A Quantization based Robust Image Watermarking Algorithm in Wavelet
Vidyasagar M. Potdar, Elizabeth Chang
School of Information Systems, Curtin University of Technology, Perth, Western Australia
e-mail: vidyasagar.potdar, email@example.com
Abstract — Watermarking which belong to the field of • Section VII Discussion will be given of the ex-
information hiding has seen a lot of research interest recently. isting literature.
Watermarking is used in B2B and B2C e-commerce markets • Section VII will continue our algorithm proposal.
for a variety of reasons including security, content protection,
• Section VIII will show our results.
copyright management, trust management, content
authentication, tamper detection and privacy. Recently many • Section IX concludes.
watermarking techniques have been proposed to support these
applications but one major issue with most of the II. DIGITAL WATERMARKING TECHNIQUE
watermarking techniques it that these techniques fail in the
presence of severe attacks. This has been a major threat to The process of embedding a watermark in a multimedia
content providers in the e-commerce scenario because if the object 1 is termed as watermarking. A Watermark can be
digital content is dramatically changed then it would be considered as a kind of a signature, which reveals the owner
difficult to prove the existence of a watermark in it and of the multimedia object. Content providers want to embed
consequently its ownership. To tackle this security threat watermarks in their multimedia objects (digital content) for
towards ownership issues we propose a quantization based several reasons like copyright protection, content authenti-
watermarking algorithm which offers incredible performance
cation, tamper detection etc. A watermarking algorithm
in presence of malicious attacks which try to remove ownership
information. The proposed scheme is backed up with excellent embeds a visible or invisible watermark in a given multi-
results. media object. The embedding process is guided by use of a
secret key, which decides the locations within the multimedia
Index Terms— Watermark Detection, Watermarking, DWT, object (image) where the watermark would be embedded.
Quantization Once the watermark is embedded it can experience several
attacks because the multimedia object can be digitally proc-
I. INTRODUCTION essed. The attacks can be unintentional (in the case of im-
ages, low pass filtering or gamma correction or compression)
Watermarking is a method of hiding proprietary informa- or intentional (like cropping). Hence, the watermark has to
tion in digital media like photographs, digital music, or be very robust against all these possible attacks. When the
digital video. The ease with which digital content can be owner wants to check the watermarks in the possibly at-
exchanged over the Internet has created copyright in- tacked and distorted multimedia object, s/he relies on the
fringement issues. Copyrighted material can be easily ex- secret key that was used to embed the watermark. Using the
changed over peer-to-peer networks, and this has caused secret key, the embedded watermark sequence can be ex-
major concerns for those content providers who produce tracted. This extracted watermark may or may not resemble
these digital contents. In order to protect the interest of the the original watermark, because the object might have been
content providers these digital contents can be watermarked. attacked.
In this paper we provide a survey of the latest techniques that
are employed to watermark images. Image watermarking
techniques can be applied to digital videos as well. How-
ever, this paper will limited to the context of the image do-
The paper is organized into the following sections:
• Section II - We describe the process of digital
• Section III - We discuss the main requirements of
• Section IV - Information on some main applica-
tions of watermarking will follow.
• Section V - Will discuss different kinds of wa-
termarks that could be embedded.
• Section VI introduces wavelet transform.
Multimedia object refers to images, video and audio clips or any digital
content that can be used for the purpose of information hiding.
Hence, to validate the existence of a watermark, either the E. Unambiguity
original object is used to compare and ascertain the water-
The extracted watermark must identify the ownership
mark signal (non-blind watermarking), or a correlation
unambiguously, which means the errors in an extracted wa-
measure is used to detect the strength of the watermark signal
termark must be as low as possible. Craver et al. (1997)
from the extracted watermark (blind watermarking). In
argued that an attacker can watermark a watermarked image
correlation based detection, the original watermark sequence
and hence claim ownership, is called a counterfeiting attack
is compared with the extracted watermark sequence, and a
. So a watermarking algorithm should be able to resist
statistical correlation test is used to determine the existence
of the watermark.
IV. WATERMARKING APPLICATIONS
III. REQUIREMENTS OF DIGITAL WATERMARKING
Before beginning the discussion on watermarking algo-
There are three main requirements of digital rithms we discuss the main B2B and B2C watermarking
watermarking. They are transparency, robustness and applications.
capacity. A. Copyright Protection
A. Transparency or Fidelity Watermarking can be used to protecting redistribution of
copyrighted material over the untrusted network like Internet
The digital watermark should not affect the quality of the or peer-to-peer (P2P) networks. Content aware networks
original image after it is watermarked. Cox et al. (2002) (p2p) could incorporate watermarking technologies to report
defines transparency or fidelity as ‘perceptual similarity or filter out copyrighted material from such networks.
between the original and the watermarked versions of the
cover work’ . Watermarking should not introduce visible B. Content Archiving
distortions because if such distortions are introduced it re- Watermarking can be used to insert a digital object iden-
duces the commercial value of the image. tifier or a serial number to help archive digital contents like
B. Robustness images, audio or video. It can also be used for classifying
and organizing digital contents. Normally digital contents
Cox et al. (2002) defines robustness as the ‘ability to detect are identified by their file names; however, this is a very
the watermark after common signal processing operations’ fragile technique, as file names can be easily changed.
. Watermarks could be removed intentionally or unin- Therefore, embedding the object identifier within the object
tentionally by simple image processing operations like con- itself reduces the possibility of tampering, and hence it can
trast or brightness enhancement, gamma correction etc. be effectively used in archiving systems.
Hence watermarks should be robust against a variety of such
attacks. Stirmark2 classifies attacks into four basic catego- C. Meta-data Insertion
ries, attacks that try to remove watermarks totally, attacks Meta-data refers to data that describes data. Images can be
that try to remove the synchronization between the embedder labelled with its content and can be used in search engines.
and the detector, cryptographic attacks and protocol attacks. Audio files, for example, can carry the lyrics or the name of a
C. Capacity or Data Payload singer. Journalists could use photographs of an incident to
insert the cover story of the respective news. Medical X-rays
Cox et al. (2002) define capacity or data payload as ‘the could store patient records.
number of bits a watermark encodes within a unit of time or
work’ . This property describes how much data should be D. Broadcast Monitoring
embedded as a watermark to successfully detect during ex- Broadcast Monitoring refers to the technique of
traction. Watermark should be able to carry enough infor- cross-verifying whether the content that was supposed to be
mation to represent the uniqueness of the image. Different broadcasted (on TV or Radio) has really been broadcasted or
applications have different payload requirements . not. Watermarking can also be used for broadcast monitor-
D. Security ing. This has major application in commercial advertisement
broadcasting where the advertiser wants to monitor whether
According to Kerckhoff’s principle the security of a their advertisement was actually broadcasted at the right time
cryptosystem depends on the secrecy of the key and not on and for right duration.
the cryptographic algorithm. Same rule applies to water-
marking algorithms, i.e. the watermarking algorithms must E. Tamper Detection
be public but watermark embedding should be based on a Digital content can be detected for tampering by embed-
secret key . ding fragile watermarks. If the fragile watermark is de-
stroyed or degraded, it indicates the presence of tampering
and hence the digital content cannot be trusted. Tamper de-
2 tection is very important for some applications that involve
Stirmark is a benchmark to test robustness of watermarking algorithms.
highly sensitive data like satellite imagery or medical im-
http://www.petitcolas.net/fabien/watermarking/stirmark agery. Tamper detection is also useful in a court of law
where digital images could be used as a forensic tool to prove number of matching pixel values in compared images and
whether the image has been tampered with. D denotes the number of different pixel values in compared
F. Digital Fingerprinting
Digital Fingerprinting is a technique used to detect the VI. WAVELET PRELIMINARIES
owner of the digital content. Fingerprints are unique to the
In the last few years wavelet transform has been widely
owner of the digital content. Hence, a single digital object
studied in signal processing in general and image compres-
can have different fingerprints, because they belong to dif-
sion in particular. In some applications wavelet based wa-
termarking schemes outperforms DCT based approaches.
One such scheme is proposed here . Before discussing the
V. WATERMARKS AND WATERMARK DETECTION
literature we point out the advantages of the DWT transform
Basically there are three main types of watermarks that can and its characteristics.
be embedded within an image.
A. Characteristics of DWT
A. Pseudo-Random Gaussian Sequence 1) The wavelet transform decomposes the image into three
A Gaussian sequence watermark is a sequence of numbers spatial directions i.e. horizontal, vertical and diagonal.
comprising 1 and -1, which has an equal number of 1’s and Hence wavelets reflect the anisotropic properties of
-1’s. It is termed as a watermark with zero mean and one HVS more precisely .
variation. Such watermarks are used for objective detection 2) Wavelet Transform is computationally efficient and can
using a correlation measure. be implemented by using simple filter convolution.
3) Magnitude of DWT coefficients is larger in the lowest
B. Binary Image or Grey Scale Image Watermarks bands (LL) at each level of decomposition and is smaller
for other bands (HH, LH, HL) .
Some watermarking algorithms embed meaningful data in
4) The larger the magnitude of the wavelet coefficient the
the form of a logo image instead of a pseudo-random gaus-
more significant it is.
sian sequence. Such watermarks are termed as binary image
5) Watermark detection at lower resolutions is computa-
watermarks or grey scale watermarks. Such watermarks are
tionally effective because at every successive resolution
used for subjective detection. Based on the type of water-
level there are few frequency bands involved.
mark embedded, an appropriate decoder has to be designed
6) High resolution sub-bands helps to easily locate edge
to detect the presence of watermark.
and textures patterns in an image.
Watermark Detection B. Advantages of DWT over DCT
If it’s a pseudo random gaussian sequence then hypothesis 1) Wavelet transform understands the HVS more closely
testing is done to detect the presence of watermark. Suppose than the DCT.
W is the original watermark bit sequence and W’ is the ex- 2) Wavelet coded image is a multi-resolution description of
tracted watermark bit sequence then we can calculate bit image. Hence an image can be shown at different levels
error rate (BER) to detect the presence of watermark. If the of resolution and can be sequentially processed from
BER is zero it indicates the presence of watermark however low resolution to high resolution.
if it is one it indicates absence of watermark. BER is cal- 3) Visual artefacts introduced by wavelet coded images are
culated as follows. Suppose D is the retrieved signal and N is less evident compared to DCT because wavelet trans-
the number of bits in watermark then: form doesn’t decompose the image into blocks for
processing. At high compression ratios blocking arte-
⎪1 if Wi ≠ Wi ⎪
BER (W , W ' ) =
∑D facts are noticeable in DCT; however, in wavelet coded
images it’s much clearer.
⎪0 if Wi = Wi ⎪
⎩ ⎭ N 4) DFT and DCT are full frame transform and hence any
change in the transform coefficients affects the entire
Normalized Correlation Coefficient can also be used to image except if DCT is implemented using a block
detect the presence of watermark. based approach. However DWT has spatial frequency
NC (W , W ' ) =
∑W W ' locality which means if signal is embedded it will affect
the image locally . Hence a wavelet transform
∑Wi 2 ∑W 'i2 provides both frequency and spatial description for an
To detect the presence of a logo watermark it can be C. DWT Decomposition
achieved by doing subjective comparison by visual inspec-
tion, whereas an objective analysis could be made by cal- Wavelet transforms use wavelet filters to transform the
culating similarity, a measure which would be outlined as image in the wavelet domain. There are many available fil-
S ters; however, the most commonly used filters for water-
follows: S R= where S denotes the marking are Haar Wavelet Filter, Daubechies Orthogonal
( S + D) Filters and Daubechies Bi-Orthogonal Filters. Each of these
filters decomposes the image into several frequencies. Single nique. The authors argue that if the watermark is embedded
level decomposition gives four frequency representations of in the low frequency components it is robust against low pass
the images. These four representations are called the LL, LH, filtering, lossy compression and geometric distortions on the
HL, HH sub-bands as shown in Fig.2. other hand if the watermark is embedded in high frequency
components it’s robust against contrast and brightness ad-
justment, gamma correction, histogram equalization and
cropping and vice-versa. Thus to achieve overall robustness
against a large number of attacks the authors propose to
LL1 HL1 embed multiple watermarks in low frequency and high fre-
quency bands of DWT .
Kundur and Hatzinakos (2004) present image fusion wa-
termarking technique. They use salient features of the image
to embed the watermark. They use a saliency measure to
LH1 HH1 identify the watermark strength and later embed the water-
mark additively. Normalized correlation is used to evaluate
the robustness of the extracted watermark. Later the authors
propose another technique termed as FuseMark  which
includes minimum variance fusion for watermark extraction.
Fig.2 Single level Decomposition using DWT Here they propose to use a watermark image whose size is a
factor of the host by 2xy.
VII. RELATED WORK Tao and Eskicioglu (2004) present an optimal wavelet
based watermarking technique. They embed binary logo
Wavelet based watermarking techniques are gaining at- watermark in all the four bands. But they embed the wa-
tention because of the upcoming JPEG2000 standard. This termarks with variable scaling factor in different bands. The
standard used the Discrete Wavelet Transform to compress scaling factor is high for the LL sub band but for the other
the image. In this section we discuss wavelet based water- three bands its lower. The quality of the extracted watermark
marking algorithms. We classify these algorithms based on is determined by Similarity Ratio measurement for objective
their decoder requirements as Blind Detection or Non-blind calculation .
Detection. As mentioned earlier blind detection doesn’t Ganic and Eskicioglu (2005) inspired by Raval and Rege
require the original image for detecting the watermarks; (2003) propose a multiple watermarking technique based on
however, non-blind detection requires the original image. DWT and Singular Value Decomposition (SVD). They ar-
After initial discussion on these algorithms we focus on the gue that the watermark embedded by Raval and Rege (2003)
main quantization based watermarking algorithms which use scheme is visible in some parts of the image especially in the
a logo as a watermark. low frequency areas, which reduces the commercial value of
the image. Hence they generalize their technique by using all
A. DWT based Non-Blind Watermark Detection the four sub bands and embedding the watermark in SVD
domain. The core technique is to decompose an image into
This technique requires the original image for detecting
four sub bands and then applying SVD to each band. The
the watermark. Most of the techniques surveyed in this sec-
watermark is actually embedded by modifying the singular
tion use a smaller image as a watermark and hence cannot
values from SVD . Fig. 3 gives complete technical de-
use correlation based detectors for detecting the watermark;
tails of the parameters used in these algorithms.
as a result they rely on the original image for informed de-
All these algorithms discussed here require the original
tection. The size of the watermark image (normally a logo) is
image for detecting the presence of watermark which is not
smaller compared to the host image.
feasible in all scenarios. Hence we now discuss some blind
Xia et al. (1997) present a wavelet based non-blind wa-
watermarking algorithms which embed an image logo as a
termarking technique for still images where watermarks are
added to all bands except the approximation band . A
multi-resolution based approach with binary watermarks is B. DWT based Blind Watermark Detection
presented by Hsu and Wu . Here both the watermark
logo as well as the host image is decomposed into sub bands Voyatzis and Pitas (1999) who presented the ‘toral auto-
and later embedded. Watermark is subjectively detected by morphism’ concept provide a technique to embed binary
visual inspection however an objective detection is em- logo as a watermark, which can be detected using visual
ployed by using normalized correlation. models as well as by statistical means. So in case the image
. Lu et al. (2001) present another robust watermarking is degraded too much and the logo is not visible, it can be
technique based on image fusion . They embed a grey- detected statistically using correlation . Watermark em-
scale and binary watermark which is modulated using the bedding is based on a chaotic (mixing) system. Original
‘toral automorphism’ described in . Watermark is em- image is not required for watermark detection. However the
bedded additively. The novelty of this technique lies in the watermark is embedded in spatial domain by modifying the
use of secret image instead of host image for watermark ex- pixel or luminance values. A similar approach is presented
traction and use of image dependent and image independent for the wavelet domain , where the authors propose a
permutations to de-correlate the watermark logos . Ra- watermarking algorithm based on chaotic encryption.
val and Rege (2003) present a multiple watermarking tech-
Some recent work in quantization based watermarking is called the approximate band and it contains most of the en-
more interesting where a logo is embedded as a watermark ergy. In the algorithm we decompose the image into four
and the detection is blind. Tsai et al. (2000) present a scalar levels and embed the watermark in HL, LH, HH sub-bands.
quantization based watermarking algorithm. They embed Here we assume the size of the watermark logo is in multiple
the watermark in the middle and low frequency components of the sub-band size. We propose a quantization based wa-
of the wavelet sub-bands. The algorithm shows robustness termarking algorithm. We incorporate implicit visual
against JPEG compression only. However, it’s not robust masking by embedding the watermark in the LH, HL and HH
against other geometric and image processing attacks is not sub-bands.
Chen et al. (2003) present another quantization based A. Embedding Algorithm
watermarking algorithm which improves on the Tsai algo- Suppose the original image is an 8 bit grey scale image of
rithm by incorporating variable quantization, resistance size S x S and watermark is a binary image of size S/16 x
against a wide range of attacks like blurring, noising, sharp- S/16.
ening, scaling, cropping and compression. They employ the
concept of digital signature and time stamps to overcome Step 1: Wavelet Transform
counterfeit attacks. Technical details of these algorithms are The original image I is decomposed by one level to obtain
presented in Table 1. LL, LH, HL and HH sub-bands.
Step 2: Block Mean Intensity Calculation
For each sub-band except the LL sub-band, starting at the top
left corner divide the sub-band into non-overlapping blocks
DWT Based Non-Blind Watermarking Algorithms
of 8x8 and calculate their mean intensity values of the
Grey High and wavelet coefficients. Then find the minimum and maximum
Haar 1 Scale Low pass
Image bands mean intensity values and use that to construct a quantization
All Bands table. Divide the mean intensity values into N equal sets. N is
Eskicioglu 2004 Logo
Kundur and Daubechies
a positive constant that can be chosen to control the trade-off
Hatzinakos 2004 10 pt wavelet
bands between robustness and transparency. For example if the
2nd Level range of mean intensity values is [20, 350] and the quanti-
Raval and Rege Not Speci- Binary
2 LL and
2003 fied Logo
HH bands zation step size being 40 we can generate a quantization table
Hsu and Wu Daubechies Binary Not
like this and assign binary values to it. For example:
Chae and Man- Not
Haar 1 Scale 0- 41- 81- 121- 161- 201- 241- 281- 321-
junath 1998 Specified
Image 40 80 120 160 200 240 280 320 360
Kundur and Binary Low Pass 1 0 1 0 1 0 1 0 1
Not Spec. 4
Hatzinakos 1997 Logo Bands
DWT Based Blind Watermarking Algorithms
Chen et al. 2003 Not Spec. 4
LL Step 3: Watermark Permutation
Binary Before embedding the watermark Wo , the watermark
Tsai et al. 2000 Not Spec. 3 LL
Binary Visually should be permuted to W p by a seed S and a pseudo random
Lu et al. Not and Grey Signifi-
1999 Spec. scale
permutation function f (.) . The seed is secret.
Daubechies Not Water- High pass
10 pt wavelet Spec. mark bands Step 4: Quantifying the DWT Coefficients
After obtaining the quantization table Q , we can quantify
Table 1: List of DWT based Blind and the DWT coefficients by the following method. We use one
Non-Blind Watermarking Algorithms block of 8x8 to embed one watermark bit. Hence we can
embed S/16 x S/16 bits in an image of size S x S. The em-
The main issues with these quantization based algorithms bedding begins by finding the nearest reconstruction point
are that it only tackles a subset of attacks. For example based on the quantization step size. Suppose the binary
Tsai’s algorithm is only robust against JPEG compression watermark bit is 1 and the mean intensity of the first block is
however Chen’s algorithm is doesn’t tackle geometric at- 158. Based on this value we compare with the quantization
tacks like rotation or image processing attacks like gamma table and find that the nearest reconstruction point for 1 is in
correction. Hence we propose a new algorithm which is ro- the range 161-200 hence the mean intensity of that block
bust against Cropping, Gamma Correction, JPEG Compres- should be uniformly scaled so that the new mean is centred
sion, JPEG2000, Resizing, Rotation (90, 180 degrees), Salt on 180 which is the median value for the range 160-200. The
and Pepper. selected block is shown in the table below.
VIII. PROPOSED SCHEME 0- 41- 81- 121- 161- 201- 241- 281- 321-
40 80 120 160 200 240 280 320 360
1 0 1 0 1 0 1 0 1
Discrete wavelet transform is a technique using which a
2D image can be transferred from spatial domain to fre- The scaling can be done as follows; suppose the original
quency domain. After one level DWT an image I is de- mean is M o and the final mean should be M f then the scal-
composed into four sub-bands LL, HL, LH, and HH. LL is
ing parameter can be calculated as M f / M o . We apply scal- termark is recovered. Using the secret seed S and the in-
ing to individual wavelet coefficients within the 8x8 block. verse permutation function f (.) we can recover the water-
Step 5: Inverse Quantization A comparative study is shown in Table 2 which shows the
We apply inverse wavelet transform to generate the wa- proposed algorithm and its comparison with the existing
termarked image. The secret keys used in this algorithm are work.
the seed S , the quantization table Q . These are termed as
verification keys. IX. RESULTS AND DISCUSSION
In order to prove the feasibility of our algorithm we con-
Step 6: Digital Signature and Time Stamping
ducted several experiments. The most classical example
The keys are digitally signed. Suppose DS = SignKey ( S , Q)
image of Lena was used for experiments. This image ( I ) is
Where Sign Key ( S , Q ) is the digital signature signed by the shown in the following figure. The watermark logo ( W ) that
owner’s private key. This digital signature is now time is used in the experiments is shown next. The size of the
stamped by a trusted third party like CA e.g. Verisign. The original image is 512x512 pixel grey scale image whereas
CA then computes T = T Key ( D s ) where TKey denotes the the size of the watermark logo is 32x32 pixel.
time stamp by the CA’s private key. The time stamped digital
signature is then stored together with the verification keys
and is used for proving the ownership of the content.
Proper- Tsai, 2000 Barni, Chen, Proposed
ties 2001 2003 Algorithm
Blindness Yes Yes Yes Yes
Quantization Constant Variable Variable Variable
Quantization Scalar Scalar Scalar Scalar
HVS No Yes Yes Implicit
Sub-bands Middle-low High LL HH, HL, LH (a) (b)
Counterfeit Not Dis- Not Dis- Timestamp Timestamp
Attack cussed cussed
Visual Yes No Yes Yes Fig. 2 (a) Image used for Watermarking
Pattern (b) Binary Image used as a Watermark
Robustness JPEG Compression, Compression, JPEG
against Compression Cropping, Cropping, JPEG 2000
Attacks Morphing Blurring, Cropping We first decomposed the image I into four sub-bands and
Sharpening, Salt and embedded the watermark W in all the bands except LL. We
Scaling Pepper then extracted the watermark without applying any attacks.
Correction, To our surprise the watermark extracted from diagonal
180 deg) sub-band (HH) was not similar to the original. This is at-
tributed to the rounding errors inherent with the D sub-band
Table 2: Comparison of various proposed methods because of its small values. The extracted watermarks in
based on DWT Table 3.
B. Extraction Algorithm
The extraction algorithm begins by verifying the digital sig-
nature and the time stamp. If the timestamp and the signature
cannot be verified the algorithm doesn’t proceed to the next HL LH HH
step. If the verification is positive the extraction algorithm
Table 3: Extracted watermarks without applying
begins and uses the quantization table Q and the secret seed
S to recover the watermark. The original image in not re-
quired during extraction. After embedding the watermark the output image is as
shown in Fig 3.
Step 1: Wavelet Transform
The original image I is decomposed by one level wavelet
transform to obtain LL, LH, HL and HH sub-bands.
Step 2: Watermark Recovery
For each sub-band except the LL sub-band, starting at the
top left corner we divide the sub-band into non-overlapping
blocks of 8x8 and calculate their mean intensity values of the
wavelet coefficients. These values are then compared with
the quantization table Q to generate the watermark bit. All
the watermark bits are thus generated and the permuted wa-
Fig 3: Watermarked Lena Image Finally we added some salt and pepper noise to see the
robustness against this kind of attack and it’s clearly visible
A. Results that the watermark is extracted.
After embedding the watermark we applied seven differ- Since we incorporate digital signature and time stamping
ent attacks and recovered watermarks which were visually technique the algorithm is also robust against counterfeit
recognizable. We attacked the watermarked image with the attacks.
following attacks Cropping, Gamma Correction, JPEG The inputs required for the extraction algorithms are not
Compression, JPEG2000, Resizing, Rotation (90, 180 de- many. We only require the quantization table and the secret
grees), Salt and Pepper. The results are shown in Table 4. seed for permuting the watermark. If the quantization table
is not available we can still manage if we have the quantiza-
Attacks HL LH HH tion step size and its range. If the range is not available we
JPEG 2000 can use identify the range from the watermarked image but
this results in reduced quality of the extracted watermark. If
the quantization table is not available the extracted water-
mark is again of a lower quality as shown in Table 4. Here
we constructed the quantization table based on the water-
marked image by gathering the range and using the quanti-
zation step size.
Our algorithm achieves good results by using smaller
Gamma block size. We use a block size of 8x8 to quantize the coef-
Correction ficients; however the scheme proposed by Meerwald 
used block size of 16x16 coefficients or 256 coefficients. At
Cropping the same time our algorithm offers more robustness to at-
tacks compared to the previously published schemes which
are listed in Table 2.
Resizing Using this algorithm grey scale image could also be em-
bedded as a watermark. We could just consider greyscale
image as a set of multiple binary images. The most signifi-
Rotation 90 cant bits planes of the grey scale image could be embedded in
this manner. A grey scale watermark has a greater prob-
ability of survival because it preserves the contextual rela-
Rotation 180 tionship.
Pepper In this paper we presented a quantization based
watermarking algorithm where we used an binary logo as a
Table 4: Attack Characterization watermark. We embedded the logo in HH, HL and LH
sub-bands of the first level wavelet decomposition. The
B. Discussion and Analysis proposed algorithm is shown to be robust against the
following attacks Cropping, Gamma Correction, JPEG
The proposed algorithm is shown to be robust against Compression, JPEG2000, Resizing, Rotation (90, 180
seven major attacks. degrees), Salt and Pepper.
Robustness against gamma correction attack is one of the
main features of this algorithm. Although distortions exist
the watermark is still visible.
The algorithm also showed robustness against geometric
attacks like cropping. The image was uniformly cropped  Cox, IJ, Miller, ML & Bloom, JA 2002, Digital Watermarking, Mor-
from all sides. However since the watermark was pseudo gan Kaufmann Publisher, San Francisco, CA, USA.
randomly permuted an attacker cannot remove the water-  Schneier, B., ‘Applied Cryptography’, WILEY, 2nd Edition.
 Craver, S., Menon, N., Yeo, B.L., Yeung, M., ‘Can visible watermarks
mark by removing a major portion of the image. The reason resolve rightful ownership?’ Proc. SPIE Storage and Retrieval for Still
is that the informative bit in the watermark i.e. the characters Image and Video Databases. Pp. 310-321, 1997
A, Z, P, N are uniformly spread all over the image and hence  Guan-Ming Su, ‘An Overview of Transparent and Robust Digital
cannot be removed totally. However even after cropping the Image Watermarking’. Available online at
image the watermark is still recovered.  Chae, J.-J., Mukherjee, D., Manjunath, B.S., ‘A robust embedded data
Image scaling operation is a severe attack because it re- from wavelet coefficients’, In Proceedings of the SPIE Electronic
sults in loss of data. We scaled the image from 512 x 512 to Imaging, Storage and Retrieval fro Image and Video Database, Vol.
256 x 256 and back. However even then the watermark is 3312, pp. 308-317, San Jose, CA, USA, Jan 1998
 Kundur D., and Hatzinakos, D., ‘A Robust Digital Image Water-
still visible in HL and LH sub-bands. marking Scheme using Wavelet-Based fusion,’ Proc. IEEE Int. Conf.
This algorithm also shows robustness against geometric On Image Processing, Santa Barbara, California, vol. 1, pp. 544-547,
attacks. We rotated the image by 90 degrees and 180 degrees October 1997.
and still the watermark is recovered completely.
 Chen, T.-Z., Horng, G., Wang, S.-H., ‘A Robust Wavelet Based Wa-
termarking Scheme using Quantization and Human Visual System
Model’, in Proceedings of the Pakistan Journal of Information and
Technology, vol. 2., no. 3., pp. 212-230, 2003
 Tsai, M. J., Yu, K.-Y., Chen, Y.-Z., ‘Joint Wavelet and spatial trans-
formation for digital watermarking’ in IEEE Transactions on Con-
sumer Electronics, Feb. 2000, vol. 46, no. 1, pp. 241-245
 Voyatzis, G., Pitas, I., ‘Digital Image Watermarking using Mixing
Systems’, in Computer Graphics, Elsevier, vol. 22, no. 4,pp. 405-416,
 Xiao, W., Ji, Z., Zhang, J., Wu, W., ‘A watermarking algorithm based
on chaotic encryption’, in Proceedings of IEEE Region 10 Conference
on Computers, Communications, Control and Power Engineering
TENCON, vol. 1, pp. 545-548, 28-31 Oct. 2002
 Kundur. D., Hatzinakos, D., ‘Digital Watermark using Multiresolution
Wavelet Decomposition’, in Proc. IEEE Int. Conf. On Acoustics,
Speech and Signal Processing, Seattle, Washington, vol. 5, pp.
2969-2972, May 1998.
 Xie, L., Boncelet, G., Acre, G.R., ‘Wavelet transform based water-
marking for digital images’, in Optics Express, vol. 3, no. 12, Dec
 Hsu, C.-T., Wu, J.-L., ‘Multiresolution Watermarking for Digital Im-
ages’, in IEEE Transactions on Circuits and Systems – II: Analog and
Digital Signal Processing, vol. 45, no. 8, pp. 1097-1101, August 1998
 Raval, M.S., Rege, P.P., ‘Discrete wavelet transform based multiple
watermarking scheme’, Conference on Convergent Technologies for
Asia-Pacific Region, TENCON 2003, vol. 3, pp. 935 – 938, 15-17 Oct.
 Kundur. D., Hatzinakos, D., ‘Towards Robust Logo Watermarking
using Multiresolution Image Fusion,’ IEEE Transactions on Multi-
media, vol. 6, no. 1, pp. 185-198, February 2004
 Tao, P & Eskicioglu, AM 2004, 'A Robust Multiple Watermarking
Scheme in the Discrete Wavelet Transform Domain', in Symposium on
Internet Multimedia Management Systems V, Philadelphia, PA.
 Lu, C. S., Huang, S.-K., Sze, C.-J., Liao, H.-Y., ‘A new watermarking
technique for multimedia protection,’ in Multimedia Image and Video
Processing, L. Guan, S.-Y. Kung, and J. Larsen, Eds. Boca Raton, FL:
CRC, 2001, pp. 507--530.
 Ganic, E., Eskicioglu, A. M., ‘Robust digital watermarking: Robust
DWT-SVD domain image watermarking: embedding data in all fre-
quencies’, Proceedings of the 2004 multimedia and security workshop
on Multimedia and Security, September 2004, pp. 166 – 174.
 Voloshynovskiy, S, Deguillaume, F & Pun, T 2000, 'Content Adaptive
Watermarking based on a Stochastic Multiresolution Image', in Tenth
European Signal Processing Conference (EUSIPCO'2000), Tampere,
Finland, September 5-8 2000.
 Tao, P., Eskicioglu, A.M., ‘A Robust Multiple Watermarking Scheme
in the Discrete Wavelet Transform Domain’, in Symposium on Internet
Multimedia Management Systems, Philadelphia, PA. October 25-28,
 Lee, C., Lee, H., ‘Geometric attack resistant watermarking in wavelet
transform domain,’ in Optics Express vol. 13, no. 4, pp. 1307-1321
 Lu, C-S., Liao, H-Y., M., Huang, S-K., Sze, C-J., ‘Cocktail Water-
marking on Images’, 3rd International Workshop on Information
Hiding, Dresden, Germany, Sep 29-Oct. 1, 1999