=Paper=
{{Paper
|id=Vol-1623/paperls1
|storemode=property
|title=Robust Image Watermarking Technique Based on Genetic Algorithm Optimization and Even Odd Modulation
|pdfUrl=https://ceur-ws.org/Vol-1623/paperls1.pdf
|volume=Vol-1623
|authors=Alexander Bahrushin, Galina Bahrushina, Ruslan Bazhenov, Kiseon Kim, Rudolf Tsoy
|dblpUrl=https://dblp.org/rec/conf/door/BahrushinBBKT16
}}
==Robust Image Watermarking Technique Based on Genetic Algorithm Optimization and Even Odd Modulation==
https://ceur-ws.org/Vol-1623/paperls1.pdf
Robust Image Watermarking Technique Based
on Genetic Algorithm Optimization
and Even Odd Modulation
Alexander Bahrushin1, Galina Bahrushina2, Ruslan Bazhenov1,
Kiseon Kim3, Rudolf Tsoy1
1
Amur State University named for Sholom Aleichem, Birobidzhan, Russia
stripylife@yahoo.com, r-i-bazhenov@yandex.ru,
rudolft55@mail.ru
2
Pacific State University, Khabarovsk, Russia
gal_bah@mail.ru
3
Gwangju Institute of Science and Technology, South Korea
kskim@gist.ac.kr
Abstract. Robustness is one of the important issues in watermarking. In
particular, robustness against JPEG compression is especially actual.
Nowadays there are many different applications requiring image
compression, such as Internet, satellite imaging, remote sensing,
multimedia, preservation of art work and so on. In this paper, an effective
watermarking technique based on the genetic algorithm optimization and
even-odd modulation in frequency domain is presented. The genetic
algorithm finds the optimum watermark strength and location for its
embedding in a spectrum of the cover image. Experimental results have
demonstrated high values of robustness and fidelity of the proposed scheme.
Keywords: image watermarking, genetic algorithm optimization, even-odd
modulation, robustness, fidelity, frequency domain, JPEG compression,
discrete cosine transform
1 Introduction
Digital watermarking refers to the process of embedding an authentication
message called watermark into some content, for instance, the digital image, which
uniquely identifies its ownership. Regardless of the particular applications any
watermarking scheme must meet two main requirements. The visual quality of the
watermarked image should be high and the watermark should withstand against any
image processing in one way or another. Such processing may include lossy
compression, filtering, added noise and many others. In watermarking terminology,
any processing that may prevent detection of the watermark is called an attack. In
general case the image is sent on the Internet as the compressed image to optimally
use the bandwidth of the network. In such applications the image compression is
essential to save storage space and transmission time.
Copyright © by the paper's authors. Copying permitted for private and academic purposes.
In: A. Kononov et al. (eds.): DOOR 2016, Vladivostok, Russia, published at http://ceur-ws.org
416 A. Bahrushin, G. Bahrushina, R. Bazhenov, K. Kim and R. Tsoy
The embedded watermark should possess two key qualities, robustness and
fidelity. Robustness implies the strength of the watermark to sustain any attacks.
Fidelity accounts for the quality of the watermarked image.
At present there are two popular techniques in watermarking: when an image is
processed in spatial domain and in transform domain.
Spatial domain based watermarking focuses on modifying the pixels of images.
It directly loads the values of a watermark into the image pixels. However spatial
domain techniques have weak robustness against common image processing and
attacks which may easily destroy the watermark [3].
Transform domain watermarking techniques are suitable first of all in such
applications where robustness is of prime concern. The idea of transformed
watermarking is reduced to embedding a watermark by altering the coefficients in
frequency domain obtained with a help of discrete Fourier transform (DFT) [4],
discrete cosine transform (DCT) [5] and discrete wavelet transform (DWT) [1,2].
Since the human eyes are more sensitive to low frequency distortions, the
watermark should be embedded into the high frequency coefficients to attain better
perceptional invisibility. However, the watermark hidden in the high frequency
domain might be discarded after lossy compression and other attacks. So to make
effective withstand against such attacks it is preferable to embed the watermark into
the lower frequency range of the image.
To achieve its high robustness against different attacks and in the same time to
provide the fidelity the even-odd modulation (EOM) technique is developed. It has
been shown in this work that the EOM technique provides very high fidelity of the
image even in a case when the watermark is embedded into the lower frequency range
of the image. So in order to increase robustness of the scheme against such dangerous
attacks as JPEG compression it is decided to change the values of the lower frequency
coefficients while embedding the watermark to the image.
Another way to improve the performance of watermarking schemes is to make
use of artificial intelligent techniques. The watermarking can be viewed as an
optimization problem. Therefore, it can be solved by Genetic Algorithm (GA), which
allows improving the fidelity of the watermarked image while keeping the robustness
of the scheme against image manipulations. There are a lot of publications devoted to
GA. For example, Huang and Wu [6] proposed a watermarking scheme based on the
DCT and GA. They embed the watermark into the image by selectively modifying the
middle-frequency coefficients of the image. The GA is applied to search for the
locations to embed the watermark in the DCT coefficient blocks such that the quality
of the watermarked image is optimized.
In papers [8-12] the surveys of digital watermarking with GA are presented.
In this paper, a new blind scheme of digital image watermarking is proposed. To
achieve high robustness of it against different attacks and in the same time to provide
high fidelity the even-odd modulation technique is developed, which embeds the
watermark in low frequency DCT coefficients by changing their parity. To find the
best position for the watermark embedding the GA is used which allows to improve
the robustness and fidelity of the scheme. Comparison has been made between the
proposed scheme and others presented in the literature to examine the image fidelity
and to evaluate robustness.
Robust Image Watermarking Technique Based on Genetic Algorithm 417
2 Watermark embedding procedure
The proposed scheme uses the DCT, which creates a frequency spectrum of the
entire original gray-level image I of size M N . This spectrum is divided into three
main regions namely low frequencies sub-band (LF), middle frequencies sub-band
(MF) and high frequencies sub-band (HF). The watermark W of size G H is
embedded within the LF sub-band of the area by modifying the coefficients of the
LF sub-band as shown in Fig.1.
Assume that the coefficients a(m, n) of the obtained spectrum are represented
by matrix A of size M N , where m 1, M , n 1, N . Then in general case any
coefficient may be written as an integer decimal number:
a(m, n) a(m, n) P 110 P 1 a(m, n) P 2 10 P 2 ... a(m, n)k 10 ... a(m, n)110 a(m, n)010 ,
k 1 0
where P is the total quantity of digits,
k is the number of digit in the coefficient a(m, n) , counting from right to left,
0 k P .
Let the digital watermark W be a binary matrix of size G H formed by a
pseudo-random number generator (PRNG).
Assume that the embedding of the watermark W within the region is carried
out in such a way that if w( g , h) 1 then it must comply with even coefficient
a(m, n) and, respectively, if w( g , h) 0 then it must comply with odd
coefficient a(m, n) . Let also admit that the coordinates m and n of upper left angle
of the region are given as it is shown in Fig. 1.
n
m
G LF MF HF
H
M
N
Fig. 1. Matrix A of size M N , (LF - low frequencies, MF - middle
requencies, HF - high frequencies)
418 A. Bahrushin, G. Bahrushina, R. Bazhenov, K. Kim and R. Tsoy
Then the embedding process may be described as follows:
if w( g , h) 1 and a(m 1 g , n 1 h) is odd then a(m 1 g , n 1 h)
turns into even number,
if w( g , h) 1 and a(m 1 g , n 1 h) is even then a(m 1 g , n 1 h)
remains unchanged,
if w( g , h) 0 and a(m 1 g , n 1 h) is odd then a(m 1 g , n 1 h)
remains unchanged,
if w( g , h) 0 and a(m 1 g , n 1 h) is even then a(m 1 g , n 1 h)
turns into odd number,
where g 1, G and h 1, H .
However the smallest noise can change the parity of the coefficient. Really, it is
just enough to add “1” to the coefficient to change its parity. In this case the
robustness of the watermarking scheme when it exposed to a smallest noise will be
extremely low. To increase the robustness of the scheme it is proposed to change the
parity of the more significant digits of the coefficients. For example, one watermark
bit may be embedded into the second digit a(m, n)1 of the coefficient. Then the noise
with the same value will not change the parity of the digit a(m, n)1 . So it is easy to
make a conclusion that the more significant digit of the coefficient is used to embed a
bit of the watermark, the more difficult to change its parity and, consequently, the
scheme will be more robustness to noise. However it should be noted that there may
be a situation where the addition to the coefficient a(m, n) of any number might
change the parity of more significant digits. Let, for example, the coefficient
a(m, n) has a value 3999 (that is P 4 ) and the watermark bit has been embedded into
the digit a(m, n) 3 . Then a noise with any level will change the parity of this digit.
To avoid such situations it is proposed to modify the coefficient a(m, n) in the
following way. Suppose that the digit a(m, n) 3 was chosen for watermark bit
embedding. To “protect” it against influence of noise it is proposed to assign to the
previous digit a(m, n) 2 a value that is greater than zero but less than nine, for
example, the value of four. After such modification the number 3999 will be turned to
number 3499. Then if a(m, n) 3499 the digit a(m, n) 3 will not change its parity
when the noise levels are in the range from -499 to 500. After the watermark has been
embedded the inverse DCT is made using the modified matrix  and as a result the
watermarked image Iˆ is obtained.
Note that the choice of the concrete digit for embedding of watermark’s bit
plays the same role as the choice of an amplifier coefficient value in watermarking
[3,16]:
Sˆ S W ,
where S is a matrix representing a spectrum of the original image;
Ŝ is a matrix representing a spectrum of the watermarked image;
is an amplifier coefficient.
Robust Image Watermarking Technique Based on Genetic Algorithm 419
3 Watermark extraction procedure
The aim of this procedure is to form a matrix Ŵ , which in ideal case should be
identical to the matrix W . In connection with that the direct DCT is first performed
over the image Iˆ . Admit that the watermark bits have been embedded to the k -th
digit of the coefficients within the region . Then by known coordinates m and n
the upper left angle of the region is found and within that region the corresponding
ˆ ( g , h) of the matrix Ŵ :
digits of the coefficients are analyzed to form elements w
ˆ ( g , h) 1 , otherwise wˆ ( g , h) 0 ,
if a(m 1 g , n 1 h) is even then w
where g 1, G and h 1, H .
So as a result the matrix Ŵ will be formed, which represents the extracted
watermark.
4 Genetic algorithm
Any optimization problem is modeled in GA by defining the chromosomal
representation, fitness function (FF) and the GA operators [7-11]. It starts with some
randomly selected population made of individuals, each corresponding to a solution
of the problem. An individual in the population is called chromosome. The FF is used
to evaluate the quality of each solution so that chromosomes with high quality will
survive and form a population for the next generation. The GA recombines a new
generation to find the best optimal solution. The three GA operators, crossover,
selection, and mutation, are applied to the chromosomes repeatedly to determine the
In the watermarking the amplification coefficient is used in a way to provide
balance between fidelity and robustness of the watermarked scheme. Since these
requirements are conflicting, then the value of should be an optimization
parameter. The value of determines the strength of the watermark and can be
varied within some range. In the proposed scheme the strength of the watermark is
determined by choosing the concrete digits of the frequency coefficients.
The FF is calculated as a combination of the Peak Signal Noise Ratio (PSNR)
and the Normalized Coefficient (NC):
FF PSNR 100 NC .
The PSNR evaluates the fidelity of the watermarked image (quality index) and is
computed as
B2
PSNR 10 log 10 ,
1 M 1 N 1
[s1 (l, k ) s 2 (l , k )]2
M N l 0 k 0
where B is a maximum pixel brightness, for example, 255;
s1 (l , k ) and s 2 (l , k ) are pixels brightness of the images I и Iˆ , respectively.
420 A. Bahrushin, G. Bahrushina, R. Bazhenov, K. Kim and R. Tsoy
The NC evaluates the robustness of the watermarking scheme and is computed as
G H
w( g , h) wˆ ( g , h)
g 1 h 1
NC ,
G H G H
w ( g , h) wˆ ( g , h)
g 1 h 1
2
g 1 h 1
2
where w( g , h) and w ˆ ( g , h) represent the original W and the extracted Ŵ
watermarks, respectively.
The NC determines the similarity between the original watermark W and the
extracted watermark Ŵ from the attacked watermarked image (robustness). The FF
is increased proportionately with the increasing of PSNR and NC. The NC has been
multiplied by factor 100 since its normal values are in the range 0 - 1, while the
values of PSNR may lie in a wide range (mainly from 30 to 100). The FF is computed
for all chromosomes in the population and the best chromosomes with the
corresponding fitness value are saved. Genetic operators like mutation and crossover
are performed on the selected parents to make new offspring which are included in
the population to form the next generation.
The chromosomes with low fitness values are discarded through this process.
The discarded chromosomes will be replaced by new offspring after executing the
crossover and mutation genetic operators.
Crossover is responsible for producing better offspring by inheriting high-
quality genes from their parents. In this work a crossover rate of 0.7 is used.
Mutation refers to the random alteration of the value in some positions of some
chromosomes. The mutation is usually selected with a probability between 0 and 1,
so that only a small portion of the genes in the chromosomes will be selected to
be muted. In this study the mutation rate equal to 0.05 is selected.
The whole process of robustness and fidelity optimization using GA can be
described as follows.
1. A cover image and a watermark image are chosen.
2. The cover image is transformed by DCT into frequency domain and the area
(see Fig. 1) of the obtained spectrum is divided into eight parts every of
which has size 8×16.
3. The watermark is also divided into eight parts every of which has size 8×16 (see
Fig. 3).
4. Coming from a given value of PSNR the digit of low-frequency coefficients is
set in which the bits of watermark must be embedded.
5. The first generation of GA chromosomes based on the watermark parameters is
created. A different watermarked image is generated for each chromosome.
6. The fidelity of each watermarked image is evaluated by calculating the
corresponding value of PSNR.
7. A chosen attack on the watermarked image is applied.
8. The watermark from the attacked image is extracted.
9. The robustness is evaluated by calculating NC between the original and
extracted watermarks .
Robust Image Watermarking Technique Based on Genetic Algorithm 421
10. The FF for the obtained PSNR and NC values is computed.
11. The chromosomes with the best FF values are selected.
12. The new population is generated by performing the crossover and mutation
functions on the selected chromosomes.
13. The steps from 5 to13are repeated until a predefined iteration number has been
reached.
The entire process is repeated for several generations until the best solutions are
obtained as showed in Fig. 2.
Steps from 1 to 4
Create initial
population
Parent selection
Crossover
Mutation
No Stop
criteria
Yes
Stop
Fig. 2: The chart of a typical GA
5 Simulation and results
The proposed scheme is implemented in MATLAB software. The results
obtained during different experiments are compared with the results presented in
[1,2,5,14-18]. Therefore for correct comparison of results those 8-bit gray-level test
images of size 512×512 and the watermarks that have been used in these works were
chosen.
422 A. Bahrushin, G. Bahrushina, R. Bazhenov, K. Kim and R. Tsoy
The binary watermarks of dimensions 16 32 and 32 32 pixels are formed by
PRNG. To evaluate the quality of the watermarked image obtained by the proposed
scheme the PSNR is used.
The strength of the watermark is varied by embedding its bits into different
digits of frequency coefficients. Before embedding the watermark is divided into
eight parts as shown in Fig. 3.
w1 w2 w3 w4
w5 w6 w7 w8
Fig. 3. Watermark divided into eight parts which must be embedded into the area
The embedding positions of the different parts of a watermark within the low
frequency spectrum of the cover image are determined by chromosome. Fig. 4 shows
the sample of chromosome.
010 001 110 111 011 101 000 100
w1 w2 w3 w4 w5 w6 w7 w8
Fig. 4. A 24-bit chromosome
The performance of the extraction scheme was evaluated by measuring fidelity
and robustness. The NC was used to measure the image quality of the watermark after
extraction.
The Bit Error Rate (BER) is also employed to measure the image quality of the
watermark after extraction. The BER has been calculated as the number of incorrectly
decoded bits divided by the total number of embedded bits in the cover image:
G H
1
BER V ( g , h) ,
G H g 1 h 1
1, if w( g , h) wˆ ( g , h)
where V ( g , h) .
0, if w( g , h) wˆ ( g , h)
In accordance with the proposed scheme the watermark bit should be embedded
to the coefficient digit in order to change its parity. Naturally, more significant digits
Robust Image Watermarking Technique Based on Genetic Algorithm 423
of the number representing a coefficient are less susceptible to noise. But at the same
time, the more significant digit is selected for embedding a watermark bit, the higher
distortion of a number representing the coefficient. In order to determine these
relationships three versions of the proposed scheme were tested in the following way.
First of all note that for convenience, it is desirable to introduce the values of
coefficients in integer form while after the DCT the coefficients are represented in a
real form as shown in Table 1. Therefore these real numbers are converted into
integers by multiplying them by 10000 (see Table 2).
During the experimental research the watermark bits have been embedded into
different digits of frequency coefficients, namely into the digit when k 7 (version
~aaaaaa ) and into the
~aaaaaaa ), into the digit when k 6 (version B, aa
A, a
digit when k 3 (version C, aaaaa~aaa ).
The robustness of the proposed scheme under JPEG compression was compared
with other schemes. Four images were selected in the experiment. The quality factor
(or, for short, QF) was used to determine the degree of loss in the compression
process. In general, JPEG recommends QF of 75–95 for visually indistinguishable
distortions, and QF of 50–75 for merely acceptable quality.
Table 1. Coefficients obtained after the DCT
-390.3282 -47.4748 584.4324 -135.2312
- 334.2655 844.9531 535.6707 -1044.2649
1164.7977 -348.4653 39.8763 -84.5748
- 1445.1743 1310.2793 -848.0618 -334.0621
Table 2. Coefficients in integer form
-3903282 -474748 5844324 -1352312
-13342655 8449531 5356707 - 10442649
11647977 -3484653 398763 -845748
- 14451743 13102793 -8480618 -3340621
The experimental results obtained with the proposed scheme and with other
schemes collected from the original papers [1,2,5,14-16] are shown in Table3. The
424 A. Bahrushin, G. Bahrushina, R. Bazhenov, K. Kim and R. Tsoy
values of NC and PSNR presented here are obtained using standard gray-level image
Lena comprising 512×512 pixels, in conjunction with 16×32 watermark.
Table 3. Dependence of NC vs. factor QF
Proposed scheme
QF [1] [2] [14] [15] [16]
A B C
NC
1.5 - - - - - 0.59 0.59 0.69
3 - - - - - 0.67 0.61 0.61
5 - - - - - 0.85 0.62 0.61
10 - 0.42 0.35 0.39 0.61 0.98 0.65 0.62
15 - 0.56 0.52 0.72 0.84 0.99 0.76 0.62
20 - 0.70 0.69 0.91 0.96 0.99 0.97 0.63
25 - 0.80 0.81 0.96 0.99 1 0.99 0.64
30 0.24 0.86 0.85 0.98 0.99 1 0.99 0.64
40 0.37 0.94 0.92 0.99 1 1 1 0.66
50 0.49 0.97 0.95 1 1 1 1 0.69
70 0.75 1 0.96 1 1 1 1 0.99
90 1 1 0.99 1 1 1 1 1
Lossy compression is so common procedure in signal processing that watermark
robustness to this attack is highly desirable. From Table 3 and Table 4 it is seen that
the proposed scheme has the best robustness while keeping a high quality. All 512
bits have been accurately extracted when QF is up to 20. Even when QF decreases to
1.5, nearly 60% watermark bits can be correctly extracted, which demonstrates a
significant advantage over other schemes.
Table 3 also shows that the proposed scheme (version A) guarantees the
watermark detection with NC 1 for all values of JPEG attacks ranging from 20%.
Table 4 presents the average values of PSNR obtained with the proposed
scheme after processing 4 images of size 512×512 with watermark of size 16×32 and
results in [2,5,15,16]. It is seen that the obtained values of PSNR are in between 58.7
dB and 143.1. So the results obtained with the proposed scheme demonstrate the
highest quality of the images after the watermark has been embedded.
Another comparison was carried out applying the gray-level Lena image of size
512 512 with the watermark of 32 32 . The JPEG compression was implemented to
the watermarked image and the experimental results are presented in Fig. 5, which
reflects the dependence of PSNR vs. QF when using different versions of the
proposed scheme and the scheme described in [5].
Robust Image Watermarking Technique Based on Genetic Algorithm 425
Table 4. The PSNR after watermark embedding
Papers [2] [15] [5] [16] Proposed scheme
A B C
PSNR (dB) 42.2 41.5 43.3 42.2 58.7 99.5 143.1
130
120
110
100
90
PSNR
80
C
70
60
A
50
40
[5]
30
20 40 60 80 100
QF
Fig. 5. The plots of some simulation results reflecting dependence of PSNR vs. QF
for different schemes.
The values of PSNR received for four different images of size 512×512 when
the watermark of size 32×32 was embedded are shown in Table 5. It is easy to see
that the obtained results are much better than in [5,17].
Table 5. The values of PSNR for four different images
Quality Tested Proposed scheme
index images [17] [5]
A B C
Man 30.85 43.32 56.7 89.9 121.9
Couple 33.62 44.42 56.7 88.9 120.5
PSNR Plane 33.69 44.42 57.8 87.6 118.7
Lake 31.57 42.48 56.8 89.1 122.2
426 A. Bahrushin, G. Bahrushina, R. Bazhenov, K. Kim and R. Tsoy
The results of the watermark robustness against various attacks are presented in
Table 6. It is seen that the values of BER and NC obtained by schemes [5,18] are
better than in a case of the developed scheme. Still this scheme resists these four
kinds of attacks in a certain scope.
Table 6. Different attacks
Type of attack
Cropping Gaussian Median Low pass
Papers 1/8 ½ noise 0.005 5×5 filter
3×3
[18] 0.05 0.28 0.23 0.27 0.09
[5] 0.04 0.24 0.06 0.17 0
BER Proposed 0.28 0.33 0.26 0.23 0.25
scheme (A)
[18] 0.954 0.811 0.855 0.821 0.938
[5] 0.967 0.864 0.670 0.900 1
NC Proposed 0.670 0.576 0.721 0.839 0.772
scheme (A)
Conclusion
A blind watermarking scheme for gray-level still images using 2D DCT has
been proposed. The scheme embeds the watermark in low frequency DCT
coefficients by changing their parity. It was found that the use of the EOM technique
and GA improved the robustness against JPEG compression and some other attacks
such as additive noise, cropping and filtering. Comparison has been made between the
proposed scheme and others presented in the literature to examine the image fidelity
and evaluate robustness. The results show that the developed scheme can reach higher
values of PSNR and NC, respectively.
Further researches can be concentrated on the development of the proposed
scheme by using the characteristics of the human visual system. It would be also
interesting to make the evaluation of GA convergence depending on a certain class of
images.
References
1. Wang, S.H., Lin, Y.P.: Wavelet tree quantization for copyright protection
watermarking. IEEE Transactions on Image Processing, vol. 13, no. 2, pp. 154–
165 (2004)
2. Lin, W. H., Horng, S.J., Kao, T.W., Fan, P., Lee, C.L., Pan, Y.: An efficient
watermarking method based on significant difference of wavelet coefficient
quantization. IEEE Trans. Multimedia, vol. 10, no. 5, pp. 746–757 (2008)
3. Hussein Jamal A.: Spatial Domain Watermarking Scheme for Colored Images
Based on Log-average Luminance. Journal of computing, vol. 2, issue 1, pp.100
- 103 (2010)
Robust Image Watermarking Technique Based on Genetic Algorithm 427
4. Shahin Shaikh, Manjusha Deshmukh.: Digital Image Watermarking in DCT
Domain. International Journal of Emerging Technology and Advanced
Engineering. Certified Journal, vol. 3, issue 4, pp. 289-293 (2013)
5. Ma Qiomel, Wu Lijun, Du Jianhong, Chen Gouxi, Yang Qiuxiang.: An adaptive
all-odd transformation watermark scheme” TELKONNIKA Indonesian Journal
of Electrical Engineering, vol. 12, no. 5, pp. 4107 – 4114 (2014)
6. Ali Al-Haj.: Combined DWT-DCT Digital watermarking, Journal of Computer
Science, vol. 3, issue 9, pp. 740-746 (2007)
7. Huang, C. H., Wu, J. L.: A Watermark Optimization Technique Based on
Genetic Algorithms. Proc. SPIE Int. Conf. Visual Communications and Image
Processing, San Jose, CA, USA., vol. 3971, pp. 516-523 ( 2000)
8. Sachin Goyal, Roopam Gupta, Ashish Bansal, Survey of Digital Watermarking
with Genetic Algorithms, CSI COMMUNICATIONS, pp. 512-522 (2010)
9. Vagesh Porwal, Siddharth Gupta.: A survey on image watermarking in
Frequency Domain using Genetic Algorithm. International Journal of Advanced
Research in Computer Engineering & Technology, vol. 4, issue 4, pp. 1135 -1145
(2015)
10. Sowmi, T., Kumar, E.S.: A Survey on Fisher Criterion based Genetic Algorithm
for Feature Selection Method. South Asian Journal of Engineering and
Technology, vol.2, no.12, pp. 1–8 (2016)
11. Swapnil Bhosale, Ashish Nikalje, Swapnil Borude, Himalaya Wadhwani, Latika
Desai.: Survey on Data Security. International Journal of Innovative Research in
Computer and Communication Engineering. Certified Organization, vol. 4, issue
2, pp. 1174-1177 (2016)
12. Udaya Rani, Shreedevi B. Choudri, Vinayak Murthy.: Survey on Robust and
Reversible Watermarking for Relational Data. International Journal of Computer
Applications, vol. 139, no.9, pp. 31-34 (2016)
13. Wenmin Song, Xiuyan Sun, Cheng Liu and Linlin Tang.: A New Watermarking
Frame Based on the Genetic Algorithms and Wavelet Packet Decomposition.
Journal of Information Hiding and Multimedia Signal Processing, vol. 6, number
3, pp. 613 – 621 (2015)
14. Lin, W.H., Wang, Y.R., Horng, S.J., Kao, T.W. and Pan, Y.: A blind
watermarking method using maximum wavelet coefficient quantization. Expert
Systems with Applications, vol. 36, no. 9, pp. 11509–11516 (2009)
15. Ma, B., Wang, Y., Li, C., Zhang, Z. and Huang, D.: A robust watermarking
scheme based on dual quantization of wavelet significant difference. Proceedings
of the Pacific-Rim Conference on Multimedia (PCM '12), pp. 154-159 (2012)
16. Li Chunlei, Song Xiaowei, Liu Zhoufeng, Aihua Zhang and Yang Ruimin.: A
Robust Watermarking Scheme Based on Maximum Wavelet Coefficient
Modification and Optimal Threshold Technique. Journal of Electrical and
Computer Engineering, vol. 2015, article ID 370615, pp. 11-18 (2015)
17. Bami, M., Bartolin, F., Piva, A.: Improved wavelet-based watermarking through
pixel-wise masking. IEEE Trans. Image Processing, 5(10), pp. 783-791 (2001)
18. Ersong Huang, Jinhua Liu, Ruhong Wen. Multiplicative watermarking algorithm
based on wavelet visial model. Journal of Computer Applications, vol. 3, no.8,
pp. 2165-2173 (2011)