Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.

5, October 2019

ROBUST IMAGE WATERMARKING METHOD USING

WAVELET TRANSFORM
Omar Adwan

Computer Information Systems Department, The University of Jordan, Amman – Jordan

ABSTRACT
In this paper a robust watermarking method operating in the wavelet domain for grayscale digital images
is developed. The method first computes the differences between the watermark and the HH1 sub-band of
the cover image values and then embed these differences in one of the frequency sub-bands. The results
show that embedding the watermark in the LH1 sub-band gave the best results. The results were evaluated
using the RMSE and the PSNR of both the original and the watermarked image. Although the watermark
was recovered perfectly in the ideal case, the addition of Gaussian noise, or compression of the image
using JPEG with quality less than 100 destroys the embedded watermark. Different experiments were
carried out to test the performance of the proposed method and good results were obtained.

KEYWORDS
Watermarking, data hiding, wavelet transform, frequency domain

1. INTRODUCTION
Digital watermarking is a relatively new area, and the search is still attracting many researchers in
the area. Over the past few years, digital watermarking has emerged as a leading research area to
solve problems of ownership and content authentications for digital multimedia data including
audio, image and video. [1].

Digital watermarking can be defined as the practice of hiding information called message or
watermark into a digital content such as audio, images and video, in such a way that the
embedded message should be imperceptible (difficult to notice) and robust against different
manipulations that may attempt to remove or alter it [2-4]. For example, a message might be
hidden within an image.

Watermarks serve mainly as tools for digital copyright protection and to prove the ownership of
the original work. [5]. In addition, Digital watermarking may also be used for a wide range of
applications, such as: Source tracking when different recipients get differently watermarked
content, Broadcast monitoring, i.e. television programs contains watermarked video from
television agencies, Video authentication, Software crippling on screen casting and video editing
software programs, to encourage users to purchase the full version to remove it, ID card security,
Fraud and Tamper detection, Content management on social networks, and others.

Watermarking methods can be broadly divided into two categories: spatial domain methods,
which embed the data by directly manipulating the pixel values of the cover image, and frequency
(transform) domain methods, which embed the data by modulating the frequency domain
coefficients [7].

DOI: 10.5121/sipij.2019.10503 29
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

It has been reported that embedding data in the frequency domain of an image can be much more
robust than embedding in the spatial domain (8-9].

Wavelets have become a key technique in different applications including compression (JPEG-
2000) and image watermarking [1], The multiresolution aspect of wavelets is helpful in managing
good distribution of the message in the cover in terms of robustness versus visibility. The wavelet
transform decomposes an image into multilevel spatial-frequency decomposition. Figure 1 shows
a one level decomposition. The main objective of wavelet transform is to define the powerful
wavelet basis functions and find efficient methods for their computation. Fourier methods are not
always good tools to recapture the signal or image, particularly if it is highly non–smooth [10].
A digital watermark is called robust with respect to transformations if the embedded information
may be detected reliably from the marked signal, even if degraded by any number of
transformations. Typical image degradations are JPEG compression, rotation, cropping, additive
noise, and quantization. For video content, temporal modifications and MPEG compression often
are added to this list. A digital watermark is called imperceptible if the watermarked content is
perceptually equivalent to the original; unwater marked content [11]. In general, it is easy to
create either robust watermarks or imperceptible watermarks, but the creation of both robust and
imperceptible watermarks has proven to be quite challenging. Robust imperceptible watermarks
have been proposed as a tool for the protection of digital content, for example as an
embedded no-copy-allowed flag in professional video content [11], [12].

A digital watermark is called "fragile" if it fails to be detectable after the slightest modification.
Fragile watermarks are commonly used for tamper detection (integrity proof). Modifications to an
original work that clearly are notice-able commonly are not referred to as watermarks, but as
generalized barcodes [12].

In this paper, I propose a robust image watermarking method based on Discrete Wavelet
Transform (DWT). The method first computes the differences between the watermark and the
HH1 sub-band values of the cover image, and then embeds these differences in one of the
frequency sub-bands. Different frequency sub-bands have been tested to find which frequency
sub-band can give best results.

Section 2 of this article introduces some of the previous work related to watermarking methods
namely the spatial and the frequency based methods. Section 3 presents the basics of the Discrete
Wavelet Transform DWT. Section 4 describes the proposed method for watermarking. Section 5
presents the experimental results, and finally section 6 concludes this article.

2. PREVIOUS WORK
Several methods have been proposed in literature based on both spatial and frequency domains.
The most straight-forward method of watermark embedding in the spatial domain is the Least
Significant Bit (LSB) substitution method, where the Most Significant Bit (MSB) of the
watermark is substituted with the LSB of the cover object [9], [12]. The advantages of the LSB
methods are that they are simple and can be easily applied to any image. However, any addition
of noise can defeat the watermark [12].

Frequency (or transform) domain methods became more attractive and many methods have been
proposed in literature based on the frequency domain. These methods are more robust than those
based on spatial domain [3, 8]. Most of these methods are based on the Discrete Cosine
Transform (DCT). A review of different DCT based methods can be found in [8].

30
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

Discrete Wavelet Transform (DWT) methods became more attractive in the last years. This is
because DWT can divide the image into different sub-bands, where each sub-band has some
different coefficients [13]. This gives more freedom to researchers to embed messages. Our
algorithm is based on the DWT and it will be discussed in the next section. In [16] the authors
describe a watermarking method using wavelet transform. The method embeds the wavelet
coefficients of the watermark and the image into different resolution levels. The wavelet
coefficients of the watermark are modulated using a human visual model constraint based on a
measure called saliency before the embedding process.

In [17], the authors performed one decomposition level using DWT. The watermark is added to
the HH1 sub-band. They then compare the difference between the resultant HH1 sub-band (with
watermark added) with the HH1 sub-band of the original image by calculating their cross
correlations. If there is a peak in the cross correlations, the watermark is considered detected,
otherwise they consider the HL1 and the LH1 sub-bands. If the watermark is still not detected, a
new level of the DWT is computed. This process is continued until the watermark is detected or
the last level of the DWT is reached.

In [16], the authors develop a simple watermarking method operating in the frequency domain for
grayscale digital images. The proposed method embeds the message in the LL1 sub-band
produced by the Discrete Wavelet Transform (DWT) of the original image by using the Least
Significant Bit (LSB) substitution method. Three cover images were used to test the performance
of the proposed method. Different experiments were carried out to test the performance of the
proposed method and the results show that little and unnoticeable degradation, while the
watermark was recovered perfectly, which is considered an acceptable result.

Other methods based on wavelet transform have been proposed in [1, 17 - 18]. The evaluation of
digital watermarking schemes may provide detailed information for a watermark designer or for
end-users, therefore, different evaluation strategies exist [21]. Often used by a watermark
designer is the evaluation of single properties to show, for example, an improvement. Mostly,
end-users are not interested in detailed information. They want to know if a given digital
watermarking algorithm may be used for their application scenario, and if so, which parameter
sets seems to be the best.

3. DISCRETE WAVELET TRANSFORM

Discrete Wavelet Transform (DWT) methods became more attractive in the last years [17]. DWT
provides multiresolution Analysis, where the image is analyzed at different frequencies with
different resolutions. When an image is passed through series of low pass and high pass filters,
DWT decomposes the image into a set of band limited components, called sub-bands [3]. The
decomposition is performed so that the sub-bands can be reassembled to reconstruct the original
image without error. Each decomposition level produces four non overlapping sub-bands [15-16]:
LL1, HL1, LH1, and HH1 as shown in Figure 1, where the first letter corresponds to applying
either a low pass (L) or high pass (H) filtering to the rows, and the second letter corresponds to
applying either a low pass or high pass filtering to the columns of the image; the number after the
two letters refers to the decomposition level. The LL1 sub-band represents the approximate
image, which holds most of the information of the original image, while the other three sub-bands
are the high frequency (detail) components of the original image.

31
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

LL1 HL1

LH1 HH1

Figure 1. One level image decomposition.

The LL1 band can be further decomposed for another level of decomposition and so on.

4. PROPOSED METHOD
The information to be embedded in a signal is called a digital watermark, although in some
contexts the phrase digital watermark means the difference between the watermarked signal and
the cover signal. The signal where the watermark is to be embedded is called the host signal. A
watermarking system is usually divided into three distinct steps, embedding, attack, and
detection. In embedding, an algorithm accepts the host and the data to be embedded, and
produces a watermarked signal [20].

Our proposed method is based on the wavelet transform discussed above. The method starts by
decomposing the image into its four sub-bands using the Daubechies DWT (db2). The method
first computes the differences between the watermark and the HH1 sub-band values of the cover
image, and then embeds these differences in one of the frequency sub-bands. Different frequency
sub-bands has been tested to find which frequency sub-band can give best results.

The embedding steps are as follows:

1. Read the cover image.

2. Read the watermark, W.
3. Perform DWT to the cover image.
4. Find the differences between the watermark and the LH1 values.
5. Embed the watermark in the LH1 sub-band as follows:
Diff = (LL1 – W)/k,where k is a constant (k = 5).
LH1 = LH1 + Diff.
6. Apply inverse Discrete Wavelet Transform (IDWT) to obtain the watermarked image,
WI.
The extracting steps are as follows:
1. Read the watermarked image (WI).
2. Perform DWT to WI
3. Extract the watermark, W as follows:
W = (LH1 –LL1)*k.

The robustness of the watermark was improved by introducing a watermark constant k chosen
experimentally (k = 5 in all of our experiments). I should note that larger value of k increases
robustness at the expense of visual quality of the watermarked image.

32
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

5. EXPERIMENTAL RESULTS
In order to test the performance of our proposed method, the well-known Lena images are used as
a cover image and the eye image is used as a watermark. Both images are shown in Figure 2.

Figure 2. (a) Lena (cover) image. (b) Eye (watermark) image.

Table 1 shows the Root Square Mean Square Error (RMSE) and the Peak Signal to Noise Ratio
(PSNR) of both the original and the watermarked images when the HL1, LH1 and HH1 sub-
bands are embedded. The Table shows that embedding the watermark in the LH1 sub-band gave
best results. Therefore, the LH1 sub-band will be used for embedding the watermark in our
experiments.

Table 1. RMSE and PSNR for different cover and watermarked images

Sub-band RMSE PSNR

HL1 1.71 43.4
LH1 1.53 44.4
HH1 1.60 44.1

Figure 3 shows in (a) the watermarked Lena image and in (b) the extracted watermark. The
Figure shows that the watermarked image shows little no noticeable degradation, while the
watermark was recovered perfectly.

33
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

Figure 3. (a) The watermarked Lena image. (b) The extracted watermark.

Figure 4 below, shows the watermarked image when adding Gaussian noise. The figure shows
that the watermark was recovered.

Figure 4. (a) The noised Lena image. (b) The extracted (noised) watermark.

Figure 5 shows the watermarked image when JPEG compression is applied, with quality = 50.
The figure shows that the watermark was also recovered.

34
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

Figure 5. (a) The compressed Lena image. (b) The extracted (compressed) watermark.

Figure 6 show the watermarked image when rotated by 30 degrees is applied. The watermarked
image is also cropped. The figure shows that the watermark was recovered.

Figure 6. (a) The rotated and cropped Lena image. (b) The extracted (rotated and cropped) watermark.

The LL1 sub-band represents the approximate image can be fed again in the filter bank and the
same process is repeated again which gives further sub images HH2, HL2, LH2 and LL2 .LL2 is
the next approximation image and can be further compressed as shown in Figure 7. Applying
muti-level decomposition as shown in figure 8 is left to future work.

35
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

Figure 7. Multi-level decomposition. Figure 8. Multi-level decompositions on Lena image

6. CONCLUSION
Watermarks serve as tools for digital copyright protection and to prove the ownership of the
original work. In this paper we propose a simple method for image watermarking. The method
applies both DWT and LSB techniques. The results show that embedding the watermark in the
LH1 sub-band gave the best results. The results were evaluated using the RMSE and the PSNR of
both the original and the watermarked image. Although the watermark was recovered perfectly in
the ideal case, the addition of Gaussian noise, or compression of the image using JPEG with
quality less than 100 destroys the embedded watermark.

The LL1 sub-band can be further fed the filter bank and the same process is repeated again which
gives further sub images HH2, HL2, LH2 and LL2 .LL2 is the next approximation image and can
be further compressed. Applying muti-level decomposition is left to future work.

REFERENCES

[1] J. Dugelay and S. Roche, "A servey of current watermaking techniques", in S. Katzenbeisser and F.
Petitcolas (eds), Information hiding techniques for steganography and digital watermarking, Artech
House, USA, pp. 121-148, 2000.

[2] I. Cox, M. Miller, J. Bloom, J. Fridrich and T. Kalker “Digital watermarking and steganography”,
Morgan Kaufman, 2008.

[3] R. Gozalez, R. Woods, Digital Image Processing, 3rd ed., Prentice Hall, 2008..

[4] M. Kutter and F. Hartung, "Introduction to Watermarking Techniques", in S. Katzenbeisser and F.

Petitcolas (eds), Information hiding techniques for steganography and digital watermarking, Artech
House, USA, pp. 97-120, 2000.

[5] S. Lai and F. Buonaiuti, "Copyright on the internet and watermarking", in S. Katzenbeisser and F.
Petitcolas (eds), Information hiding techniques for steganography and digital watermarking, Artech
House, USA, pp. 191-213, 2000.

[6] I. Cox, M.L. Miller, J.M.G. Linnartz, T. Kalker, “A Review of Watermarking Principles and
Practices” in Digital Signal Processing for Multimedia Systems, K.K. Parhi, T. Nishitani, eds., New
York, New York, Marcel Dekker, Inc., 1999, pp. 461-482.

36
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019
[7] U. Qidwai and C. Chen, Digital image processing: An algorithmic approach with Matlab, CRC Press,
2010.

[8] Cox, M. Miller, J. Kilian, F. Leighton and T. Shamoon, "Secure spread spectrum watermarking for
multimedia", IEEE Transactions on Image Processing, Vol. 6, No. 12, pp. 1673-1687, 1997.

[9] N. Johnson and S. Katzenbeisser, “A survey of steganographic techniques,” in S. Katzenbeisser and F.

Petitcolas (eds), Information hiding techniques for steganography and digital watermarking, Artech
House, USA, pp. 43-78, 2000.

[10] A.H.M. Jaffar Iqbal Barbhuiya1 , K. Hemachandran (2013), “Wavelet Tranformations & Its Major
Applications In Digital Image Processing”, International Journal of Engineering Research &
Technology (IJERT) Vol. 2 Issue 3, March - 2013 ISSN: 2278-0181

[11] Khan, Asifullah; Mirza, Anwar M. (October 2007). "Genetic perceptual shaping: Utilizing cover
image and conceivable attack information during watermark embedding". Information Fusion. 8 (4):
354-365. doi:10.1016/j.inffus.2005.09.007.

[12] C. Shoemaker, Hidden Bits: "A Survey of Techniques for Digital Watermarking",
http://www.vu.union.edu/~shoemakc/watermarking/, 2002. Last access: June, 2012.

[13] M. Weeks, "Digital signal processing using Matlab and Wavelets, 2nd ed.", Jones and Bartlett
publisher, 2011.

[14] D. Kundur and D. Hatzinakos, "A robust digital watermarking method using wavelet-based fusion",
in Proceeding of the International conference on image processing, Santa Barbara, pp. 544-547, 1997.

[15] X. Xia, C. Boncelet and G. Arce, "Wavelet transform based watermark for digital images", Optics
Express, Vol. 3, No. 12, pp. 497-511, 1998.

[16] O. Adwan, et al., "Simple Image Watermarking Method using Wavelet Transform", Journal of Basic
and Applied Science, Vol. 8, No. 17, pp. 98-101, 2014.

[17] B. Gunjal and S. Mali, "Secured color image watermarking technique in DWT-DCT domain",
International journal of computer science, engineering and information technology, Vol. 1, No. 3, pp.
36-44, 2011.

[18] P. Reddy, M. Prasad and D. Rao, "Robust digital watermarking of images using wavelets",
International journal of computer and electrical engineering, Vol. 1, No. 2, pp. 111-116, 2011.

[19] G. Langelaar, I. Setyawan, R.L. Lagendijk, “Watermarking Digital Image and Video Data”, in IEEE
Signal Processing Magazine, Vol. 17, pp. 20-43, 2000.

[20] Tanya Koohpayeh Araghi, Azizah B T Abdul Manaf (2017), “Evaluation of Digital Image
Watermarking Techniques“, International Conference of Reliable Information and Communication
Technology, IRICT 2017: Recent Trends in Information and Communication Technology pp 361-
368.

[21] A.S.Kapse1, Sharayu Belokar2, Yogita Gorde3, Radha Rane4, Shrutika Yewtkar, (2018) “Digital
Image Security Using Digital Watermarking”. International Research Journal of Engineering and
Technology (IRJET), Volume: 05 Issue: 03 | Mar-2018.

37
 Signal & Image Processing: An International Journal (SIPIJ) Vol.10, No.5, October 2019

AUTHORS

Omar Adwan is an Associate Professor in the Department of Computer Information

System at the University of Jordan, where he has been since 2010. From 2012 to 2016 he
served as Department Chair. He received a B.S. in Computer Science from Eastern
Michigan University in 1987. Dr Adwan received his M.S. and Ph.D. in Computer
Science majoring in Software Engineering from The George Washington University.
Currently Dr. Adwan is the Deputy Dean of Student Affairs at the University of Jordan.
Dr Adwan research interests are in Image Processing, software engineering, with focus
on software testing, software analytics, software security, intelligent software
engineering, and both data mining and machine learning.

38