International Journal of Computer and Communication Engineering, Vol. 3, No.

5, September 2014

A Digital Image Watermarking Algorithm Based on DWT

DCT and SVD
Md Saiful Islam and Ui Pil Chong

 spatial domain approaches, the watermark is embedded

Abstract—Digital watermarking techniques have been directly to the pixel locations [13], [14]. Embedding the
developed to protect the copyright of multimedia objects such as watermark in the spatial domain is the direct method. It has
text, audio, video, etc. In this paper, we propose a new digital various advantages like less computational cost, high capacity,
watermarking algorithm with gray image based on discrete
wavelet transform (DWT), 2 dimensions discrete cosine
more perceptual quality but less robust and it mainly suits for
transform (DCT) and singular value decomposition (SVD) for authentication applications. In transform domain approaches,
robust watermarking of digital images in order to protect digital a mathematical transform is applied to the original image to
media copyright efficiently. One of the major advantages of the embed watermark into the transform coefficients, then apply
proposed scheme is the robustness of the technique on wide set inverse transform to get the embedded image. It has more
of attacks. Experimental results confirm that the proposed robust, less control of perceptual quality and mainly suits for
scheme provides good image quality of watermarked images.
copyright application. The most frequent used methods are
Index Terms—Digital image watermarking, DWT, DCT discrete cosine transform (DCT) domain [15], [16], discrete
PSNR, SVD. wavelet transform (DWT) domain [17], singular value
decomposition (SVD) domain [18]. They now come into
more widespread used as they always have good robustness to
I. INTRODUCTION common image processing.
In the present globalization, the availability of the Internet In this paper a DCT DWT SVD based blind watermarking
and various image processing tools opens up to a greater technique has been used for embedding watermark. A new
degree, the possibility of downloading an image from the watermarking algorithm based on DWT, DCT and SVD, for
Internet, Manipulating it without the permission of the digital image indicate that this algorithm combines the
rightful owner. For reason such as this and many others, advantages of these three transforms. It can proof the
image authentication has become not only an active but also imperceptibility and robustness very well. Moreover, the
vital research area. Embedding watermarks [1]-[4] in both algorithm is robust to the common image process such as
signals and images can cause distortion in them. Filtering, Gaussian noise, Rotation and Salt and Pepper.
In general, a successful watermarking scheme should satisfy The remainder of the paper is organized as follows: - In
the following fundamental requirements. Section II, we briefly describe the literature of Discrete
1) Imperceptibility: the perceptual difference between the Cosine Transform, Discrete Wavelet Transform and Singular
watermarked and the original documents should be Value Decomposition related to watermarking. Section III
unnoticeable to the human eye, i.e. watermarks should presents our proposed algorithm, while the simulations and
not interfere with the media being protected. data analysis are described in Section IV. Finally, we make
2) Trustworthiness [5]–[8]: a satisfactory watermarking some conclusions about our proposed method.
scheme should also guarantee that it is impossible to
generate forged watermarks and should provide
trustworthy proof to protect the lawful ownership. II. LITERATURE REVIEW
3) Robustness [9]–[12]: an unauthorized person should not
A. Discrete Wavelet Transform
be able to destroy the watermark without also making the
document useless, i.e., watermarks should be robust to The basic idea of discrete wavelet transform (DWT) in
signal processing and intentional attacks. In particular, image process is to multi-differentiated decompose the image
after common signal processing operations have been into sub-image of different spatial domain and independent
applied to the watermarked image like filtering, frequency district. After the original image has been DWT
re-sampling, cropping, scaling, digital-to-analog, transformed, the image is decomposed into four sub-band
analog-to-digital conversions, compression, geometric images by DWT: three high frequency parts (HL, LH and HH,
transformation, rotation, etc., they should still be named detail subimages) and one low frequency part (LL,
detectable. named approximate sub-image). In Fig. 1, 2 level wavelet
Generally, watermarking can be classified into two groups: transform process of the image is shown, HL, LH, HH are the
spatial domain methods and transform domain methods. In horizontal high frequency, the vertical high frequency and the
diagonal high frequency part respectively and LL is the
Manuscript received January 20, 2014; revised March 15, 2014. This approximation low frequency part.
work was supported in part by the University of Ulsan. The energy of the high-frequency part (horizontal, vertical
The authors are with the University of Ulsan, Ulsan, South Korea (e-mail: and diagonal part) is less, which represent the information of
saiful05eee@yahoo.com, upchong@ulsan.ac.kr).

DOI: 10.7763/IJCCE.2014.V3.349 356

 International Journal of Computer and Communication Engineering, Vol. 3, No. 5, September 2014

the original image, such as the texture, edge, etc. The low  1
frequency part concentrates most of the energy of the image  ,v  0
 N
and represents an important component and it can be  (v )  
decomposed continuously. The energy of the image is  2 , v  1, 2 ,......., N  1
diffused better and the stronger image intensity can be 
 N
embedded, with the more levels the image is decomposed by
wavelet transform. Hence, the wavelet decomposing levels C. Singular Value Decomposition
adopted in the algorithms can be chosen as far as possible. The singular value decomposition (SVD) is a factorization
of a real or complex matrix, with many useful applications in
LL HL signal processing and statistics.
The fundamental properties of SVD from the viewpoint of
LH HH
Fig. 1. Wavelet decomposition.
image processing applications are: i) the singular values (SVs)
of an image have very good stability, i.e., when a small
B. Discrete Cosine Transform perturbation is added to an image, its SVs do not change
The Discrete Cosine Transform is a very popular transform significantly; and ii) SVs represent intrinsic algebraic image
function that transforms a signal from spatial domain to properties.
frequency domain and it has been used in JPEG standard for In this section, we describe a watermark casting and
image compression due to good performance. As a real detection scheme based on the SVD.
transform, DCT transforms real data into real spectrum and From the viewpoint of linear algebra, we can observe that a
therefore avoids the problem of redundancy. The popular discrete image is an array of nonnegative scalar entries, which
block-based DCT transform segments an image may be regarded as a matrix. Let such an image be denoted by
non-overlapping block and applies DCT to each block. This A. Without loss of generality, we assume in the subsequent
result in giving three frequency sub-bands: low frequency sub discussions that A is a square image, denoted by A  E N  N ,
band, mid-frequency sub-band and high frequency sub-band. where E represents either the real number domain or the
DCT-based watermarking is based on two main facts. The complex number domain. The SVD of A is defined as
first one is that most of the signal energy lies at
A  X T
H
low-frequencies sub band which contains the most important
parts of the image and second one is that high frequency
components of the image are usually removed through where X  E N  N and T  E N  N are unitary matrices and
compression and noise attacks [19]. E
N N
is a diagonal matrix with nonnegative numbers on
There are four established types of DCT’s, i.e., DCT-I,
the diagonal and zeros on the off diagonal. The nonnegative
DCT-II, DCT-III, and DCT-IV. The DCT-II is widely applied
components of  represent the luminance value of the image.
in signal processing because it is asymptotically equivalent to
Changing them slightly does not change the image quality and
the Karhunen–Loeve Transform (KLT) for Markov-1 signals
they also don’t change much even on attacks, watermarking
with a correlation coefficient that is close to one [20]. For
algorithms normally make use of these two properties.
example, JPEG image compression is also based on the
The unique property of the SVD transform is that the
DCT-II [21]. The two-dimensional DCT is usually used in
potential N2 degrees of freedom (DOF) or samples in the
digital image processing. Given an image A of size N×N, the
original image now get mapped into
DCT of the image is defined as:
M 1 N 1
  N DOF
C ( u , v )   ( u ) ( v )   f ( x, y ) 
x0 y0 N ( N  1)
X  DOF
  ( 2 x  1) u    ( 2 y  1) v  2
co s   co s  
 2M   2N 
N ( N  1)
T  DOF
And the inverse transform is defined as 2
M 1 N 1
Totaling N2 DOF
f ( x, y )     ( u ) ( v ) C ( u , v ) 
u0 v0
SVD has many good mathematical characteristics.
  ( 2 x  1) u    ( 2 y  1) v 
co s   co s  
 2M   2N 
III. PROPOSED METHOD
where
This section presents the methods for embedding and
 1 extraction of hidden data. In this paper a DCT DWT SVD
 ,u  0 based blind watermarking technique has been used for
 M
 (u )   embedding watermark. We use the DCT DWT SVD for host
 2
, u  1, 2 ,......., M  1
image and we select the middle frequency to embed

 M watermark.

357
 International Journal of Computer and Communication Engineering, Vol. 3, No. 5, September 2014

The main task of this work has performed into following Perceptual quality of the watermarked image is measured
steps: by calculating PSNR between host and watermarked image, at
 Watermark Embedding the receiver side, watermark is extracted from the
1) Apply one-level Haar DWT to decompose the host image watermarked image. Extracted watermark is evaluated by
A, into four sub-bands i.e. ALL, AHL, ALH, and AHH. measuring its correlation with the original watermark. The
2) Consider AHL and is divided into 8×8 square blocks. PSNR value is calculated at different gain factor, when the
Perform 2D DCT to each block, collect the DC value of gain factor value is to be high the PSNR value of the image
each DCT coefficient matrix D1(x, y) together to get a increases (Shown in Table I).
new matrix M1.
3) Now consider ALH and find the Coefficient matrix D2(x, y)
and another new matrix M2, same as step 2.
4) Apply SVD to M1 and M2, obtain M1=U1S1V1T and
M2=U2S2V2T.
5) Let B of size 64×64 to represent the watermark image.
Divide the B into two parts: B1 and B2.
6) Modify the singular values S1 and S2 (in step 5) with B1
(a) Original (b) Watermarked
and B2 respectively and apply SVD to them,
S1+αB1=U1*S1*V1T* and S2+αB2=U2*S2*V2T*
7) For the coefficient matrix D1(x, y) in step 2 and D2(x, y) in
step 3, change each DC value to M1*(x, y) and M2*(x, y),
obtain new coefficient matrix D1*(x, y) and D2*(x, y)
respectively. Apply inverse DCT to each D1*(x, y) and
D2*(x, y) to produce the watermarked middle frequency
band AHL* and ALH* (c) Rotation (d) Salt and pepper
8) The watermarked image, AW is obtained by performing
the inverse DWT using two sets of modified DWT
coefficient (AHL* and ALH*) and two sets of non-modified
DWT coefficient (ALL and AHH).
 Watermark Extraction
1) Apply one-level Haar DWT to decompose the
Watermarked image (possibly attack) AW into four
sub-bands: ALL, AHL**, ALH**, and AHH. (e) Filtering (f) Gaussian noise
2) Divide both of AHL** and ALH** into 8×8 square blocks Fig. 2. Grayscale image "Lena".
separately, apply DCT to each block. Collect the DC
value to get matrix M1** for AHL** and M2** for ALH**.
3) Apply SVD to M1** and M2**, i.e. M1**= U1**S1**V1T** and
M2**= U2**S2**V2T**.
4) Compute C1 = U1*S1**V1T* and C2 = U2*S2**V2T*
5) Extract the watermark image from each sub-band, i.e.,
B1*= (C1- S1)/ α and B2*= (C2-S2).
6) We get the watermark image by combining the results of
step 5: B*= B1*+ B2*. (a) Original (b) Watermarked

IV. SIMULATION AND ANALYSIS

In order to testing the robustness of the proposed algorithm,
substantial testing are performed. In the simulation, we test
different manipulations on the four well-known and standard
grayscale image that are "Lenna", "Boat", "Opera House" and
"Pepper". The original images (host image) are shown in Fig. (c) Rotation (d) Salt and pepper
2(a), Fig. 3(a), Fig. 4(a) and Fig. 5(a) and the watermarked are
shown in Fig. 2(b), Fig. 3(b), Fig. 4(b) and Fig. 5(b)
respectively. The watermark, as shown in Fig. 6(a), is used in
our simulation. Simulation results show that the quality of
watermarked image is promising.
To test and verify the robustness of our watermarking
algorithm, the watermarked image is attacked by Gaussian
Noise, Filtering, Rotation and Salt and Pepper. These are
(e) Filtering (f) Gaussian Noise
shown in Fig. 2-Fig. 5. Fig. 3. Grayscale image "Baboon".

358
 International Journal of Computer and Communication Engineering, Vol. 3, No. 5, September 2014

V. CONCLUSION
In this paper, a novel watermarking method based on
DWT-DCT-SVD is proposed. This novel method gives
successful results comparing to methods using different cover
images. Results show that the new method is very robust
against different attacks like Gaussian Noise, Salt and Pepper,
(a) Original (b) Watermarked filtering and Rotation. Therefore, the proposed algorithm is a
good method for authentication of image materials.

ACKNOWLEDGMENT
This work is supported by University of Ulsan.

REFERENCES
(c) Rotation (d) Salt and pepper [1] C. Rey and J. L. Dugelay, “A survey of watermarking algorithms for
image authentication,” Journal on Applied Signal Processing, vol.
2002, issue 6, pp. 613-621, 2002.
[2] G. Voyatzis, N. Nikolaidis, and I. Pitas, “Digital watermarking: an
overview,” in Proc. Ninth European Signal Processing Conference,
Rhodes, Greece, September 8–11, 1998, pp. 9–12.
[3] V. Potdar, S. Han, and E. Chang, “A survey of digital image
watermarking techniques,” in Proc. 3rd IEEE-International
Conference on Industrial Informatics, Frontier Technologies for the
Future of Industry and Business, Perth, WA, August 10, 2005, pp.
(e) Filtering (f) Gaussian Noise
709-716.
Fig. 4. Grayscale image "Opera House". [4] N. J. Harish, B. B. S. Kumar, and A. Kusagur, “Hybrid robust
watermarking technique based on DWT, DCT and SVD,”
International Journal of Advanced Electrical and Electronics
Engineering, vol. 2, issue 5, 2013.
[5] J. Zhao, E. Koch, and C. Luo, “Digital watermarking in business today
and tomorrow,” Commun. ACM, vol. 41, no. 7, pp. 67–72, 1998.
[6] W. Zeng, “Digital watermarking and data hiding: technologies and
applications,” in Proc. ICISAS, 1998, vol. 3, pp. 223–229.
[7] S. Craver, N. Memon, B. L. Yeo, and M. M. Yeung, “Resolving
rightful ownerships with invisible watermarking techniques:
(a) Original (b) Watermarked limitations, attacks and implications,” IEEE J. Select. Areas Commun.,
vol. 16, no. 4, pp. 573–586, 1998.
[8] L. T. Qiao and K. Nahrstedt, “Watermarking schemes and protocols for
protecting rightful ownership and customer’s rights,” J. Vis. Commun.
Image Represent., vol. 9, no. 3, pp. 194–210, 1998.
[9] I. J. Cox, J. Kilian, F. T. Leighton, and T. Shamoon, “Secure spread
spectrum watermarking for multimedia,” IEEE Trans. Image
Processing, vol. 6, pp. 1673–1687, Dec. 1997.
[10] N. Nikolaidis and I. Pitas, “Copyright protection of images using
(c) Rotation (d) Salt and pepper robust digital signatures,” in Proc. ICASSP, May 1996, vol. 4, pp.
2168–2171.
[11] N. Nikolaids and I. Pitas, “Robust image watermarking in the spatial
domain,” Signal Process., vol. 66, no. 3, pp. 385–403, 1998.
[12] M. Barni, F. Bartolini, V. Cappellini, and A. Piva, “A DCT-domain
system for robust image watermarking,” Signal Process., vol. 66, no. 3,
pp. 357–372, 1998.
[13] L. Xie, G. R. Arce, A. Bash, and E. B. Basch, “Image enhancement
toward soft image authentication,” in Proc. IEEE International
(e) Filtering (f) Gaussian Noise Conference on Multimedia and Expo (ICME), 2000, vol. 1, pp.
Fig. 5. Grayscale image "Boat". 497-500.
[14] J. Fridirich, “Image watermarking for tamper detection,” in Proc. IEEE
International Conference on Image Processing (ICIP), 1998, vol. 2,
pp. 404-408.
[15] C. Y. Lin and S. F. Chang, “Seme-fragile watermarking for
authenticating JPEG visual content,” in Proc. SPIE Int. Conf. Security
and Watermarking of Multimedia Contents II, 2000, vol. 397, pp.
140-151.
[16] M. Wu and B. Liu, “Watermarking for image authentication,” in Proc.
(a) Original (b) Recovered IEEE Int. Conf. Image Processing (ICIP), 2000, vol. 2, pp. 437-441.
Fig. 6. Watermark image. [17] D. Kundur and D. Hatzinakos, “Digital watermarking for telltale
tamper proofing and authentication,” Proceedings of the IEEE, vol. 87,
issue 7, pp. 1167-1180, July 1999.
TABLE I: DIFFERENT VALUE OF PSNR FOR DIFFERENT IMAGES
[18] R. Sun, H. Sun, and T. Yao, “A SVD and quantization based
Image PSNR(in dB)
semi-fragile watermarking for image authentication,” in Proc. Int.
Lena 51.318 Conf. Signal Processing (ICSP), 2002, vol. 2, pp. 26-30.
Baboon 51.209 [19] C. C. Lai and C. C. Tsai, “Digital image watermarking using discrete
Opera 51.193 wavelet transform and singular value decomposition,” IEEE
Boat 50.998 Transactions on Instrumentation and Measurement, vol. 59, no. 11, pp.
3060-3063, Nov. 2010.

359
 International Journal of Computer and Communication Engineering, Vol. 3, No. 5, September 2014

[20] K. R. Rao and P. Yip, Discrete Cosine Transform: Algorithms, Ui Pil Chong received the B.S. degree in electrical
Advantages, and Applications, Boston, MA: Academic, 1990. engineering from University of Ulsan, Korea, in1978,
[21] W. Pennebaker and J. Mitchell, JPEG Still Image Data Compression and M.S. degree in electrical engineering from Korea
Standard, New York: Van Nostrand Reinhold, 1993. University, Seoul, Korea in 1980. He studied in field
of computer engineering of Oregon State University
and received M. S. degree in 1985 and received Ph. D.
Md Saiful Islam is a lecturer of electrical and degree at New York University (POLY), NY, USA
electronic engineering at Premier University, 1997.
Bangladesh. He received B.Sc. degrees in electrical In January of 1997, Dr. Chong joined the School of
and electronic engineering from the Chittagong Computer Engineering and Information Technology of the University of
University of Engineering and Technology, Ulsan in Ulsan City, Korea where he has been promoted to be full professor
Bangladesh in 2010. since 2006. He has published more than 210 journal papers, conference
Now, he is doing his M.Sc. in School of Electrical papers in the area of digital signal processing, fault detection and diagnosis
and Computer Engineering at University of Ulsan, in the plants, biomedical engineering, computer music, and multimedia
South Korea and working as research assistant in applications. He also holds the 10 Korean patents. He is a member of IEEE
digital signal processing and new media lab (DSPNM) at the same since 1993 and Eta Kappa Nu since 1995. He served as the president of
university. His research interests include digital signal processing, image Korean Society of Digital Arts Media from 2004 to 2007. Also he served as
processing, and green energy. He is currently researching the following the dean of Graduate School of Information and Communication from 2004
advanced field: radar signal processing. to 2006.
Currently, he is the head of Whale Research Institute in University of
Ulsan and vice president of the Korea Institute of Signal Processing and
Systems.

360