Robust Image Watermarking Theories and Techniques: A Review: Vol.12, February2014
Robust Image Watermarking Theories and Techniques: A Review: Vol.12, February2014
Robust Image Watermarking Theories and Techniques: A Review: Vol.12, February2014
ABSTRACT
Over the past several decades, digital information science has emerged to seek answers to the question: can any
technique ensure tamper-resistance and protect the copyright of digital contents by storing, transmitting and
processing information encoded in systems where digital content can easily be disseminated through communication
channels? Today it is understood that the answer is yes. This paper reviews the theoretical analysis and performance
investigation of representative watermarking systems in transform domains and geometric invariant regions. Digital
watermarking is a technology of embedding watermark with intellectual property rights into images, videos, audios,
and other multimedia data by a certain algorithm. The basic characteristics of digital watermark are imperceptibility,
capacity, robustness and false positive of watermarking algorithm and security of the hiding place. Moreover, it is
concluded that various attacks operators are used for the assessment of watermarking systems, which supplies an
automated and fair analysis of substantial watermarking methods for chosen application areas.
1. Introduction
Because of the fast and extensive growth of network detected from the watermarked product, for
technology, digital information can be distributed with protecting digital content copyright and ensuring
no quality loss, low cost and nearly instantaneous tamper-resistance, which is indiscernible and hard to
delivery. Protection of multimedia content has remove by unauthorized persons.
recently become an important issue because of the
consumers’ insufficient cognizance of the ownership Digital watermarking is seen as a partial solution to
of intellectual property. Thus, over the past several the problem of securing copyright ownership [80].
decades, digital information science has emerged to Essentially, watermarking is defined as the process
seek answers to the question: can researchers of embedding sideband data directly into the
ensure tamper-resistance and protect the copyright samples of a digital audio, image, or video signal.
of digital contents by storing, transmitting, and Sideband data is typically “extra” information that
processing information encoded in systems where must be transmitted along with a digital signal, such
digital content can easily be disseminated through as block headers or time synchronization markers. It
communication channels? Today it is understood that is important to realize that a watermark is not
the answer is yes, and many research groups around transmitted in addition to a digital signal, but rather
the world are working towards the highly ambitious as an integral part of the signal samples. The value
technological goal of protecting the ownership of of watermarking comes from the fact that regular
digital contents, which would dramatically protect sideband data may be lost or modified when the
inventions represented in digital form for being digital signal is converted between formats, but the
vulnerable to illegal possession, duplication and samples of the digital signal are (typically)
dissemination [83]. Digital watermarking [16] is the unchanged[72].
process of embedding or hiding digital information
called watermark into a multimedia product, and then To clarify this concept further, it is useful to
the embedded data can later be extracted or consider an analogy between digital watermarks
122 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
and paper watermarks. Watermarks have extracted watermark during the detection process. In
traditionally been used as a form of authentication other words, the watermark still exists in the
for legal documents and paper currency. A watermarked image, but its positions have been
watermark is embedded within the fibers of paper changed. Therefore, while traditional watermarking
when it is first constructed, and it is essentially systems require the creation of a framework of the
invisible unless held up to a light or viewed at a resilience to watermarked data geometrical
particular angle. More importantly, a watermark is modifications, creation and enforcement of
very difficult to remove without destroying the synchronization errors correction of such frameworks
paper itself, and it is not transferred if the paper is is now possible. Besides facilitating more efficient
photocopied. The goals of digital watermarking are copyrighted protection and robustness against
similar; in the next section, it will be shown that desynchronization, adaptation of geometrically
digital watermarks require similar properties. invariant image features can potentially offer a
greater robust capacity to detect watermarks without
Before the concept of watermarking can be explored synchronization errors, especially when applied to
further, three important definitions must first be survive local distortions such as random bending
established. A host signal is a raw digital audio, attacks. Development of such a framework is an
image, or video signal that will be used to contain a essential starting point for organizations that wish to
watermark. A watermark itself is loosely defined as a improve or replace currently existing watermarking
set of data, usually in binary form, that will be stored algorithm-based pixel frequency or other transform
or transmitted through a host signal. The watermark coefficients for watermark embedding, and develop a
may be as small as a single bit, or as large as the set of means to establish and maintain feature-based
number of samples in the host signal itself. It may be watermarking of geometric distortions correction.
a copyright notice, a secret message, or any other
information. Watermarking is the process of In the first section, the properties of the general
embedding the watermark within the host signal. watermarking frameworks that are exploited in the
Finally, a key may be necessary to embed a process of encoding and detecting watermarking
watermark into a host signal, and it may be needed are shortly reviewed. A survey of the key digital
to extract the watermark data afterwards[16]. image watermarking algorithms and techniques is
presented subsequently. The characteristics of
Up to now, two traditionally-used strategies, spatial- watermarking systems are described for evaluating
domain [68] and transform domain [28][80] the performance of watermarking systems. There
techniques have been developed for digital image are five important issues that are usually considered
watermarking. The former category is designed to in the most practical application; they are
insert directly a watermark into the original image by highlighted in the following subsections. In addition,
a factor, which would lead to fair-quality watermarked digital watermarking is described as an efficient
images. The latter approach, for taking advantage of method for the protection of ownership rights of
perceptual properties, is devised to embed a digital audio, image, video and other data types. It
watermark into the frequency-domain of the original can be applied to different applications including
images. These types of watermarking schemes have digital signatures, fingerprinting, broadcast and
good performances of robustness in comparison to publication monitoring, authentication, copy control,
the most common signal processing manipulations and secret communication. Watermarking attacks
such as JPEG compression, filtering, and addition of can be classified into two broad categories:
noise [16][14][40][49][59]. Signal processing destruction attacks: including image compression,
operators are applied to watermarked images for image cropping, spatial filtering, among others; and
removing the watermark or decreasing its energy so synchronization attacks: including image rotation,
that the extracted watermark is unrecognizable or image shifting and pixelhine deletion. The chapter
insufficient as the validate evidence. Unfortunately, lists and describes some of these conventional
the ineffectiveness of existing traditional attacks in the following sections.
watermarking algorithms is described by the
robustness against unintentional or malicious For constructing geometric invariant watermarking,
geometric attacks [37]. Geometric attacks induce four mainstream schemes are introduced by literature
synchronization errors between the original and the reviews on watermarking algorithms robust to the
JournalofAppliedResearchandTechnology 123
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
geometrical distortions. Most of these efforts confine purchaser of the multimedia data over public
to theoretically analyzing and quantifying the effect of channels. Using variations of symmetrical and
the global and local affine transform to the asymmetrical styles or forms of encrypting data,
performance of the watermarking algorithms. the permutated original multimedia contents are
non-recognizable in appearance, unsystematic,
2. Watermarking backgrounds and disorderly [82]. Although encryption algorithms
can be applied to avoid illegal access to digital
With the rapid proliferation of globally-distributed contents, it appears that encryption by itself is not
computer networks technologies and popularity of sufficient enough to prevent an unauthorized pirate
multimedia systems, fashionable and economical from illegally replicating multimedia content and
digital recording and storage devices have made it protect multimedia data all along its lifetime. Once
possible to construct the platform, where it became multimedia content has been decrypted into its
considerably facilitated to not only acquire, original style and the protection of information is
represent, replicate, distribute and transmit invalidated for further manipulations because there
multimedia contents in digital formats without is no degradation of quality in subsequent works
degradation of quality, but also manipulate them. and no verification differences between one copy
General-purpose computers and graphics editing and any other derivative copy. Therefore,
programs provide ultimate playgrounds for an unauthorized replicating copy and transmission of
amateur who does not have any ripe experience multimedia data cannot be obstructed [83].
and professional skills. He conveniently processes
an image or tamper specific objects without Steganography [54][77] represents a technique
remaining any appreciable traces, for introducing that is used to convey communicating secret data
the best in your digital images, reconstructing them by writing hidden messages into an appropriate
into anything you can imagine, and demonstrating multimedia carrier, e.g., audio, image and video.
them in extraordinary ways. In the multimedia The existence of the message is suspected except
publishing industries, researchers, scientists, and by the sender and intended receiver. Unlike
practicing engineers attach close importance to cryptography techniques, the goal of steganography
pervasive advancements because the unauthorized is to conceal the very existence of the hidden
manipulation and the unrestricted reproduction of messages, together with avoiding arouse suspicion
original digital multimedia can easily be and not attracting attention to themselves. However,
disseminated through communication channels steganography conventionally involves associating
such as the Internet. Consequently, there is an secret point-to-point transmission and
urgent demand for effective techniques to ensure communication. Thus, steganography approaches
tamper-resistance and prevent pirates from causing are typically not resistant to transformation of the
damage to the owners of digital content. Because of carriers, or hold only restricted robustness.
possible copyright issues, the copyright and
intellectual property of digital multimedia data should The definition of digital watermarking [16-17]
be protected from illegal possession, duplication and emerges while trying to overcome the limitations of
dissemination. Three complementary techniques are encryption and steganography in enforcement and
being introduced: encryption, steganography and protection of intellectual property rights. Compared
watermarking [16][67][24]. to the idea of encryption, the watermark
information is inserted into its original form and
Encryption [44] is a conspicuous and secure does not hinder users from listening to, viewing,
technique to converse data into a scrambled code watching, or manipulating the content. And unlike
that can be distributed and deciphered through a steganography, digital watermarking technologies
private or public network. Generally speaking, in are to establish the identity of information to avoid
both research and application fields, encryption the unauthorized embezzlement. Generally,
and cryptographic algorithms serve copyright additional information is embedded directly into the
owners as an approach to protect the secure original multimedia or host signal which is useful
transmission of confidential multimedia data and valuable, and the message itself is not
between a distributor or publisher and the unnecessary to be secret.
124 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
JournalofAppliedResearchandTechnology 125
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
126 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
JournalofAppliedResearchandTechnology 127
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
The above represent a few example applications Removal attacks intent to accomplished removing
where digital watermarks could potentially be of watermarks from the host image. This classification
use. In addition, there are many other applications includes lossy compression, denoising, demodulation,
for rights management and protection like tracking quantization, averaging and collusion attacks.
use of content, binding content to specific players,
automatic billing for viewing content, broadcast 6.1.1 Denoising and lossy compression attacks
monitoring, among others. From the variety of
potential applications exemplified above, it is clear This category of attacks is relatively broad and
that a digital watermarking technique needs to contains common image processing operators such
satisfy a number of requirements. Since the as lossy compression, image denoising and
specific requirements vary with the application, quantization. Image denoising, also understood as
watermarking techniques need to be designed filtering, is mainly related to maximum likelihood, a
within the context of the entire system in which minimax criterion or a minimum mean square error, a
they are to be employed. Each application imposes maximum a posteriori probability. The resulting
different requirements and would require different filtering operator is decided by the selected criteria,
types of invisible or visible watermarking schemes and also by the priors on the cover image and the
or a combination thereof. watermark. Compression is a popular scheme for
attacking watermarked images or audio. Two
6. Attack operators common compression schemes: lossy compression,
such as VQ compression, and JPEG compression for
The assessment of watermarking systems supplies image processing have lately been considered to
an automated and fair analysis of substantial have approximately the same impact on noise
watermarking methods for chosen application removal as denoising. For the attackers to remove
areas. At present, numerous investigators and the hidden watermarks, they may compress the
researchers utilize their own designed assessment watermarked images with some other VQ codebooks
systems, which does not require the capability of and decode the VQ indices to get the reconstruction.
comparison each other. Therefore, The The VQ compression schemes are effective for
assessment procedure can be very complicated, attacking some of the existing algorithms. Both lossy
and the current research is on assessment compression and denoising can importantly diminish
methods with unique attacks for images (for the capacity of the watermarking channel
example, attainable tools Optimark [51], StirMark establishing the output of various substitutable
[65], Checkmark [10] or for particular applications channels to zero for each bit of watermark.
128 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
JournalofAppliedResearchandTechnology 129
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
130 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
limited in the N u M matrix with real entries. It is and labels the resulting subimages. LL (the
noted that the SVD applies more generally to approximation) which is the coarse overall shape,
complex-valued rectangular matrices, while we covers the low-frequency components that contain
restrict our discussion to real-valued, square most of the energy in the image and LH (horizontal
details), HL (vertical details) and HH (diagonal
matrices. The singular value decomposition of A details) which represent higher-frequency detailed
is represented by, information have the finer scale wavelet
coefficients according to the filters used to
A UȈVT (3) generate the sub-image. The wavelet components
are then used to obtain the next coarse overall
Nu N shape by further iterating LL in this process and we
Where U and V R are the unitary matrix, and get the details (LL1, LH1, HL1, and HH1). The size
ȈR Nu N of the input data reduces gradually. The size at
is a diagonal matrix and the superscript
each succeeding octave is one-fourth the size of
T denotes matrix transposition. The diagonal
the previous one. This process is repeated several
ı times until the desired final scale is obtained as
elements of Ȉ , denoted by i are called the
singular values of A and these are assumed to be shown in Figure 3. The Figure shows the sketch of
the decomposition of an image in three resolution
ıi t ıi 1 levels through DWT. In the analysis of
arranged in decreasing order . The
Ui reconstruction, the inverse DWT is performed by a
columns of U denoted by ̓s are called the left similar structure with the corresponding synthesis
singular vectors while the columns of V denoted by subimages, consisting of an inverse the
Vi approximation coefficients and the detail
̓s are called the right singular vectors of A. It coefficients. This feature makes the watermark
ıi Vi Ui more robust in comparison with spatial methods
is easy to see that , and satisfy: against various distortions. The applications of
DWT are efficient due to two important reasons.
AVi ıi U i One: it has the irregular distribution of the inverse
(4)
transform value in a pyramidal wavelet domain.
Two: it has the low linear complexity which
U iT A ı i VIT (5) requires lower computational cost O(n) when
compared with the computational cost consumed
In [42], the watermarking scheme is fundamentally by Fourier and cosine transforms O(n log(n)).
flawed algorithm, this is because one attacker can
always claim that this watermark was the embedded
one and he can claim ownership of the
watermarked image, using the singular vectors of
any fake watermark in the detection stage. For
correcting fault scheme, a SVD-based watermarking
algorithm, which explores the optimal scaling factors
of watermark embedding, is presented.
JournalofAppliedResearchandTechnology 131
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
Watermarking algorithms robust to the geometrical Recently, one of the most straightforward solutions
distortions have been the focus of research [81]. to cope with desynchronization attacks is composed
Several approaches to this problem include of inserting templates along with the watermark or
exhaustive search, synchronization pattern/template, embedding a periodic watermark pattern into the
invariant domain and implicit synchronization using image. These additional templates [52][64] are used
image features are widely used. Most of the as artificially embedded references for purposes of
proposed geometrical transform invariant algorithms resynchronization. There is insensitive information
are actually only RST invariant. Also the systematic carried in them. The template-based approach
analysis of the watermarking algorithm performance performs the watermark retrieval process by
under geometrical distortion has begun to draw asserting the presence of a watermark and
great attention. Most of these efforts confine to estimating and compensating the severe geometric
theoretically analyzing and quantifying the effect of transformation of the watermarked image for
the global affine transform to the performance of the accomplishing resynchronization patterns [41].
watermarking algorithms. Local distortions are more Therefore, the registration pattern is identified and
and more regarded as a necessary test scenario in offered resilience to geometric attacks [7].
benchmarking software. However, it is difficult to
theoretically analyze its effect on watermark The template-based algorithms introduced in this
detection due to its complexity. section all are relevant to adaptive determination in
the strength of the templates for constraining the
8.1 Exhaustive Watermark Search search space. The contraction of searching space
results in being rather susceptible to the credibility of
Exhaustive search technique [31] is one of system. Meanwhile, the template is restricted to the
commonly used candidate approaches to coping number of synchronization points, which tends to be
with the watermark desynchronization problem. This the counterpart of unaccepted false positive
approach performs the process of watermark probabilities in the parameter estimation of the
detection over a training sequence containing each applied affine transformation. Also the process of the
pilot geometrical inverse deformation. The watermark successful watermark detection relies on the precise
is recovered by searching each hypothetical detection of the template because the detection error
distortion parameter. Obviously, it seems to be tolerance is unacceptable for the inaccuracy of the
difficult to efficiently evaluate deformation detected position. It causes the unwarranted
parameters for the absence of the original image synchronization for shrinking the search space.
132 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
Although an amount of progress for utilizing scheme based on the Fourier–Mellin transform [39].
template-based watermarking techniques, these A rotation, scale and translation invariant domain is
ignore the perceptual similarity between the original obtained using the log-polar mapping (LPM) and the
and watermarked image. The insertion of the Fourier transform invariance properties to
watermark and the template should take the translations. In the resulting log-polar map, rotations
embedding position and strength into account and scaling come down to translations. In practice,
carefully. And besides, the algorithms compromise this solution can be implemented for simple affine
the data payload of the watermark for keeping the transformations, but it is inapplicable as soon as the
restricted fidelity of the original image after the image undergoes local geometrical deformations.
embedding process, since the embedded template Moreover, problems of approximation due to the
tries to decrease the number of embedded discrete nature of the images, plus the reduction of
information that consist of the template to guarantee the embedding space make the watermark weakly
synchronization and avoid potential visible artifacts. resistant to low-pass filtering and lossy
compression. Alghoniemy and Tewfik [2] present
In addition, security aspect has been paid little another approach where the watermarking space is
regard, because these algorithms are particularly defined as a canonical, normalized space based on
vulnerable to template removal attacks [23][25][7]. the geometric image moments. These moments are
If these templates are provided with characteristic used to transform the image into a form that is
features of independence from the host image, independent to rotation, scale, and
then the specific characteristics can be exploited to horizontal/vertical reflection. The watermark is
destruct the synchronization pattern. In another embedded in this space, and then the inverse
word, the templates are used to be inserted as transformation is applied to obtain the final
pseudorandom noise patterns and some specific watermarked image. During detection, the moments
operations, for example filtering operators, can are calculated again and used to estimate the
filter potentially out the local maxima. Then, the normalization parameters. Once the image is
applied templates are eliminated easily by the normalized, the watermark can be detected. Rather
malicious attack and the applied various geometric than embedding the watermark in an invariant
transformations are unable to be recognized during subspace, Solachidis and Pitas propose creating a
the template detection process. In brief, the major self-similar watermark, and to embed it then into the
limitation is that this kind of approach has a severe DFT domain[63]. Thus, their method is robust to
influence on the resistance of affine translation, since it does not affect the DFT
transformations confronting with threats and risks magnitude. Since the watermark is made up of
of the template attack. identical sectors, the detection is possible even after
a degree rotation. The self-similar properties of the
Self-synchronizing watermarks [29][21] are watermark also allows for the reduction of the
susceptible to removal or estimation attacks in number of different frequency sampling steps where
much the same way as template-based methods, the detection should be performed when the image
because an attacker can use knowledge about the has been cropped and scaled.
watermark’s periodic tiling to remove it. For
example, attackers can easily remove peaks in the 8.4 Feature Watermarking
autocorrelation function or the frequency domain
by filtering in the spatial or the frequency domain, Feature-based watermarking has raised a number of
thus rendering the watermark vulnerable to available algorithms over the last few years. For
subsequent geometric attacks. solving watermark synchronization problems, feature-
region detection is the preferred strategy to resist
8.3 Invariant Domain against local geometric distortions. Generally
speaking, content-based synchronization
Another solution consists in embedding the watermarking schemes follow the same basic
watermark in a geometrical invariant subspace. In process: detected feature points are localized at the
[12], it suggested using histogram specification to local maxima while non-maxima suppression that
hide a watermark invariant to geometrical distortions. eliminates pixels that are not local maxima, and the
In [60] and [38], they proposed a watermarking final set of features is determined by analysis of
JournalofAppliedResearchandTechnology 133
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
threshold. Afterwards, extracted feature points are patches as the inserting modules. The rectangular
applied to identify regions for watermark insertion in watermark is transformed to be a polar-mapped
the host image. At the receiver side, the feature watermark, and inversely polar-mapped to determine
points are detectible without synchronization error. inserting modules before watermark insertion. The
The observation of feature-based synchronization watermark correspondence detection is achieved by
has resulted in various algorithms known as region- the circular convolution. The invariant watermark of
based watermarking [22][35][36][61][62]][73][74][75]. rotation is obtained by the translation characteristic of
The feature points-based approach is a technique the polar-mapped circular pads. In [33], Harris̽
using localized watermarking algorithms. It discovers Laplace approach is applied to extract the scale
the watermark using stable feature points of images, invariant feature points relied on the scale selection
where the watermark is independently inserted into theory using Harris corner points. The watermark is
the corresponding each local region [29]. Hence the inserted after affine the normalization according to
feature-based process can be invariant to local the local characteristic scale at each feature point.
geometrical deformations so that it is an encouraging The characteristic scale is defined by the scale,
approach to solve the robustness against geometrical where the normalized scale-space representation of
deformations in the watermarking scheme with blind an image achieves a maximum value, and the
detection. In [7], it extracted feature points of the host characteristic orientation relied on the angle of the
image using Harris detector and produced a dominant axis of an image.
Delaunay tessellation on the group of stable feature
points. Then, the watermark was inserted in each In [22], a robust watermarking approach was
triangle of the tessellation. But the Harris feature proposed combining Tchebichef moments and the
points were not invariant to scaling [46] so that the local circular regions (LCRs). LCRs are shaped by
approach is not resistant to the attack of scaling. Harris-Laplace detector, and Tchebichef moments
are applied to acquire the global characteristics of
In [69], it produced a feature extraction process the LCRs. In [62], the affine invariant point detector
named Mexican Hat wavelet scale interaction. Image [46] was discussed to detect feature points. For a
normalization was individually exploited to chosen feature point, an elliptical feature region
unoverlapping image circles with the firm radius and (viz., affine covariant region) is used and formed
the center at the obtained feature points. In each for inserting in spatial domain. Before embedding,
image circle, two 32 32 blocks were selected for the watermark, it is geometrically transformed into
embedding watermark. In spite of the fact that the an elliptical pattern according to the shape of the
system displays experimentally the robustness to the region. This method has a modest robustness and
majority of attacks, it is attackable to scaling provides a potential idea for resistance against
transforms because these transform can cause the complicated geometric distortions.
alterations of content for two blocks in the image
circles with a firm radius. In [35], it presented the 9. Conclusion
Harris feature points from the image with the
normalized scale and inserted the watermark within With the rapid proliferation of globally-distributed
each disk region in the center at the localized stalest computer network technologies and popularity of
Harris points and firm radius. The presented method multimedia systems, fashionable and economical
can be resistant to global geometrical digital recording and storage devices have made it
transformations, including scaling, rotation and possible to construct the platform, which has
moderate translation. Nevertheless, the presented considerably make it easy to not only acquire,
scheme cannot be robust against cropping because represent, replicate, distribute, and transmit
the normalization of scale may be sensitive to multimedia contents in digital formats without
cropping in nature. Recently, many researchers have degradation of quality, but also to manipulate
turned interest orientation to the resynchronization of them. However, the transmission of information
watermark using the scale invariant feature detectors over various networks is often unsafe. This lack of
relied on the scale-space theory in the pattern security can be critical depending on the nature of
recognition fields, such as Harris-Laplace, scale the transmitted media. This issue gave rise to
invariant feature transform (SIFT) [33]. In [33], the digital watermarking, a research field which deals
SIFT feature is exploited to produce the circular with the process of embedding information into
134 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
digital data in an inconspicuous manner. This in [7].Bas P., Chassery J.M., and Macq B., 2002.
turn enables information security for various Geometrically invariant watermarking using feature point.
multimedia applications including copyright IEEE Transactions on Image Processing, 11(9):1014–1028.
protection. In this paper, different watermarking [8].Cayre F., Fontaine C. and Furon T., 2005.
techniques have been reviewed and analyzed. Watermarking security, part I: theory, In: Security,
This is based on image processing in spatial and Steganography and Watermarking of Multimedia Contents
transform domain. Different techniques using VII, Proceedings of SPIE. 5681.
singular value decomposition and discrete wavelet
transform in transform domain have been [9].Cayre F., Fontaine C. and Furon T., 2005.
reviewed. Additionally, the analysis of these Watermarking security, part II: practice, In: Security,
Steganography and Watermarking of Multimedia
techniques has been represented in the form of Contents VII, Proceedings of SPIE. 5681.
tables considering different factors of watermarking
like embedding imperceptibility, capacity, security, [10] Checkmark Benchmarking, http://watermarking.unige.
robustness and false positive. It can be concluded ch/Checkmark/, 2006
that various attacks techniques are used to the
assessment of watermarking systems, which [11] Christian K., Jana D., Andreas L., 2006. Transparency
benchmarking on audio watermarks and steganography,
supplies an automated and fair analysis of SPIE conference, at the Security, Steganography, and
substantial watermarking methods for chosen Watermarking of Multimedia Contents VIII, IS&T/SPIE
application areas. Furthermore, watermarking Symposium on Electronic Imaging.
algorithms robust against the geometrical distortions
have been the focus of research. Several [12] Coltuc D. and Bolon P., 1999. Robust watermarking
approaches to this problem, including exhaustive by histogram specification. Proc. IEEE Int. Conf. Image
search, synchronization pattern/template, invariant Processing. 2:236–239.
domain and implicit synchronization using image [13] Cox I.J. and Linnartz J.P.M.G.,1998. Some general
features, are widely used. methods for tampering with watermarks. IEEE J.
Selected Areas Communi., 16(4)587–593.
References
[14] Cox I. J., Miller M. L., and Bloom J. A. 2001. Digital
Watermarking. San Francisco, CA: Morgan Kaufman.
[1].Alghoniemy M. and Tewfik A. H., 2000. Geometric
distortion correction in image watermarking. Proc. SPIE
[15] Cox I., Miller M., Bloom J., Fridrich J., and Kalker T.,
Security and Watermarking of Multimedia Contents II ,
Digital Watermarking and Steganography, 2007.
3971:82- 89.
Multimedia Information and Systems. 142–143.
[2].Alghoniemy M. and Tewfik A.H., 2000. Geometric [16].Cox I.J. and Miller M.L. 1997. A review of
distortion correction through image normalization. Proc. watermarking and the importance of perceptual
IEEE Int. Conf. Multimedia and Expo. 3:1291–1294. modeling. Proc. SPIE Electronic Imaging ‘97, Storage
and Retrieval for Image and Video Databases.
[3].Alghoniemy M. and Tewfik A.H., 2006. Progressive
quantized projection approach to data hiding. IEEE [17].Cox, I.J. 1996. Secure spread spectrum watermarking
Transactions on Image Processing, 15(2):459-472. for images, audio and video. International Conference on
Image Processing, 234–246.
[4].Alvarez R.M. and Perez G.F. 2002. Analysis of pilot-
based synchronization algorithms for watermarking of [18].Cox, I.J., Kilian J., Leighton, F.T., and Shamoon,
still images. Signal Processing: Image Communication. T.,1997. Secure spread spectrum watermarking for
17(8): 611 – 633. multimedia. IEEE Transactions on Image Processing,
6(12), 1673–1678.
[5].Andreas L., and Jana D., 2006. Profiles for
Evaluation - the Usage of Audio WET. SPIE conference, [19]Craver S., Memon N., Yeo B.L., Yeung M.M., 1997.
at the Security, Steganography, and Watermarking of Can invisible watermark resolve rightful ownerships?
Multimedia Contents VIII, IS&T/SPIE Symposium on Fifth Conference on Storage and Retrieval for Image and
Electronic Imaging. Video Database. 3022:310–321.
[6].Barni M., 2005. Effectiveness of exhaustive search and [20]Deguillaume F., Csurka G., Pun T., 2000.
template matching against watermark desynchronization. Countermeasures for unintentional and intentional video
IEEE Signal Process. Lett. 12(2):158–161. watermarking attacks. IS&T/SPIE Electronic Imaging.
JournalofAppliedResearchandTechnology 135
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
[21].Delannay D. and Macq B., 2000. Generalized 2D [34] Bao P, Ma X., 2005. Image adaptive watermarking
Cyclic Patterns for Secret Watermark Generation. Proc. using wavelet domain singular value decomposition.
IEEE Int’l Conf. Image Processing, 2:72-79. IEEE Trans. Circuits Syst. Video Technol. 15(1): 96-102
[22].Deng C., Gao X., Li X. and Tao D., 2009. A local [35].Li L.D., and Guo B.L., 2007. Localized image
Tchebichef moments-based robust image watermarking. watermarking in spatial domain resistant to geometric
Signal Process., 89( 8):1531–1539. attacks, Int. J. Electron. Commun. (AEU), 63:123–131.
[23].Dong P. and Galatsanos N., 2002. Affine [36].Lichtenauer J., Setyawan I., Kalker T., and
transformation resistant watermarking based on image Lagendijk R., 2003. Exhaustive geometrical search and
normalization. Proceedings of International Conference the false positive watermark detection probability. Proc.
on Image Processing, 3:489 – 492. SPIE Security and Watermarking of Multimedia
Contents. 5(2):203–214.
[24].ElbaúÕ, E. (2012). Robust MPEG Watermarking in
DWT Four Bands. Journal of Applied Research and [37] Licks V. and Jordan R., 2005. Geometric attacks on
Technology, 10(2). image watermarking systems. IEEE Multimedia
Magazine,12(3):68–78.
[25] Herrigel A., Voloshynovskiy S., and Rytsar Y., 2001.
The watermark template attack. Proceedings of SPIE: [38] Lin C. Y., Wu M., Bloom J.A., Cox I.J., Miller M.L.,
Security and Watermarking of Multimedia Contents III, and Lui Y. M., 2000. Rotation, scale, and translation-
394–405. resilient public watermarking for images. Proc. SPIE—
Security and Watermarking of Multimedia Contents II,
[26].Holliman M., Memon N., Yeung M., 1999. 3971:90–98.
Watermark estimation through local pixel correlation.
IS&T/SPIE Electronic Imaging’99, Session: Security and [39].Lin F. and Brandt R.D., 1993. Towards absolute
Watermarking of Multimedia Contents,134–146. invariants of images under translation, rotation and
dilatation. Pattern Recognit. Lett., 14(5):369–379.
[27].Kalker T., 2001. Considerations on watermarking
security. Proceedings of the IEEE Multimedia Signal [40] Lu C. S., Huang S. K., Sze C. J. and Mark Liao H.
Processing MMSP01 workshop, 201–206. Y. 2000. Cocktail watermarking for digital image
protection. IEEE Trans. Multimedia, 2(4):209–224.
[28].Kim T. Y., Choi H., Lee K., and T. Kim 2004. An
asymmetric watermarking system with many embedding [41] Lu W., Lu H.T., Chung F.L., 2006. Feature based
watermarks corresponding to one detection watermark. watermarking using watermark template match. Appl.
IEEE Signal Processing Tellers, 2:375-377. Math Comput. 177(1):377–86.
[29].Kutter M., 1999. Watermarking Resisting to [42].Luis P.F., Pedro C. and Fernando P., 2005.
Translation, Rotation, and Scaling. Proc. SPIE Information- Theoretic Analysis of Security in Side-
Multimedia Systems and Applications. 3528:423-431. Informed Data Hiding, Information Hiding, 131–145.
[30].Kutter M., Bhattacharjee S.K., Ebrahimi T., [43].Macq, B., Dittmann, J., Delp, E.J., 2004.
1999.Toward second generation watermarking schemes. Benchmarking of Image Watermarking Algorithms for
Proceedings of the IEEE International Conference Image Digital Rights Management, Proceedings of the IEEE,
Processing, 1:320–323. Special Issue on: Enabling Security Technology for
Digital Rights Management, 92(6): 971–984.
[31].Kutter M., Jordan F., and Bossen F., 1998. Digital
watermarking of color images using amplitude
[44].Matthew R. 1989. One the derivation of a chaotic
modulation. J. Electron. Imag. 7(2):326–332.
encryption algorithm. Cryptologia 8 (1): 29–42.
[32].Langelaar G.C., Lagendijk R.L., Biemond J., 1998.
Removing spatial spread spectrum watermarks by non- [45].Merhav,N. 2005. An information-theoretic view
linear filtering. Proceedings of the European Signal ofwatermark embedding-detection and geometric
Processing Conference. attacks. Proceedings of WaCha 05.
[33].Lee H.Y., Kim H., Lee H.K., 2006. Robust image [46] Mikolajczyk K., Schmid C., 2002. An affine invariant
watermarking using local invariant features, Opt. Eng. interest point detector. Proceedings of European
45 (3):1-11 Conference on Computer Vision, 2350:128–142.
136 Vol.12,February2014
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTaoetal./122Ͳ138
[57].Petitcolas, F., Anderson R.J., Kuhn M.G., 1999. [71].Voloshynovskiy S., 2001.Attacks on Digital
Information Hiding: A survey. Proceedings of the IEEE Watermarks: Classification, Estimation-Based Attacks, and
(special issue) 87(7):1062–1078. Benchmarks. IEEE Communications Magazine.
39(8):118–126.
[58] Pitas I., 1996. A method for signature casting on
digital images. Proceedings of the IEEE International [72] Vukmiroviü, S., Erdeljan, A., Imre, L., & ýapko, D.
Conference on Image, 215–218. (2012). Optimal Workflow Scheduling in Critical
Infrastructure Systems with Neural Networks. Journal of
[59] Prado-Molina, J., Peralta-Higuera, A., Palacio-Prieto, J. Applied Research and Technology, 10(2).
L., & Sandoval, R. (2010). Airborne high-resolution digital
JournalofAppliedResearchandTechnology 137
RobustImageWatermarkingTheoriesandTechniques:AReview,HaiTao etal./122Ͳ138
[74] Yuan W., Ling H., Lu Z., Yu Y., 2006. Image content-
based watermarking resistant against geometrical
distortions. Proceedings of the Eighth International
Conference on Signal Processing, 2632–2635.
[75] Zheng D., Liu Y., and Zhao J., 2007. A survey of
RST invariant image watermarking algorithms. ACM
Comput. Surv., 39(2):1–91.
138 Vol.12,February2014