust and multiresolution image
watermarking for progressive wavelet image coding, 6th International Conference on Wireless Commu-
nications Networking and Mobile Computing (WiCOM). pp 1
2. Aslantas V(2007) An SVDbased digital image watermarking using genetic algorithm. Int J Electron Commun: 14
3. Bao P, Ma X (2005) Image adaptive watermarking using wavelet domain singular value decomposition.
IEEE Trans Circuits Syst Video Technol 15(1):96102
4. Barni M, Bartolini F, Piva A (2001) Improved wavelet-based watermarking through pixel-wise masking.
IEEE Trans Image Process 10(5):783791
5. Belkacem S, Dibi Z, Bouridane A (2007) A masking model of HVS for image watermaking in the DCT
domain. IEEE Int Conf Electron, Circ Syst: 330
6. Benhocine A, Laouamer L, Nana L, Pascu AC (2013) New images watermarking scheme based on
singular value decomposition. J Inf Hiding Multimedia Signal Process 4(1):918
7. Carballeira P (2012) Analysis of pixel-mapping rounding on geometric distortion as a prediction for view
synthesis distortion, Proceeding of 3DTV-Conference: the true vision - capture, transmission and display
of 3D video (3DTV-CON) 2012, pp 14
8. Coello Coello CA, Toscano Pulido G (2001) A micro-genetic algorithm for multi objective optimization.
Proc Genet Evolut Comput Conf
9. Cui L, Li W (2011) Adaptive multiwavelet-based watermarking through JPW masking. IEEE Trans Image
Process 20(4):10471060
10. Genetic Algorithms http://www.cs.ucc.ie/~dgb/courses/tai/notes/handout12.pdf
11. Hamed M, Hossein Khosravi R, Mohammad R, Akbarzadeh T(2010) Anewadjustable blind watermarking based
on GAand SVD, Proceeding of International Conference on Innovations in Information Technology (IIT), pp 610
12. He X, Chang S (2011) An adaptive image watermarking algorithm based on neural network. Int Conf
Intell Comput Technol Autom 2:408
13. Himanshu H, Rawat S, Raman B, Bhatnagar G (2011) DCT and SVD based new watermarking scheme,
3rd International Conference on Emerging Trends in Engineering and Technology (ICETET), pp 146151
14. Hoi-Yu Tong H (1997) A perceptually adaptive JPEG coder, master thesis, University of Toronto
15. Huang HC, Chen YH (2009) Genetic fingerprinting for copyright protection of multicast media. Soft
Comput 13(4):383391
16. Huang HC, Chu CM, Pan JS (2009) The optimized copyright protection system with genetic
watermarking. Soft Comput 13(4):333343
17. Huang C-H, Wu J-L (2000) Awatermark optimization technique based on genetic algorithms. Proc SPIE 3971
18. Kang X, Huang J, Zeng W (2008) Improving robustness of quantization-based image watermarking via
adaptive receiver. IEEE Trans Multimedia 10(6):953959
19. Kumar Sharma D, Kumar Pathak V, Sahu GP (2007) Digital watermarking for secure e-government
framework Computer Society India, pp 182191
3100 Multimed Tools Appl (2014) 72:30853103
20. Kuo S-s, Johnston JD (2002) Spatial noise shaping based on human visual sensitivity and its application
to image coding. IEEE Trans Image Process 11(5):509517
21. Lai CC, Huang HC, Tsai CC (2009) A digital watermarking scheme based on singular value decomposition
and micro-genetic algorithm. Int J Innov Comput Inf Control 5(7):18671873
22. Li Q, Yuan C, Zhong Y-Z (2007) Adaptive DWT-SVD domain image watermarking using human visual
model. Int Conf Adv Commun Technol 3:1947
23. Lim SB (2010) Human visual system, http://www.siongboon.com/projects/2010-08-22/electronicballast/
HumanVisualSystem.pdf
24. Ma X, Shen X (2008) A novel blind grayscale watermark algorithm based on SVD, Proceeding of
International Conference on Audio, Language and Image Processing (ICALIP), shanghai, pp 10631068
25. Mairgiotis AK, Galatsanos NP (2008) New additive watermark detectors based on a hierarchical spatially
adaptive image model. IEEE Trans Inf For Sec 3(1):2937
26. Naiss I (2013) Computable bounds for rate distortion with feed forward for stationary and ergodic
sources. IEEE Trans Inf Theory 59(2):760781
27. Pappas TN, Safranek RJ, Chen J (2005) Perceptual criteria for image quality evaluation. In: Bovik AC (ed)
Hand book of image and video processing, 2nd edn. Academic Press, San Diego, United States of America
28. Rosiyadi D, Horng S-J, Fan P, Wang X, Khan MK, Pan Y (2012) An efficient copyright protection
scheme for e-government document images. IEEE Multimedia 19(3):6273
29. Si H, Li C-T Maintaining information security in e-government through steganology, Department of
Computer Science, University of Warwick, Coventry CV4 7AL, UK
30. Sing K (2011) edge detector, http://yumnamkirani.hpage.us/edge_detection_74376639.html
31. Verma A, Tapaswi S (2009) A novel reversible visible watermarking technique for images using Noise
Sensitive Region Based Watermark Embedding (NSRBWE) approach, IEEE EUROCON 09, pp 1374
32. Voloshynovskiy S, Pereira S, Pun T, Eggers JJ, Su JK (2001) Attacks on digital watermarks: classification,
estimation based attacks, and benchmarks. IEEE Commun Mag 39(8):118126
33. Wang J (2008) Design of e-government security system based on information security model. Inf Manag,
Innov Manag Ind Eng 2: 359362, ISBN: 978-0-7695-3435-0, 1921
34. White GB, Jones S (2007) Minitrack E-government information security, system sciences, 2007. HICSS
2007. 40th Annual Hawaii International Conference on, E-ISBN: 0-7695-2755-8, pp 9797
35. Xue S, Teng X (2007) Adaptive blind watermarking based on improved noise sensitivity. Fourth Int Conf
Fuzzy Syst Knowl Discov 4:731734
36. Yan Y, Wei C, Li S (2009) Block-based adaptive image watermarking scheme using just noticeable
difference. IEEE Int Work Imaging Syst Tech: 377380
37. Zhang X, Han G, Zou K, Li W, Li B (2007) An effective mechanism based on watermark for e-government
information security. Int Conf Converg Inf Technol- IEEE:580585
38. Zhao F, Liu G, Ren F (2009) Adaptive blind watermarking for JPEG2000 compression domain, 2nd
International Congress on Image and Signal Processing, pp 1
39. Zhu S-m, Liu J-m (2008) Adaptive watermarking scheme in hybrid DWT-DCT transform based on human
visual system. Int Symp Knowl Acquis Model: 668
Shi-Jinn Horng received his PhD. degree in computer science from the National Tsing Hua University in 1989.
Currently, he is a professor in the Department of Computer Science and Information Engineering, National Taiwan
University of Science and Technology. His research interests include multimedia, biometrics, and parallel algorithms.
Multimed Tools Appl (2014) 72:30853103 3101
Didi Rosiyadi received his Master degree in the Computer Science, Gadjah Mada University, Yogyakarta,
Indonesia in 2003. Currently, he is a PhD Student in Department Computer Science and Information
Engineering, National Taiwan University of Science and Technology, Taiwan. His research interests include
watermarking, and multimedia.
Pingzhi Fan (FIEE, SMIEEE) received his PhD degree in electronic engineering from Hull University, UK.
He is currently a professor and a vice president of Southwest Jiaotong University, Chengdu, China. He is the
inventor of 20 patents, and the author of over 300 research papers and 8 books. His research interests include
CDMA theory and technology, information theory & coding, information security.
3102 Multimed Tools Appl (2014) 72:30853103
Xian Wang received his Ph.D. degrees in Communication and Information Systems from Southwest Jiaotong
University, Chengdu, China in 2008 and was an associate professor at the same university. He is also a
postdoctoral researcher with the Department of Computer Science and Information Engineering, National
Taiwan University of Science and Technology, Taiwan. His research interests include mobility management
and performance modeling for personal communications service network.
Dr. Muhammad Khurram Khan (M07) is currently working as associate professor and R&D Manager at
the Center of Excellence in Information Assurance (CoEIA), King Saud University, Saudi Arabia.
He is also an honorary Professor at IIIRC, Shenzhen Graduate School, China. He has published more than
100 papers in journals and conferences. His areas of interest are biometrics, and multimedia security.
Multimed Tools Appl (2014) 72:30853103 3103