Al-Khwarizmi

Engineering
Journal
Al-Khwarizmi Engineering Journal,Vol. 12, No. 2, P.P. 124- 133 (2016)

Digital Image Watermarking Using Arnold Scrambling and Berkeley

Wavelet Transform

Iman M.G. Alwan* Enas Muzaffer Jamel**

*,** Department of Computer Science /College of Education for Women / University of Baghdad
*Email:[email protected]
**Email: [email protected]

(Received 26 April 2015; accepted 26 November 2015)

Abstract

Embedding an identifying data into digital media such as video, audio or image is known as digital watermarking. In
this paper, a non-blind watermarking algorithm based on Berkeley Wavelet Transform is proposed. Firstly, the
embedded image is scrambled by using Arnold transform for higher security, and then the embedding process is applied
in transform domain of the host image. The experimental results show that this algorithm is invisible and has good
robustness for some common image processing operations.

Keywords: Berkeley Wavelet Transform, image watermarking, Arnold Transform, watermark attack.

1. Introduction techniques. In the first one the cover media is not

required while in the second one, the original
Due to rapid distribution of digital files such media is required for extraction.
as text, audio, still images or video over fast Mainly, two alternative watermarking methods
communication systems, the plans for verification exist; spatial and frequency (transform) domains.
of digital contents originality and copyright The spatial domain techniques are not powerful
ownership are usually demanded. Digital even to the insignificant of altering or removal
watermarking is a new solution for content attempts (attacks), hence these techniques have
protection. It draws significant interests and become less common. On the other hand, the
becomes an active research field [1]. Generally transform domain techniques have taken the most
digital watermark is the technology of adding interests by the researchers and almost all of the
verification message or hidden copyright known transforms have been used [3]. The
information called a watermark or label or transforms used are Discrete Cosine Transform
signature to the digital media so that it can be (DCT), Discrete Fourier Transform (DFT), and
extracted or detected later to make a confirmation Discrete Wavelet Transform (DWT) [4,5]. In [3]
about the object [2]. slantlet transform is adapted in watermarking
Several eligible characteristics should be process, while in [6] the contourlet transform is
available in watermarking system. They are used as a frequency domain watermarking system.
Imperceptibility, which means perceptual parity The curvelet transform based logo watermarking
between the original and the watermarked images, is proposed in [7].
Robustness, which means the immovability of the In [8] a two-dimensional triadic wavelet
watermark data against changes and alteration to transform is proposed, it’s called Berkeley
the watermarked file, and finally capacity, which Wavelet transform. It participates many
indicates the amount of information that can be characteristics like oriented, spatial localization,
saved in the cover media . To reveal the and frequency bandpass. It is formed by
watermark information, two techniques are 4-complete, orthogonal bases which the transform
followed. They are blind and non-blind and inverse transform computations are fast. Its
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

good computation and fine primary simulation The mother wavelets are piecewise constant
visually makes it a robust tool in watermarking functions, so each wavelet, 𝛽𝜃 ,𝜙 , can be totally
process. In [9], the authors proposed a represented as a matrix 𝑏𝜃,𝜙 of 3x3 dimensions
watermarking scheme based on Berkeley Wavelet [8]:
Transform. The embedding process is carried out
in the DC band of the transformed host image −1 0 1
1
only by sub-dividing it into non-overlapping 𝑏|,0 = 6
−1 0 1 …(1)
blocks, and replacing the binary watermarked −1 0 1
coefficient with only one designated coefficient of −1 2 −1
1
middle frequency diagonally, and then an inverse 𝑏|,𝑒 = 18
−1 2 −1 …(2)
Berkeley Wavelet Transform isperformed to get −1 2 −1
the watermarked image. This algorithm embeds a
limited amount of information since it is −1 1 0
1
embedded in one band which is (1/9) of the total 𝑏/,0 = 6
1 0 −1 …(3)
host image size. They used a host image of size 0 −1 1
(243x243) and the binary watermark image of size −1 −1 2
(27x27). 1
𝑏/,𝑒 = −1 2 −1 … (4)
In this paper, Berkeley wavelet transform is 18
2 −1 −1
used for grey scale image watermarking process,
the grey scale watermark image is first scrambled −1 −1 −1
1
by using Arnold transform to boost the security 𝑏−,0 = 0 0 0 …(5)
6
side, and then it is embedded in the whole 1 1 1
frequency bands of the transformed grey scale
host image. So the amount of embedded −1 −1 −1
1
information is greater. From simulation results, 𝑏−,𝑒 = 18
2 2 2 …(6)
the proposed method shows good performance in −1 −1 −1
0 −1 1
terms of Peak Signal to Noise Ratio (PSNR) of 1
𝑏\,0 = 6 1 0 −1 …(7)
the watermarked image and normalized
correlation (NC) between the watermark image −1 1 0
2 −1 −1
and the extracted one. The performance of the 1
𝑏\,𝑒 = 18 −1 2 −1 …(8)
system is also tested under various attacks −1 −1 2
including noise, JPEG compression, and cropping.
The results exhibit a good response of the Daughter wavelets βm ,n,s
θ,ϕ can be produced at
proposed system. Rest of this paper is organized various positions (i, j) in the x-y plane, and
as follows: The concepts of Berkeley Wavelet various scales, (s) through scaling and translation
Transfrom are introduced in Section 2. In Section of the mother wavelets βθ,ϕ by using the dilation
3, the proposed embedding algorithm will be
equation:
presented. Finally, the Simulation Results and 1
Discussion will be given in Section 4, followed by βm ,n,s s s
θ,ϕ = s 2 βθ,ϕ (3 x − i , 3 y − j ) …(9)
the conclusions in Section 5. The BWT uses triadic scaling, i.e. the sizes of
the daughter wavelets are scaled by powers of 3.
To form an orthogonal set, the possible
2. Berkeley Wavelet Transform translations in the x-y plane are locked to integer
multiples of the wavelet size, 3s. All of the BWT
The Berkeley wavelet transform is a wavelet wavelets have zero mean, and so a single DC term
basis, triadic, 2- dimensional transform. It is required to represent the mean value of an
presents an effective image representation. Its image:
computational prosperities are useful; it is β0 = u(x 3 , y 3) / 9 …(10)
complete, orthonormal, and a sparse code for Note that this DC term is not subject to the
images of natural type. BWT includes 4- pairs of dilation equation; instead, it is applied only once,
mother wavelets at 4- orientations. at the largest spatial scale. Since the BWT is a
At every pair, one of the wavelet is odd complete, orthonormal set, it is self-inverting.
symmetric, and the other is even symmetric. An image can be reconstructed from its BWT
Orthonormal, complete basis in 2- dimensions are coefficients using [8]:
resulted from the translation and scaling of the
total set plus a single constant term of wavelet.

125
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

𝐼 𝑥, 𝑦 = 𝜃∈Θ,𝜙 ∈Φ ∞ 𝑠
𝑚 ,𝑛,𝑠=0 𝑤𝑚 ,𝑛,𝑠 𝛽𝜃 ,∅ (3 𝑥 −
image can be made by computing the equation
𝑚 , 3𝑠 (𝑦 − 𝑛) …(11) (14) n-times iteratively.
θ,ϕ 𝐷(𝑥, 𝑦) 𝑟 = 𝐶 −1 𝐷(𝑥 ′ , 𝑦 ′ )𝑟−1 𝑚𝑜𝑑 𝑁 …(14)
where wm ,n,s are the BWT coefficients
where 𝐷(𝑥 ′ , 𝑦 ′ )(0) is the encrypted image’s pixel,
representing the image.
and 𝐷(𝑥, 𝑦)(𝑟) is a decrypted pixel by executing r
Figure (1) shows the output of the Berkeley
iterations. Figure (2) shows an example of Arnold
Wavelet Transform of the image Boat as it is image encryption with n=5 for pepper test image
obtained by the Berkeley Wavelet Transformation of size (256x256).
program.

(a) (b) (a) (b)

Fig. 1. The original image (a), the transformed Fig. 2. The original image (a), Arnold transformed
image (b).
image (b).

3. The Proposed System For embedding process, the host image should
be a power of 3𝑛 𝑥3𝑛 where n is an integer
The proposed system is based on embedding number, and the bit stream of the encrypted
grey scale visual watermark image of size (1/9) of watermark image is transformed into a sequence
the host image into the host grey scale image by wi(1)….wi(L), where L is the bit stream length,
using Berkeley Wavelet Transform (BWT). and 𝑤𝑖 𝑚 ∈ −1,1 , 𝑚 = 1, … . , 𝐿 . The
For security confirmation and robustness sequence is used as the watermark.
improvement, the watermark image is scrambled
before embedding into the host image. Arnold
A: Watermark Embedding Process
transform algorithm is adopted as a scrambling
scheme to the original watermark image. Arnold
Input: Host image, Secret image, 𝛼 the scaling
transform has periodicity process, so the image
factor.
can be reclaimed after the permutation concept. It
Output: Watermarked image.
is only appropriate for N×N digital images. It is
1. Read the host image I(N,N), the watermark
defined as
image w(m,m).
𝑥′ 1 1 𝑥
′ = mod N …(12) 2. Apply Berkeley wavelet transform to I(N,N).
𝑦 1 2 𝑦 Triadic pattern image decomposition is
where (x, y) represent the pixel coordinates of
applied.
the original image, while (x', y') represent the
3. Apply Arnold transform to the grey- scale
pixel coordinates of the encrypted image. Let C
watermark image.
indicate the matrix at the left of the right hand
4. Change the 2-D 8-bit scrambled watermark
side of equation (12), Im(x, y) indicate the original
pattern of size m x m into a binary sequence
image pixels, and I𝑚(𝑥 ′ , 𝑦 ′ )𝑛 indicate pixels of
w = w k k = 0, … . , m2 }, where w(k) ∈
the encrypted image acquired by n-times Arnold
0,1 . Then map the watermark information w
transform application. Thence, Arnold transform’s
into a bipolar vector
image encryption is [10]:
𝐼𝑚 𝑥 ′ , 𝑦 ′ = 𝐶 𝐼𝑚(𝑥, 𝑦) 𝑟−1 (𝑚𝑜𝑑 𝑁) …(13)
𝑤𝑖𝑏 = 𝑤𝑖𝑏 𝑘 𝑘 = 0, … . . , 𝑚 2 and 𝑤𝑖𝑏 (𝑘) is
Where r = 1, 2, ….. n, and 𝐼𝑚(𝑥 ′ , 𝑦 ′ )0 =
denoted as
𝐼𝑚(𝑥, 𝑦).
𝑤𝑖𝑏 𝑘 = (−1)𝑤𝑖 (𝑘) …(15)
To obtain 𝐼𝑚(𝑥, 𝑦)(𝑟−1) , the inverse matrix of C
can be multiplied at each side of equation (13). where k=0,1,…..,𝑚 2
This means, the encryption of the encrypted

126
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

5. Embed the binary sequence in the all 4.1 The Simulation Parameters Settings
transformed host image’s subbands according
to the formula The proposed algorithm was implemented and
𝑊𝑡𝐼𝑚𝑔 = 𝐻 + 𝛼 × 𝑤𝑖𝑏 …(16) tested on several standard images using
MATLAB Version 7.0.4.365, and was
where, WtImg is the watermarked image, H is the implemented with an Intel core i7, 2.7 GHz
host Berkeley Wavelet Transformed image, 𝛼 is processor. The host images are of size 243x243,
the scaling factor, lower value of it gives better and the watermark images are of size 80x80. To
quality for the watermarked image, and 𝑤𝑖𝑏 is the evaluate the qualityof the watermarked image,
watermark information stream. PSNR criteria were adopted. Greater values of
6. Apply the inverse of the Berkeley Wavelet PSNR indicate preferable quality of the
transform to get the watermarked image. watermarked image. Values of PSNR more than
7. End 30 dBs denote acceptable image quality, where no
significant change is made by watermarking
process [10]. The algorithm was also tested under
B: Watermark Extracting Process different kinds of signal attack. Figure (3a) and
(3b) displays the test host, and watermark images.
It is assumed that the original image, value Equation (18) represents the peak signal to noise
of 𝛼, and number of iteration for Arnold ratio [10].
255 2
scrambling technique n, are known for extraction 𝑃𝑆𝑁𝑅 = 10𝑙𝑜𝑔10 ( 𝑀𝑆𝐸 ) …(18)
procedure. 1
Input: Watermarked Image. 𝑀𝑆𝐸 = ( )2 (𝑝𝑖𝑗 − 𝑝𝑖𝑗′ )2 …(19)
𝑁
Output: Watermark secret image. where 𝑝𝑖𝑗 ’s stand for the values of the original
1. Apply Berkeley Wavelet transform for the pixel, 𝑝𝑖𝑗′ ’s stand for the values of the modified
watermarked image, to get the transformed pixel, and N is the image dimension. The
subbands. efficiency of the watermark extraction result is
2. Extract the embedded sequence from the all evaluated by using normalized correlation
subbands of the transformed image according coefficient (NC), for the extracted watermark W'
to the equation: and the original watermark W as:
3. 𝑤𝑖𝑏 = (𝑊𝑡𝐼𝑚𝑔 − 𝐻)/𝛼 … 𝑛
𝑖=1 𝑤 𝑖 𝑤 ′ (𝑖)
(17) 𝑁𝐶 𝑊, 𝑊 ′ = …(20)
𝑛 𝑤 (𝑖)2 𝑛 𝑤 ′ (𝑖)2
𝑖=1 𝑖=1
4. Convert the extracted sequence 𝑤𝑖𝑏 into 0, 1
sequence, reconstruct the scrambled mxm where (n×n) are the watermark dimensions. The
watermark image. unity value given exact matching between the
5. Apply the inverse Arnold transform to get the extracted watermark and the original watermark
watermark image. images, NC of about 0.7 or above is counted
6. End passable [11].

4. The Simulation Results and Discussion

In this part, the simulation parameters settings

of the proposed algorithm, the simulation results,
and discussion of the results are exhibited.

127
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

a. Test Host Images.

b. Test watermark Images.

Fig. 3. Test host and watermark images.

4.2. The Simulation Results watermark images. PSNR values of the

watermarked images are also included. The
The simulation results of the proposed Normalized Correlation (NC) is listed to indicate
algorithm are shown in Figure (4). It shows the the similarities between the original watermarks
watermarked images and the reconstructed and the extracted ones.

PSNR= 40.88 dB PSNR=40.66 dB PSNR=40.87 dB PSNR=40.7 dB

NC=1 NC=1 NC=1 NC=1

Fig. 4. The watermarked images and the reconstructed watermark images.

128
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

Various attacks are used to test the robustness (cropping), and JPEG compression. Figure (5)
of the watermark algorithm, such as noise shows results of applying various attacks to the
(Gaussian, Salt & Pepper), geometrical distortion watermarked test image.

PSNR= 25.22 dB Gaussian Noise mean=0, Crop 1/9 of image JPEG compression
Salt & Pepper noise variance=0.00005, PSNR=15.12 dB 30%
PSNR= 32.37 dB PSNR= 32.94 dB

NC= 0.8404 NC=0.8071 NC=0.6241 NC=0.8043

PSNR= 25.01 dB Gaussian Noise mean=0, Crop 1/9 of image JPEG compression
Salt & Pepper noise variance=0.0005 PSNR= 16.48 dB 30%
PSNR= 34.84 dB PSNR= 30.99 dB

NC=0.8950 NC=0.8071 NC= 0.8293 NC= 0.7491

PSNR = 24.90 dB Gaussian Noise mean=0, Crop 2/9 of image JPEG compression
Salt & Pepper noise(0.01) variance=0.00005 PSNR= 12.93 dB 30%
PSNR= 35.58 dB PSNR = 36.44 dB

129
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

NC= 0.7631 NC=0.6527 NC= 0.5391 NC= 0.7635

PSNR= 27.68 dB Gaussian Noise mean=0, Crop 3/9 of image JPEG compression
Salt & Pepper noise(0.004) variance=0.0005 PSNR= 8.14 dB 30%
PSNR= 31.85 dB PSNR= 30.99 dB

NC=0.8950 NC=0.6642 NC= 0.4417 NC= 0.7930

Fig. 5. Results of applying various attacks to the watermarked images.

The results of the proposed algorithm were

compared with [10] in which non blind digital Results of algorithm in Results of the proposed
[10] algorithm
image watermarking based on discrete wavelet
transform has been proposed. The watermark
image was encrypted by Arnold transform and
Cross chaotic sequence, the encrypted watermark
image is transformed into wavelet transform
domain, and the medium frequency coefficients
are extracted and embedded into the wavelet
transformed host image. Figure (6), shows the
PSNR and NC of both algorithm in [10] and the
proposed one respectively.
PSNR=33.243dB. PSNR=40.66 dB

Fig. 6. PSNR and NC of both algorithms in [10] and

the proposed.

130
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

4.3. The Discussion NC Normalized Correlation

PSNR Peak Signal to Noise Ratio
The simulation results show that the proposed WtImg watermarked image
algorithm offers good response in terms of w(m,m) watermark image
quantitative parameters (Peak-Signal-to-Noise-
Ratio and normalized correlation) and in terms of
visual quality. From Figure (4), the values of Greek letters
PSNR of the watermarked images are between
(40.66- 40.88) dB for the test images which is 𝛼 scaling factor
perfectly convenient for the human eyes since
values of PSNR greater than 30 dBs indicate β0 DC term of BWT
convenient image quality. Also the extracted βm ,n,s
θ,ϕ
daughter wavelet of BWT
watermark test images have normalized θ,ϕ BWT coefficients
wm ,n,s
correlation values of one which means exact
similarity between the original and the
watermarked images. The inviolability of the
6. References
watermark versus remove or degradation attempts
by attacks of different digital signal processing
operations like noise (Gaussian, Salt & Pepper), [1] Ramani K., Prasad E.V., Naidu V.L., and
different rates of geometrical distortion Ganesh D.,”Color Image Watermarking Bi-
(cropping), and JPEG compression is measured by Orthogonal Wavelet Transform”,
NC criteria. From the results of Figure (5), it is International Journal of Computer
seen that watermark can be still extracted despite Application, vol.11, no.9, pp. 25-29,
December, 2010.
the degradation of image quality due to different
types of attacks. Also from Figure (6), it can be [2] Ram B., “Digital Image Watermarking
noticed that the proposed algorithm offers better Technique Using Discrete Wavelet
imperceptible results. The improvement in the Transform and Discrete Cosine Transform”,
PSNR value for Boat image is 7.417 dB. International Journal of Advancements in
Research & Technology, Vol. 2, No.4, pp.19-
27,April-2013.
[3] Lafta M.M. and Alwan I.M.,” Watermarking
5. Conclusion in Image Using Slantlet Transform”, Iraqi
Journal of Science, Vol.52, No.2, pp. 225-
In this paper, a non-blind digital image 230, 2011.
watermarking scheme is proposed. It depends on [4] Bailey K., Curran K.,” An Evaluation of
using Berkeley Wavelet Transform (BWT), which Image Based Steganography Methods”,
is a complete orthonormal basis functions. The Multimedia Tools & Applications, vol. 30,
watermark image is firstly scrambled by Arnold no. 1,pp. 55-88,July 2006.
transform to achieve higher embedding security. [5] Gutub A., Fattani M.,” A Novel Arabic Text
The embedding process is done in the whole Steganography Method Using Letter Points
frequency subbands of the transformed host and Extensions”, WASET International
image. The evaluation of the proposed method in Conference on Computer, Information and
terms of PSNR and normalized correlation shows System Science and Engineering, Vienna
very good performance as far as invisibility. Also Austria, May 25-27, 2005.
the proposed method offers good response under [6] Taiyue W., Hongwei LI.,” A Novel
various image attacks. Scrambling Digital Image Watermarking
Algorithm Based on Counterlet Transform”,
Journal of Natural Science, vol. 19, no. 4,
Notation pp.315-322, 2014.
[7] Hien T.D., Miyara K., Nagata Y.,Nakao Z.,
BWT Berkeley Wavelet Transform and Chen Y.W.,” Curvelet Transform Based
H host image in BWT domain Logo Watermarking”, Innovative Algorithms
Im(x,y) original image pixel and Techniques in Automation, Industrial
Im(x ′ , y ′ )n pixels of encrypted image acquired Electronics and Telecommunications, pp
by n- times Arnold transform. 305-309, Springer 2007.
MSE Mean Square Error

131
Iman M.G. Alwan Al-Khwarizmi Engineering Journal, Vol. 12, No. 2, P.P. 133- 124(2016)

[8] Willmore B., Prenger R.J., Wu M. C-K, and [10] Pradhan C., RathS. and Kumar Bisoi A. ,
Gallant J.L.,” The Berkeley Wavelet ” Non Blind Digital Watermarking
Transform: A biologically-inspired Technique Using DWT and Cross Chaos”,
orthogonal wavelet transform”, MIT Press 2nd International conference on
Journals, Neural Computation, vol. 20, no.6, communication, computing, and security,
pp. 1537-1564, June 2008. India, Procedia Technology, Elsevier Ltd.,
[9] Ravindran R. P., Soman K.P.,” Berkeley no.6, pp. 897-904, 2012.
Wavelet Transform Based Image [11] Pan J.S., Huang H.C., and Jain L.C.,
Watermarking”, International Conference on ”Intelligent Watermarking Techniques”,
Advances in Recent Technologies in World Scientific Publishing Co. Pte. Ltd.,
Communication and Computing, India, pp. 2004.
357-359, 2009.

132
‫م جلة الخىارزمي الهنذسية المجلد‪ ، 12‬العذد‪ ، 2‬صفحة ‪)2016( 124 -133‬‬ ‫ايمان محمذ جعفر‬

‫العالمة المائية الرقمية بإستعمال تحىيلة ‪ Arnold‬و تحىيلة ‪Berkeley Wavelet‬‬

‫ايناس مظفر جميل**‬ ‫ايمان محمذ جعفر علىان*‬

‫*‪ **,‬قسن الحاسباث‪ /‬كليت التزبيت للبٌاث‪ /‬جاهعت بغذاد‬
‫*البزيذ األلكتزوًي‪[email protected] :‬‬
‫**البزيذاأللكتزوًي‪[email protected]:‬‬

‫الخالصة‬

‫تعزف عوليت تضويي البياًاث في وسط رقوي كولف الفذيى‪ ,‬أو الولف الصىتي ‪ ,‬أو الصىرة بالعالهت الوائيت الزقويت ويقذم هذا البحث خىارسهيت‬
‫العالهت الوائيت التي تعتوذ علً تحىيلت ‪ . Berkeley Wavelet‬في هذٍ الخىارسهيت الوقتزحت يتن اوال تشفيز الصىرة السزيت باستعوال تحىيلت ‪Arnold‬‬
‫لتحقيق هستىي أهاى أعلً‪ ,‬و هي ثن تٌفذ عوليت االخفاء في هجال التحىيلت للصىرة الوضيفت‪ .‬أظهزث الٌتائج العوليت جىدة هذٍ الخىارسهيت الوقتزحت وهتاًتها‬
‫لبعض عولياث النعالجت الصىريت‪.‬‬

‫‪133‬‬