Seminar Report On Image Watermarking

Seminar Report on
Multipurpose Image Watermarking

------------------------------------------------------------------------------------------
--------------------------------------------
Chapter - 1
INTRODUCTION
Digital multimedia work including music, photographs, and movies, have proliferated
during the last decade with the advent of the World Wide Web, e-commerce, and broadband
connectivity to the home. This technical and economic revolution has ushered in a brave new
world for content creators, publishers, distributors, and consumers. The advantages of digital
media are many, including efficient storage and manipulation, fast and inexpensive
distribution, and the ability to customize the media to each unique user. However, these
advantages are tempered by serious concerns regarding illegal copying, the viability of new
e-commerce business models, and the uncertainty surrounding newly enacted copyright laws.
Digital rights management concerns itself with all these issues. While rapid advances
in technology have created the current situation (where illegal digital copies can be created at
virtually no cost and transmitted almost instantly to anyone in the world), technology alone is
unlikely to provide a comprehensive solution. Clearly, business models, licensing
agreements, and new case law will have very significant roles to play.
The ease of copying digital information without any loss of quality violates the
conservation of mass property of traditional media, which inhibited wide global distribution
in the past. On the Internet today it is possible to duplicate digital information a million-fold
and distribute it over the entire world in seconds. These issues worry creators of intellectual
property to the point that they do not even consider to publish on the Internet. Copy
protection, copyright protection, and content authentication have, therefore, been the three
most important issues in the digital world. Conventional cryptographic systems permit only
valid key holders access to encrypted data, but once such data is decrypted, there is no way to
track its reproduction or retransmission [1].
To solve the problem of publishing digital images, researchers have come up with
digital image watermarking [1]. This method allows the owner of an original image to add an
invisible watermark to the digital image before publishing it. The watermark serves to claim
copyright on the image. The owner protects the watermark with a cryptographic secret key,
preventing anybody not possessing the secret key from reading or even detecting the
watermark. The watermark is also supposed to be robust against image tampering. Therefore
anybody who wants to distribute the image further will also distribute the watermark with it,
--------------------------------------------------------------------------------------
------------------------------------------

1
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
violating the copyright on the image. If the copyright holder can detect the fraud, he can
prove ownership by showing that the image contains his proper private watermark.
This scheme works well in a nice and tidy" clean-room research environment. But in
the real world" people play by different rules. The systems are not used as intended in the
research environment and hidden back doors are exploited. This report shows the problems
that current digital image watermarking schemes face.
1.1. Real World" Concerns
Unfortunately the real world" differs significantly from research labs. Computer
programmers or researchers assume that the end users will behave and think as logically" as
they do. For example image watermarking is used by a wide variety of users and should still
be equally applicable and efficient for everybody. Unfortunately this is not the case for
today's systems.
1.2. User Interface Concerns
In the real world" the users have only a limited understanding of the underlying
mechanisms of image watermarking. They do not want to spend hours of training to use one
function of their image processing software. Watermarking should be like a black box"
where the user enters his original image and by some magic the box outputs the watermarked
image. No specific user understanding should be necessary.
Usually artists are the designers of images. The observation that artists do not like to
degrade their work deliberately by inserting a watermark leads us to believe that they will
either insert a weak watermark or not insert anything at all. Security is not always a strong
enough argument to convince artists to lower their image quality.
Nowadays the user interface (UI) becomes very important: it must present the user
with a clear model of the effects of watermarking and protect him from misuse. It needs to
iron out users misconceptions about the watermarking technique. This observation is equally
true for other security software. Whitten showed in [26] that the security of the system should
not rely on user understanding but be an implicit property of the software. Therefore the UI
becomes a crucial point in any security system.
1.3. Legal Issues
--------------------------------------------------------------------------------------
------------------------------------------

2
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
The real world challenges in image copyrighting are as follows. The scenario is the
following: in a country that does not adhere to the Berne convention [27] on copyright
protection, a malicious person sets up his web-server distributing copyrighted images, music,
etc. There is no way to prevent this person from his illegal" distribution, as the country does
not provide the legal basis for prosecution. All forms of intellectual property share this
common problem. The past has shown that such scenarios are not far fetched. In fact recent
incidents prove that this attack is quite common. The situation gets worse as the Internet
expands its range with high-speed connections to countries that are traditionally" known for
copyright infringement. If these countries do not change their laws, this problem cannot be
solved trivially by technical means.
1.4. Web Spider Issues
Web-spiders which scan the web for stolen images face the same problem as stated
above: the web-server detects that the request originates from the spider and it will then not
forward any illegal images to that site, or replace them with dummy images. Access
controlled or pay-sites present another problem to the web-spider. It cannot access the
contents without paying or authenticating itself. Unfortunately we believe that the largest part
of copyright infringement comes from access-controlled sites. These problems present high
barriers for any web-spider to overcome.
Various schemes have been proposed to do rights management [13,25]. Rights
management is where the information distributor can give usage access rights to the client.
For example he could declare that only viewing, not printing of the image is possible. This
sounds feasible at first, but again, things look differently in the real world. Even if using a
tamper-proof smartcard in everybodys PC that checks the access rights prior to any action
enforces the rights, the image can be stolen anyway. While the image is displayed on the
screen, the information has to be present somewhere in the PC's memory. Therefore that
memory can also be read by another program and saved. Watermarking faces a similar
problem: the image already has to be present somewhere in memory before the watermark
can be extracted. Similar concerns are discussed with respect to DVD rights management in
[37].
--------------------------------------------------------------------------------------
------------------------------------------

3
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Another reason things look differently in the real world is that the systems are used in
ways not foreseen by the system designers. In watermarking we could imagine the following
problem: Instead of showing the stolen image at once on a web-page, Mallory chops up the
image into small blocks and creates many small images. In the web page, the images are then
arranged such that the viewer can see the original image again. Unfortunately, the individual
images are too small to carry an extractable watermark. Only if multiple blocks were merged,
the watermark could be extracted again. Therefore a web-spider could not detect the fraud,
since it checks each image individually. This mosaic attack was also discovered
simultaneously by Anderson and Petitcolas.
1.5. Hacker Circles
If Mallory wanted to remove a watermark and she knows about ways of using her
image processing toolkit with which she could potentially remove the watermark she would
probably hesitate to do so. First, she knows that there is a large fine for copyright
infringement and she is unsure if the attack would be successful. Second she is quite unsure
of which transformations to apply - she did tests on her own watermarks, but are the
commercial marks the same?
Hackers are globally well organized and already today there exist multiple programs
to remove watermarks. Therefore Mallory does not need to be a signal or image processing
expert to remove the copyright with a high certainty. The power of such underground activity
to create high quality watermark removal software should not be underestimated.
--------------------------------------------------------------------------------------
------------------------------------------

4
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter 2
WATERMARKING
2.1 INTRODUCTION
The process of embedding information into another object/signal can be termed as
watermarking. The watermark is one, which is imperceptibly added to the cover-signal in
order to convey the hidden data. The basic idea of watermarking is to embed a stealthy image
(unperceivable image) into the image needed to be secured. The stealthy image is derived
from the original image and should carry sufficient information to ensure copyright
verification. It should be imperceptible, undetectable, and robust enough to survive any
attempts to destroy it.
Cryptography has been the cornerstone [38] of technologies used to protect
intellectual property rights. For example, cable and satellite television signals are commonly
scrambled to prevent unauthorized viewing. However, cryptography only protects the work
during transmission or distribution. It provides no protection after the work is decrypted. And
all work must eventually be decrypted if consumers are to enjoy the photograph, music, or
movie. Watermarking is a technology that complements cryptography by embedding
imperceptible signals in a work. These signals remain in the work after decryption and even
after conversion to the analog world, and their use has been proposed for a variety of digital
rights management purposes.
Digital watermarking is only possible because our vision system is not perfect. A
perfect vision system can be defined as one that has the capability of distinguishing even the
slightest changes in visual stimuli. For the human visual system this is, however, not the case.
A digital watermark is best described by comparing it to a traditional paper watermark.
Traditional watermarks are added to some types of paper to offer proof of authenticity. They
are imperceptible, except when the paper is held up to a light for inspection. Similarly, digital
watermark is added to a digital image in a way that can be seen by a computer but is
imperceptible to the human eye. A digital watermark carries a message containing
information about the creator or distributor of the image, or even about the image itself.
--------------------------------------------------------------------------------------
------------------------------------------

5
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
A digital watermark is used to communicate copyright information about an image in
order to reduce copyright infringement. A person opening a digitally watermarked image in
an image-editing application, the reader receives notification through a copyright symbol
((c)) that the image contains copyright and ownership information. The digital watermark can
provide a link to complete contact details for the copyright holder or image distributor,
making it easy for the viewer to license the image, license another one like it, or commission
new work. Clearly watermarking is a form of communications, and a number of models have
been proposed. A watermarking system consists of a watermark embedder and a watermark
detector, which are analogous to a transmitter and receiver. The most common model then
views the original work as the communication channel and the watermark as the modulated
signal that conveys the message. However, unlike traditional communications, watermarking
must also pay careful attention to the fidelity of the underlying cover work. This has
sometimes been modeled as a power constraint at the embedder (transmitter). A more recent
model considers watermarking as a form of communications with side information. In this
model, the cover work is no longer treated as noise but as side information available to either
the transmitter and/or receiver. This communications perspective holds the promise of greatly
improving the channel capacity of a watermarking system.
Unlike printed watermarks, which are intended to be somewhat visible, digital
watermarks are designed to be completely invisible, or in the case of audio clips, inaudible.
Moreover, the actual bits representing the watermark must be scattered throughout the file in
such a way that they cannot be identified and manipulated. And finally, the digital watermark
must be robust enough so that it can withstand normal changes to the file, such as reductions
from lossy compression algorithms.
Ideal properties of a digital watermark have been stated in many articles and papers.
These properties include:
1) A digital watermark should be perceptually invisible to prevent obstruction of the original
image.
2) A digital watermark should be statistically invisible so it cannot be detected or erased.
3) Watermark extraction should be fairly simple. Otherwise, the detection process requires
too much time or computation.
--------------------------------------------------------------------------------------
------------------------------------------

6
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
4) Watermark detection should be accurate. False positives, the detection of a nonmarked
image, and false negatives, the non-detection of a marked image, should be few.
5) Numerous watermarks can be produced. Otherwise, only a limited number of images may
be marked.
6) Watermarks should be robust to filtering, additive noise, compression, and other forms of
image manipulation.
7) The watermark should be able to determine the true owner of the image.
2.1.1 REASONS TO DIGITALLY WATERMARK IMAGES
There are several reasons for using digital watermarks. One is simple pride of
authorship--the same reason that artists sign their paintings. Digital images are especially
prone to loss of authorship, as seen by the avalanche of images posted daily on the World
Wide Web, few of which have any reference to the photographer or illustrator.
Another reason is, more important, commercially: a digital watermark communicates
the name and rights of an image's copyright holder. With this information, the image
consumer can quickly and easily contact the image creator or distributor to license the work
or commission additional work.
Overall, digital watermarking provides creators and distributors of images three main
benefits:
Protect your valuable images by communicating your copyright.
Track down uses of your images on the Web.
Generate incremental revenue by embedding an ad in every image.
In general, there are two types of digital watermarks addressed in the existing
literature: visible and invisible watermarks. A visible watermark typically contains a visible
message or a company logo indicating the ownership of the image. On the other hand, the
invisibly watermarked digital content appears visually very similar to the original. Examples
include images distributed over Internet with watermarks embedded in them for copyright
protection. Those which fail can be classified as visible watermarks. Examples include logos
used in papers in currencies.
--------------------------------------------------------------------------------------
------------------------------------------

7
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
2.2. TYPES OF WATERMARK :
Invisible watermarks can be broadly classified into two types, robust and fragile (or
semi-fragile) watermarks.
2.2.1. Robust Watermark
A robust watermark [2]-[7] should be stuck to the document it has been embedded in,
in such a way that any signal transform of reasonable strength cannot remove the watermark.
Hence a pirate willing to remove the watermark will not succeed unless they debase the
document too much to be of commercial interest. This form of watermark is the very
challenging and attracts most research.
For any watermark to be robust, the watermark information must be embedded in the
target medium in such a way that removing this information irreparably damages the
medium.
When considering static images, the commonly recognized transformations that a
watermark should survive are:
Rotation, scaling, translation, mirroring.
Filtering (Gaussian blur, image sharpening, etc.)
Adding noise to the image, adding jitter (duplicating and removing lines/columns of
the image), cropping (removing the sides).
Color remapping (color, quantization, and adjustments in brightness or contrast).
Lossy compression (JPEG, MPEG, fractal compression).
All of these transformations can preserve the value of the image to the user. After they have
been applied, the image is still recognizably derived from the original image. This concept
can be used as the foundation of a robust watermarking method [29]; if the image can be
transformed into some space which is invariant to the value preserving transforms listed
above, the watermark can be applied in that space before reversing the transformation to
arrive back at the original image. This watermark will then also be invariant under these
transformations.
2.2.2. Semi-fragile Watermark
A (semi-) fragile watermark [8]-[11] is a mark, which is (highly) sensitive to a
modification of the stego-medium. A fragile watermarking scheme should be able to detect
--------------------------------------------------------------------------------------
------------------------------------------

8
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
any change in the signal and identify where it has taken place and possibly what the signal
was before modification. It serves at proving the authenticity of a document.
2.3 PROPERTIES OF WATERMARKING TECHNIQUES
2.3.1. Unobtrusive (Difficult to notice)
The watermark should be perceptually invisible to the viewer nor should the watermark
degrade the quality of the content. In earlier work [33,34,35,36],Cox , had used the term
imperceptible", and this is certainly the ideal. However, if a signal is truly imperceptible,
then perceptually based lossy compression algorithms should, in principle, remove such a
signal. Current state-of-the-art compression algorithms probably still leave room for an
imperceptible signal to be inserted. This may not be true of next generation compression
algorithms. Thus, to survive the next generation of lossy compression algorithms, it will
probably be necessary for a watermark to be noticeable to a trained observer.
2.3.2. Robustness
The watermark must be difficult (hopefully impossible) to remove. Of course, in theory, any
watermark may be removed with sufficient knowledge of the process of insertion. However,
if only partial knowledge is available, for example, the exact location of the watermark within
an image is unknown, then attempts to remove or destroy a watermark by say, adding noise,
should result in severe degradation in data fidelity before the watermark is lost. In particular,
the watermark should be robust to the following attacks and characteristics.
i. Universality: The same digital watermark algorithm should apply to all three media
types. This is potentially helpful in the watermarking of multimedia products. Also, this
feature is conducive to implementation of audio, image, and video watermarking algorithms
on common hardware.
ii. Tamper-resistance: The watermarking techniques should be robust to legitimate
signal distortions as well as intentional attacks to remove or tamper with the digital
watermark.
--------------------------------------------------------------------------------------
------------------------------------------

9
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
iii. Common Signal Processing: The watermark should still be retrievable even if
common signal processing operations are applied to the data. These include, digital-to-analog
and analog-to-digital conversion, resampling, requantization (including dithering and
recompression), and common signal enhancements to image contrast and color, or audio bass
and treble, for example.
iv. Common Geometric Distortions: Watermarks in image and video data should also
be immune from geometric image operations such as rotation, translation, cropping, and
scaling.
v. Subterfuge Attacks: In addition, the watermark should be robust to collusion by
multiple individuals who each possess a watermarked copy of the data. That is, the
watermark should be robust to combining copies of the same data set to destroy the
watermarks. Further, if a digital watermark is to be used as evidence in a court of law, it must
not be possible for colluders to combine their images to generate a different valid watermark
with the intention of framing a third-party.
2.3.3. Secure
The watermarked image should not reveal any clues of the presence of the watermark,
with respect to unauthorized detection, or (statistical) indefectibility or unsuspicious (not the
same as imperceptibility).
--------------------------------------------------------------------------------------
------------------------------------------

10
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter - 3
TECHNICAL PROBLEMS OF WATERMARKING
3.1. THE WATERMARKING PROCESS
watermarking.
The technical challenge that watermarking presents is the successful concealment of
the watermark signal in a much larger bandwidth medium. As [28] points out, usually an
attacker's ability to read or, worse, change or remove the watermark, is of more concern than
their ability to detect its presence. In the face of perfect compression methods, it is not clear
that it is possible to conceal the presence of a watermark.
An often-used method for covert channel communication over a higher-bandwidth
medium is spread spectrum communications. This method can also be applied to
watermarking images, as described in [30,31]. It has the advantage of having a long history in
both military and civilian digital communications, and is thus both well understood and of
--------------------------------------------------------------------------------------
------------------------------------------

11
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
proven reliability. At the current time, it appears the most promising framework for
embedding watermarks in either images or other digital media data streams.
Watermarks are most often used for copyright control. It is commonly recognized that
no watermarking method can withstand all possible attacks. Thus, in the domain of network-
accessible digital media, it may be better to regard a watermark as an aid to quickly
comparing possible copies of a copyrighted work to the original, in order to locate copyright
violations, than as an indicator of copyright in itself.
3.2. ATTACKS ON WATERMARKS
The possible attacks against watermarks are wide and varied. Cox and Linnartz [37]
and Anderson [14] present a number of such attacks, including:
Image modification attacks. These use image transformations such as those listed
above.
Bit-level attacks. If the attacker has access to a watermark presence detector, the
contents and location of the watermark can be derived. This also makes it much easier
to remove a watermark.
Watermark-insertion attacks. If the attacker has access to a watermark insertion
device, and the watermarking process is not a one-way function, it is possible to
recover the original, unwatermarked image, by pre-distorting the copy, and
rewatermarking it.
Statistical averaging attacks. The attacker uses multiple watermarked images to
estimate the watermark, and then subtracts this from the image. This is especially a
problem with video since a large number of watermarked frames is available.
Scrambling attacks. By inserting a scrambler before the watermark detector, and a de-
scrambler after it, detection of the watermarking can be avoided.
As it can be seen from this list, even with a perfect watermarking method, there are
various system-level attacks that can frustrate the secure use of watermarks in a copyright
scheme.
3.3. PRIORITIZING ATTACK RESISTANCE
As making a watermark resistant to a large number of image transformation attacks is
a difficult task, it is important to prioritize these transforms. A watermark should aim to be
most resistant to the most common attacks; as we have pointed out, complete watermark
--------------------------------------------------------------------------------------
------------------------------------------

12
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
security is an unachievable goal. It is more realistic to aim for a high probability of recovery.
A corollary of this is that a watermarking method must first of all be robust against those
transformations used by legal users of the image. It is tempting to concentrate on the various
complex attacks that malicious attackers might employ. However, this is pointless if the
method is not properly resistant to image edits made by either the original owner or valid
users of the image. Roughly in order of merit, a watermarking method must handle:
Scaling, especially (filtered) down-sampling.
GIF and JPEG compression. (Color quantization, and lossy compression.)
Simple brightness, contrast, or gamma adjustment.
Border cropping.
3.4. HANDLING SMALL IMAGES
Below a certain size it becomes difficult to embed and reliably recover a watermark
from an image. Even the better watermarking methods have difficulty retaining watermarks
in 100 by 100 pixel images that are lossily compressed. It is not clear that this problem can
ever be solved by technical means, as it is stems from a simple lack of bandwidth. As the
effective bandwidth of the picture is (roughly) proportional to its area, halving the dimensions
of it reduces the available bandwidth by 4. Often small images are also color-reduced for
efficient storage, further restricting the available bandwidth. This observation provides the
foundation of the mosaic attack described in section 2.
It should be noted that as image size drops, the concern about copyright violations
also drops. Small text extracts from larger works are allowed for personal use under the
copyright act [32]. In music, samples of a couple of seconds or less, especially if they have
been distorted in some way, are generally accepted as fair use". Longer, more recognizable
samples require copyright clearance.
Most of the literature on watermarks avoids mentioning this problem, because it
seems obvious. However end users of watermarking are often not aware of the problems
small images can cause, or if they are, are uncertain as to how large an image has to be before
its watermark is secure. This makes it essential that a watermarking system has some
mechanism for warning a user that a small-image watermark may not be robust.
In order to fulfill multipurpose applications, several multipurpose watermarking
algorithms based on wavelet transform [11] and fast Fourier transform [12] have been
--------------------------------------------------------------------------------------
------------------------------------------

13
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
presented. Traditional digital watermarking schemes are mainly based on DCT and DWT
transforms. Recently, some robust image watermarking techniques based on vector
quantization (VQ) [15][21] have been presented. [15][18] embed the watermark
information into the encoded indices under the constraint that the extra distortion is less than
a given threshold. [19] embeds the watermark bit in the dimension information of the variable
dimension reconstruction blocks of the input image. [20], [21] embed the watermark
information by utilizing the properties, such as mean and variance, of neighboring indices.
Multipurpose watermarking method based on multistage vector quantization will be
discussed.
Chapter 4
VECTOR QUANTIZATION
4.1. INTRODUCTION
Vector quantization (VQ) has become an attractive blockbased encoding method for
image compression in recent years. It can achieve a high compression ratio. In environments
such as image archival and one-to-many communications, the simplicity of the decoder
makes VQ very efficient. In brief, VQ can be defined as a mapping from k-dimensional
Euclidean space R
k
into a finite subset C={c
i
/i=0,1,,N-1} that is generally called a
codebook, where c
i
is a codeword and N is the codebook size. VQ first generates a
representative codebook from a number of training vectors using, for example, the well-
known iterative clustering algorithm [22] that is often referred to as the Generalized Lloyd
Algorithm (GLA).
--------------------------------------------------------------------------------------
------------------------------------------

14
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
In VQ, the image to be encoded is first decomposed into vectors and then sequentially
encoded vector by vector. In the encoding phase, each k-dimensional input vector x=(x
1
,x
2
,
x
k
) is compared with the codewords in the codebook C={c
0
,c
1
,c
(N-1)
} to find the best
matching codeword c
i
=(c
i1
,c
i2
,,c
ik
) satisfying the following condition:
d(x,c
i
) = min d(x,c
j
) (1)
0jN-1
i.e., the distance between x and c
j
is the smallest, where d(x,c
j
) is the distortion of
representing the input vector by the codeword c
j
, which is often measured by the squared
Euclidean distance, i.e.
1
1
( , ) ( )
k
j jl
l
d x c x c
2
(2)
Then, the index of the best matching codeword assigned to the input vector x is
transmitted over the channel to the decoder. The decoder has the same codebook as the
encoder. In the decoding phase, for each index , the decoder merely performs a simple table
lookup operation to obtain c
i
and then uses c
i
to reconstruct the input vector . Compression is
achieved by transmitting or storing the index of a codeword rather than the codeword itself.
The compression ratio is determined by the codebook size and the dimension of the input
vectors, and the overall distortion is dependent on the codebook size and the selection of
codewords.
In 1980 Linde, Buzo and Gray proposed an improvement of the Lloyds technique.
They extended Lloyds results from mono- to k-dimensional cases. For this reason their
algorithm is known as the Generalized Lloyd Algorithm (GLA) or LBG [23] from the initials
of its authors.
This algorithm is similar to the k-means algorithm.
4.2 THE ALGORITHM:
1. Determine the number of codewords, N, or the size of the codebook.
--------------------------------------------------------------------------------------
------------------------------------------

15
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
2. Select N codewords at random, and let that be the initial codebook. The initial
codewords can be randomly chosen from the set of input vectors.
3. Using the Euclidean distance measure clusterize the vectors around each codeword.
This is done by taking each input vector and finding the Euclidean distance between it
and each codeword. The input vector belongs to the cluster of the codeword that
yields the minimum distance.
4. Compute the new set of codewords. This is done by obtaining the average of each
cluster. Add the component of each vector and divide by the number of vectors in the
cluster.
m
j
ij i x
m
y
1
1
where i is the component of each vector (x, y, z, ... directions), m is the number of
vectors in the cluster.
1. Repeat steps 2 and 3 until the either the codewords don't change or the change in the
codewords is small.
This algorithm is by far the most popular, and that is due to its simplicity. Although it is
locally optimal, yet it is very slow. The reason it is slow is because for each iteration,
determining each cluster requires that each input vector be compared with all the
codewords in the codebook.
4.3 WORKING OF THE SEARCH ENGINE
Although VQ offers more compression for the same distortion rate as scalar
quantization, yet is not as widely implemented. This is due to two things. The first is the
time it takes to generate the codebook, and second is the speed of the search. Many
algorithms have been proposed to increase the speed of the search. Some of them reduce the
--------------------------------------------------------------------------------------
------------------------------------------

16
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
math used to determine the codeword that offers the minimum distortion; other algorithms
preprocess the codewords and exploit underlying structure.
The simplest search method, which is also the slowest, is full search. In full search an
input vector is compared with every codeword in the codebook. If there were M input
vectors, N codewords, and each vector is in k dimensions, then the number of multiplies
becomes kMN, the number of additions and subtractions become MN((k - 1) + k) = MN(2k-1),
and the number of comparisons becomes MN(k - 1). This makes full search an expensive
method.
4.4 MEASURE OF PERFORMANCE VQ
How does one rate the performance of a compressed image or sound using VQ?
There is no good way to measure the performance of VQ. This is because the distortion that
VQ incurs, will be evaluated by us humans and that is a subjective measure. It can always be
resorted to good old Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR).
MSE is defined as follows:
( )
1
1
M
i i
i
MSE x x
M

where M is the number of elements in the signal, or image. For example, if we wanted to find
the MSE between the reconstructed and the original image, then we would take the difference
between the two images pixel-by-pixel, square the results, and average the results.
The PSNR is defined as follows:
( )
,
`
.
|
MSE
n
PSNR
2
1 2
10 log 10
where n is the number of bits per symbol. As an example, if we want to find the PSNR
between two 256 gray level images, then we set n to 8 bits.
In the decoding phase, not the original but the watermarked codeword is used to
represent the input image block. Therefore, the VQ-based digital image watermarking will
introduce some extra distortion. Whether the original image is required or not during the
--------------------------------------------------------------------------------------
------------------------------------------

17
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
watermark extraction is dependent on the embedding method. In these algorithms, the
codebook is open for users but the partition is the secret key. Experimental results show that
these algorithms are robust to VQ compression with high-performance codebooks, JPEG
compression and some spatial image processing operations. However, these algorithms are
fragile to rotation operations and VQ compression with low-performance codebooks.
4.5 WATERMARKING ALGORITHMS BASED ON INDEX PROPERTIES
To enhance the robustness to rotation operations and VQ compression operations,
some image watermarking algorithms [21], [22] based on the properties of neighboring
indices have been proposed. In [21], the original watermark W with size A
w
* B
w
is first
permuted by a predetermined key,key
1
, to generate the permuted watermark W
p
for
embedding. The original image X with size A * B is then divided into vectors x(m,n) with
size (A/A
w
) * (B/B
w
), where x(m,n) denotes the image block at the position of (m,n). After
that, each vector x(m,n) finds its best codeword C
i
in the codebook C and the index is
assigned to x(m,n), we can then obtain the indices matrix Y with elements y(m,n), which can
be represented by
U U
1 ) / (
0
1 ) / (
0
)) , ( ( ) (

Aw A
m
Bw B
n
n m x VQ X VQ Y
=
U U
1 ) / (
0
1 ) / (
0
) , (
Aw A
m
Bw B
n
n m y (3)
For natural images, the VQ indices among neighboring blocks tend to be very similar, so we
can make use of this property to generate the polarities . After calculating the variances of
y(m,n) and the indices of its surrounding blocks with
,
`
.
|

+

+

1
1
1
1
2 2
) , (
9
1
) , (
m
m i
n
n j
j i y n m
2
1
1
1
1
) , (
9
1
,
`
.
|

+

+

m
m i
n
n j
j i y (4)
We can obtain the polarities as follows:
--------------------------------------------------------------------------------------
------------------------------------------

18
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------

U U
1 ) / (
0
1 ) / (
0
) , (
Aw A
m
Bw B
n
n m p P (5)
where
'

otherwise
T n m
n m p
, 0
) , ( , 1
) , (
2
(6)
For convenience, the threshold T is set to be half of the codebook size, N/2.Then the final
embedded watermark or the secret key, key2 , can be generated with the exclusive-or
operation as follows:
key
2
= W
p
P (7)
After the inverse-VQ operation, both the reconstructed image X and the secret key key
2
work
together to protect the ownership of the original image. In the extraction process, we first
calculate the estimated polarities P from X and then obtain an estimate of the permuted
watermark, as follows:
W
p
=key
2
P (8)
Finally, the inverse permutation operation with key
1
can be performed to obtain the extracted
watermark W. In order to embed multiple watermarks, [22] also uses the mean of indices to
generate another polarities P
1
for embedding. Experimental results show that these algorithms
are robust to many kinds of attacks, including JPEG, VQ, filtering, blurring, and rotation.
However, these algorithms have the following problem.
1) The codebook should be used as a key, because if the user possesses the same codebook,
he can also embed his own watermark in the watermarked image without any modification.
In fact, unlike traditional watermarking methods, these kinds of watermarking
algorithms do not modify the VQ compressed cover work at all. The term fingerprint or
secure fingerprint may be more appropriate, and sometimes we can call this kind of
watermark zero-watermark. In view of unification, we use the term robust watermark
instead of secure fingerprint.
--------------------------------------------------------------------------------------
------------------------------------------

19
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter - 5
MULTISTAGE VECTOR QUANTIZATION
The basic idea of multistage VQ is to divide the encoding task into successive stages,
where the first stage performs a relatively crude quantization of the input vector using a small
codebook. Then, a second-stage quantizer operates on the error vector between the original
and quantized first-stage output. The quantized error vector then provides a second
--------------------------------------------------------------------------------------
------------------------------------------

20
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
approximation to the original input vector thereby leading to a refined or more accurate
representation of the input. A third-stage quantizer may then be used to quantize the second-
stage error to provide a further refinement and so on.
In this paper, we adopt a two-stage vector quantizer as illustrated in Fig. 1.
FIG 1. Two-stage VQ
It is the simplest case and can be used to generate the general multistage vector quantizer.
The input vector x is quantized by the initial or first-stage vector quantizer denoted by VQ
1
whose codebook is C
1
={c
10
,c
11
,,c
1(N1-1)
} with size N
1
. The quantized approximation 1 x is
then subtracted from x producing the error vector e
2
. This error vector is then applied to a
second vector quantizer VQ
2
whose codebook is C
2
={c
20
,c
21
,,c
2(N2-1)
} with size N
2
yielding
the quantized output 2 e .The overall approximation x to the input x is formed by summing
the first and second approximations, 1 x and 2 e . The encoder for this VQ simply transmits a
pair of indices specifying the selected codewords for each stage and the task of the decoder is
to perform two table lookups to generate and then sum the two codewords. In fact, the overall
codeword or index is the concatenation of codewords or indices chosen from each of two
codebooks. That is to say, this is a product code where the composition function of the
decoder is simply a summation of the reproductions from the different two VQ decoders.
Thus, the equivalent product codebook C can be generated from the Cartesian product C
1
*
C
2
. Compared to the full-search VQ with the product codebook C, the two-stage VQ can
reduce the complexity from N=N
1
* N
2
to N
1
+ N
2
.
5.1. EMBEDDING PROCESS
--------------------------------------------------------------------------------------
------------------------------------------

21
VQ
1
VQ
2
+
2 e
2 e
+
_
1 x
1 x
1 x
+
x
x
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Before describing the proposed algorithm, we make some assumptions. Let X be the
original image with size A * B, let W
R
and W
F
be the binary robust and semi-fragile
watermarks with size A
w
* B
w
respectively. Here, a small visually meaningful binary image V
with size a * b is replicated periodically to obtain the binary semi-fragile watermark W
F
with
size A
w
* B
w
that is large enough for embedding.
In each stage of the proposed algorithm, only one bit is embedded in each input image
block (or vector), so the dimension of each input vector or codeword is k=(A/A
w
) * (B/B
w
).
Assume that the first-stage codebook is C
1
={c
10
,c
11
,,c
1(N1-1)
} with size N
1
= 2
n1
and the
second-stage codebook is C
2
={c
20
,c
21
,,c
2(N2-1)
} with size N
2
=2
n2
where n
1
and n
2
are natural
numbers. Thus, a binary number with n
1
+ n
2
bits, in which the first n
1
bits stand for the index
of Stage 1 and the last n
2
bits denote the index of Stage 2, can represent the overall index. The
overall codeword can be selected from the equivalent product codebook C={c
0
,c
1
,c
(N-1)
}
with size N = N
1
* N
2
. In other words, if the index in codebook C
1
is i and the index in
codebook C
2
is j, then the equivalent overall index in the product codebook C is i + j * N
2
.
In our algorithm, the robust watermark W
R
and the semi-fragile watermark W
F
are
embedded in two stages, respectively. We embed the robust watermark in the first stage and
the semi-fragile one in the second stage to enhance the robustness and transparency of the
proposed algorithm. In what follows, we describe the two-stage embedding process
separately.
5.1.1. Robust Watermark Embedding Process
In the proposed algorithm, we adopt the method [20] based on index properties to
embed the robust watermark in the first stage as shown in Fig. 2.
--------------------------------------------------------------------------------------
------------------------------------------

22
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
FIG. 2 Robust Watermark embedding process in the first stage
The original watermark W
R
is first permuted by a predetermined key, key
1
, to generate
the permuted watermark W
RP
for embedding. The polarities P is then calculated. Finally, we
generate the final embedded watermark or the secret key, key
2
, with the exclusive-or
operation. After the first-stage embedding, we can obtain the reconstructed image X and the
error image X
1
as follows:
X=VQ
1
-1
[VQ
1
[X]]
X
1
= X X
This method has two problems and can be automatically solved, which will be discussed later
in the extraction process.
--------------------------------------------------------------------------------------
------------------------------------------

23
Division
Composition
Permutation
Embedding
Index Polarities
Computation
Normal VQ
Decoder
Nearest Neighbor
VQ Encoder
Input
Vectors
Codebook C
1
Output
Vectors
+
_
Encoded Indices
P
W
RP
X
X
1
X
key
2
key
1
W
R
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
5.1.1.2. The Embedding Algorithm
Let the input image be X with size M x N. The robust watermark with VQ is to be
embedded into X and a watermarked reconstruction and the secret keys associated with the
embedded watermark is to be output. These secret keys are registered to the third party to
certify the ownership of original multimedia source.
The two binary-valued watermarks to be embedded are W
R
and W
F
. First the VQ
operation is performed and codewords with the nearest search algorithm are obtained.
Afterwards the watermarks can be embedded with the characteristics of the indices in the VQ
domain. In order to survive the picture cropping attacks, a pseudo-random number traversing
method is applied to permute the watermark to disperse its spatial-domain relationships. With
a predetermined key, key
1
, in the pseudo-random number generating system, then
W
RP
=permute(W
R
,key
1
)
Then the permuted version W
RP,
is used for embedding into VQ indices.
In the VQ encoding procedure, X is divided into vectors x, then each x finds its
nearest codeword c
k
in the codebook C, and the index k is assigned to x for transmission.
During the transmission of the VQ indices, some errors might be induced by the channels,
thus the received indices would not be identical to the transmitted ones. While decoding with
the VQ indices, the decoder merely performs a table look-up process on the received index
k to obtain c
i
and the get the reconstructed image X.
The VQ is performed with the codebook size L. The codebook, C, and the codewords
therein, c
k
, k
[0,L-1], can be represented by

C={c
0
,c
1
,,c
L-1
}.
The block at the position (m,n) of the original source X is x(m,n). After performing
VQ, the indices Y and y(m,n) can be expressed with
Y=VQ(X),
Y(m,n)=VQ(x(m,n))
C.
To embed the binary watermark into the original source, some relationships to alter
the VQ indices into binary format for further embedding need to be introduced. The
polarities, P, of the VQ indices to embed the watermarks. For natural images, the VQ indices
--------------------------------------------------------------------------------------
------------------------------------------

24
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
among the neighboring blocks tend to be very similar, this characteristic can be used to
generate P.
Calculating the mean of y(m,n) and the indices of its surrounding blocks with
( , ) m n
=
1 1
1 1
1
( , )
9
m n
i m j n
y i j
+ +

Similarly obtaining the variance of y(m,n) and the indices of its neighboring block
with

2
(m,n) =
1 1
1 1
1
( , )
9
m n
i m j n
y i j
+ +

-
2
(m,n)
The polarities based on the means , variances and both the means and the variances , can be
decided with the pre-determined threshold value.
The threshold taken = for convenience and L= codebook size. Then the watermark
W
RP
can be embedded with P by the exclusive-or operation-
key
1
=W
RP
P.
5.1.2. Semi-Fragile Watermark Embedding Process
To embed one bit in each index in the second stage, we can adopt an index
constrained vector quantization (ICVQ) encoding scheme as shown in Fig. 3.
--------------------------------------------------------------------------------------
------------------------------------------

25
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------

FIG. 3. Semi-fragile watermark embedding process in the second stage
Because each index has n
2
bits, we can select an embedding position from n
2
candidate positions. Assume that we select Position key
3
, which is considered as a key, to
embed the watermark bit, where 0 key
3
n
2
-1 . Unlike the normal VQ encoder, the
embedding process for each watermark bit can be performed by searching the best match
codeword c
2p
for each input error vector x
1
under the constraints that the key
3
th bit of index is
equal to the watermark bit to be embedded. After the normal VQ decoder, we can obtain the
reconstructed error image X
1
, as follows:
X
1
= VQ
2
-1
[ICVQ
2
[X
1
]]
Then, we can obtain the final watermarked image , as follows:
X
W
= X
+ X
1
5.2 . EXTRACTION PROCESS
--------------------------------------------------------------------------------------
------------------------------------------

26
Division
Composition
Normal VQ
Decoder
Index Constrained
VQ Encoder
Codebook C
2
Output
Vectors
Input
Vectors
+
+
X
1
X
1
X
W
Encoded Indices
Watermarking
Position key
3
W
F
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
To enhance the security of the embedding process, the equivalent product codebook C is used
in the extraction process as shown in Fig. 4, that is to say, the two-stage codebooks are used
as secret keys while the product codebook is open for users.

FIG. 4. Watermark Extraction Process
In addition, because the users do not know the codebook sizes used in two-stage VQ
either, how to segment the overall index into two stage indices is also a secret key, key
5
, to
--------------------------------------------------------------------------------------
------------------------------------------

27
Composition
Inverse
Permutation
XOR
Bit judgment
Index Polarities
Computation
Index
Segmentation
Normal VQ
Encoder
Division
Inverse
Permutation
Codebook C
u
X
w
key
4
Segmentation Position key
5
Codebook C
Input
Vectors
Encoded
Indices
Indices of Stage1
Indices of Stage2
Watermarking Position key
3
Extracted Bits
Extracted Semi-Fragile
Extracted Robust Watermark W
ER
Watermark W
EF
key
2
key
1
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
users. In order to make the embedding algorithm more secretly, we can also permute the
product codebook and then publicize the permuted codebook C
u
for users.
The extraction process can be performed without the original image and can be
described as follows. First, perform the inverse permutation operation with key
4
on Codebook
C
u
to obtain the product codebook C. Second, the watermarked image X
w
is divided into
blocks or vectors. Third, the normal VQ encoder performs the nearest neighbor codeword
search on all input vectors to obtain the encoded overall indices. Fourth, according to the two
stage codebook sizes, each overall index is segmented into two indices. One is for robust
watermark extraction; the other is for semi-fragile watermark extraction. Finally, the robust
and semi-fragile watermarks are extracted independently. For the robust watermark
extraction, the polarities P are computed from the indices of Stage 1, and then XOR operation
is performed between P and key
2
to obtain the extracted permuted robust watermark W
EPR
,
and finally perform inverse permutation operation with key
1
to obtain the extracted robust
watermark W
ER
. For the semi-fragile watermark extraction, the key
3
th
bit of each index of
Stage 2 can be checked to obtain the extracted watermark bit, where key
3
is just the
watermarking position, and then piece all extracted bits together to form the extracted semi-
fragile watermark W
EF
. From the above, it can be seen that the advantages of using ICVQ in
the semi-fragile watermarking are as follows:
1) Both the embedding and the extraction processes are very simple.
2) The extraction is blind.
3) The embedded position can be controlled by a key for more security.
In Section C above, the problem of the robust embedding technique [21] are pointed
out. However, in this algorithm, this problem can be automatically solved. Using not the two-
stage codebooks but the equivalent product codebook to extract the watermarks can solve this
problem. From Fig. 4, it can be seen that the extraction time is determined by the codebook
size of C. If N is very large, then the full search VQ encoding is rather a time-consuming
process, so fast codeword search algorithm [24] is used in the proposed algorithm. A new
fast codeword search algorithm for image vector quantization (VQ) is introduced. This
algorithm performs a fast codeword search in the Hadamard Transform (HT) domain using
the partial distance search (PDS) technique. Experimental results show that the algorithm
needs only 2-3% or the distortion calculations of the exhaustive search method
--------------------------------------------------------------------------------------
------------------------------------------

28
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter 6
DESIGN
FIG 6.1. The LBG procedure
--------------------------------------------------------------------------------------
------------------------------------------

29
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
--------------------------------------------------------------------------------------
------------------------------------------

30
Start
Initial codebook (Y0)
M=0;
D
-1
=+
New partition calculation
S
m
=P(Y
m
)
Distortion (D
m
) calculation
(D
m-1
-D
m
) / D
m
<
New codebook calculation
Y
m+1
=X(S
m
)
ELGB Block
m=m+1
Final codebook (Y
m
);
End
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
FIG. 6.2. ELBG codebook optimization
Fig. 6.3. High-level flow-chart of the ELBG block
--------------------------------------------------------------------------------------
------------------------------------------

31
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
FIG. 6.4. Detailed description of the ELBG block
--------------------------------------------------------------------------------------
------------------------------------------

32
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter 7
TERMINOLOGY
7.1. Asymmetric Watermarking :
In this case, the detection process (and in particular the detection key) is fully known
to anyone as opposed to blind watermarking where a secret key is required. So here, only a
'public key' is needed for verification and a 'private key' (secret) is used for the embedding
though. Knowledge of the public key does not help to compute the private key (at least in a
reasonable time), it does not either allow removal of the mark nor it allows an attacker to
forge a mark.
7.2. Capacity :
It describes how many information bits can be embedded into the cover data. It also
addresses the possibility of embedding multiple watermarks in one document in parallel.
7.3. Codebook :
In Vector Quatization, k-dimensional Euclidean space R
k
, is mapped into a finite
subset C={C
i
/ i=0,1,,N-1}. The subset C is generally called a codebook, where C
i
is a
codeword and N is the codebook size.
7.4. Complexity :
It describes the effort and time we need to embed and retrieve a watermark. This
parameter is essential for real-time applications.
7.5. Copy Protection :
Copy protection attempts to find ways, which limits the access to copyrighted
material and/or inhibit the copy process itself. A recent example is the copy protection
mechanism on DVDs.
7.6. Copyright Protection :
--------------------------------------------------------------------------------------
------------------------------------------

33
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Copyright protection inserts copyright information into the digital object
without the loss of quality. Whenever the copyright of a digital object is in question, this
information is extracted to identify the rightful owner. It is also possible to encode the
identity of the original buyer along with the identity of the copyright holder, which allows
tracing of any unauthorized copies.
7.7. Cryptography :
It is about protecting the content of messages (their meaning).
7.8. Digital Image :
A digital image a[m,n] described in a 2D discrete space is derived from an analog
image a(x,y) in a 2D continuous space through a sampling process that is frequently referred
to as digitization. The effect of digitization is shown in Figure 10. The 2D continuous image
a(x,y) is divided into N rows and M columns. The intersection of a row and a column is
termed a pixel. The value assigned to the integer coordinates [m, n] with {m=0,1,2,...,M-1}
and {n=0,1,2,...,N-1} is a[m,n]. In fact, in most cases a(x,y) is actually a function of many
variables including depth (z), color ( ), and time (t).
FIG. : Digitization of a continuous image.
7.9. Embedding-key :
It is a secret used to embed the mark in the original image.
7.10. Encryption :
--------------------------------------------------------------------------------------
------------------------------------------

34
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Encryption protects the images during their transmission. With encryption, an
eavesdropper does not have access to the on-line image when it is transferred. But when the
user has deciphered the image, then this image does not have any copyright protection
anymore.
7.11. Extraction-key :
It is a key used to detect or extract a watermark.
Symmetric watermarking algorithms require use the same secret key for embedding
and extraction. Asymmetric algorithms use a secret key for embedding and a public key for
extraction. Keys are built in such a way that the private key cannot be computed from the
public one.
7.12. Fingerprinting :
Fingerprints are characteristics of an object that tend to distinguish it from other
similar objects. They enable the owner to trace authorized users distributing them illegally. In
the case of encrypted satellite television broadcasting, for instance, users could be issued a set
of keys to decrypt the video streams and the television station could insert fingerprint bits
into each packet of the traffic to detect unauthorized uses.
7.13. Image Authentication :
It is the task of the watermark detector to indicate only those regions that have been
modified to change the content or meaning of the image (tamper detection); the bright pixels
in the images on the right highlight tampered areas in the image.
7.21. Robust Watermarks :
A robust watermark should be stuck to the document it has been embedded in, in such
a way that any signal transform of reasonable strength cannot remove the watermark. Hence a
pirate willing to remove the watermark will not succeed unless they debase the document too
much to be of commercial interest.
7.22. Segmentation :
--------------------------------------------------------------------------------------
------------------------------------------

35
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
In the analysis of the objects in images, it can be distinguished between the objects of
interest and "the rest." This latter group is also referred to as the background. The techniques
that are used to find the objects of interest are usually referred to as segmentation techniques -
segmenting the foreground from background.
7.23. Semi-blind Watermarking :
In some cases you may need extra information to help your detector (in particular to
synchronize its random sequence on the possibly distorted test signal). In particular some
watermarking schemes require access to the 'published' watermarked signal, that is the
original signal just after adding the watermark.
7.24. (Semi-) fragile Watermarks:
A (semi-)fragile watermark is a mark which is (highly) sensitive to a modification of
the stegno-medium. A fragile watermarking scheme should be able to detect any change in
the signal and identify where it has taken place and possibly what the signal was before
modification.
7.25. Signature :
The owner of the image may electronically sign the image (with a hash function and a
signature algorithm), but since the signature is added as a suffix to the image, it can easily be
removed by anyone who gets the image.
7.26. Spatial Watermarks :
Spatial domain, additive watermarking is the same as additive watermarking in any
domain that is a linear transformation of the spatial domain, e.g. Fourier, block DCT,
wavelet, etc. It usually means that someone has created a watermark pattern that has the same
dimensions as the original image and has added the watermark pattern to the image.
7.27. Steganography :
As the purpose of steganography is having a covert communication between two
parties whose existence is unknown to a possible attacker, a successful attack consists in
--------------------------------------------------------------------------------------
------------------------------------------

36
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
detecting the existence of this communication (e.g., using statistical analysis of images with
and without hidden information).
Steganography is a very generic concept that consists in hiding messages in a way that
eavesdroppers or any monitors do not even know that there is a communication and a
message is being sent. Many data hiding techniques were invented for this purpose. Those
techniques inspired the development of watermarking algorithms for copyright protection.
7.28. Transparency :
It relates to the properties of the human sensory system. A transparent watermark
causes no perceptible artifacts or quality loss.
7.29. Vector Quantization :
The objective of VQ is the representation of a set of feature vectors
k
R X x by a
set, Y = {y1 , ..., y
NC
}, of N
C
reference vectors in
k
R
. Y is called codebook and its elements
codewords. The vectors of X are called also input patterns or input vectors. So, a VQ can be
represented as a function:
Y X q :
.
7.30. Verification :
It procedure distinguishes between private verification similar to symmetric
cryptography and public verification like in asymmetric cryptography. Furthermore, during
verification, we differ between invertible and noninvertible techniques, where the first one
allows the reproduction of the original and the last one provides no possibility to extract the
watermark without alterations of the original.
7.31. Visible and Invisible Watermarking :
Visibility is a term associated with the perception of the human eye. A watermarked
image in which the watermark is imperceptible, or the watermarked image is visually
identical to its original constitutes a invisible watermarking. Examples include images
distributed over internet with watermarks embedded in them for copyright protection. Those
which fail can be classified as visible watermarks. Examples include logos used in papers in
currencies.
--------------------------------------------------------------------------------------
------------------------------------------

37
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
7.32. Watermarking :
watermarking.
Watermarking is the robust embedding of a copyright information (e.g. time and date,
copyright identifiers or simply a correlation pattern) into a content. This content may be a text
or audio content but most of the time watermarking is applied to still or moving images.
Chapter 8
EXAMPLES
8.1. Original Lena Image
8.2 Original Robust Watermark
--------------------------------------------------------------------------------------
------------------------------------------

38
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
8.3. Original Semi-fragile Watermark
--------------------------------------------------------------------------------------
------------------------------------------

39
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
8.4. Extracted watermarks under no attacks :
The watermarked images along with extracted watermarks under no
attacks and NHS values are depicted in Fig. 10.
Watermarked Image Under No
attack
NHS = 1 NHS = 1
8.6. Spatial-Domain Image Processing Attacks :
Several spatial-domain image processing techniques, including
image cropping, median filtering, blurring, high-pass filtering, contrast
enhancement, and brightness enhancement are performed on the
watermarked image. The attacked images along with extracted
watermarks are depicted in Fig. 12. Except for the case of high-pass
filtering, the robust watermark can successfully survive with NHS > 0.7.
Although the NHS value of the extracted robust watermark in the high-
pass filtering case is somewhat smaller in this algorithm, the information
conveyed therein is still recognizable. For the case of image cropping in
the upper-left corner, the extracted semi-fragile watermark can locate the
cropping position. For each case, the semi-fragile watermark can be used
to verify the authenticity of the watermarked image.
--------------------------------------------------------------------------------------
------------------------------------------

40
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Image cropping NHS = 0.9109 NHS = 0.8654
Median filtering NHS = 0.9454 NHS = 0.8381
Image Blurring NHS = 0.7732 NHS = 0.5430
--------------------------------------------------------------------------------------
------------------------------------------

41
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Image High Pass filtering NHS = 0.5847 NHS = 0.4790
Contrast Enhancement NHS = 0.8603 NHS = 0.4880
Brightness Enhancement NHS = 0.8115 NHS = 0.3896

--------------------------------------------------------------------------------------
------------------------------------------

42
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter 9
APPLICATIONS
First applications, which came to mind, were related to copyright protection of digital
media. For everyone it is extremely easy to duplicate digital data and this even without any
loss of quality. Similar to the process in which artist artistically signed their paintings with a
brush to claim their copyrights, artists of today can watermark their work and hide for
example their name in the image. Hence, the embedded watermark will allow identifying the
owner of the work.
It is clear that this concept is also applicable to other media such as digital video and
audio. Especially the distribution of digital audio over the Internet in the MP3 format is
currently a big problem. In this scenario digital watermarking may be useful to set up a
--------------------------------------------------------------------------------------
------------------------------------------

43
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
controlled audio distribution and provide efficient means for copyright protection, usually in
collaboration with international registration bodies such as the IDDN-Inter Deposit Digital
Number.
In the field of data security, watermarks may be used for certification, authentication, and
conditional access. Certification is an important issue for official documents, such as identity
cards or passports.
Example of a protected identification card. The identification number "123456789" is
written in clear text on the card and hidden as a watermark in the identity photo.
Digital watermarking allows to mutually link information on the documents. That
means that some information is written twice on the document: for instance, the name of a
passport owner is normally printed in clear text and is also hidden as an invisible watermark
in the photo of the owner. If anyone would intend to counterfeit the passport by replacing the
photo, it would be possible to detect the change by scanning the passport and verifying the
name hidden in the photo does not match any more the name printed on the passport. The
picture below shows a printing machine from Intercard for various types of plastic cards.
--------------------------------------------------------------------------------------
------------------------------------------

44
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Printing machine for identification cards.
Another application is the authentication of image content. The goal of this
application is to detect alterations and modifications in an image. The three pictures below
illustrate an example of this application. The picture on the left shows an original photo of a
car that has been protected with a watermarking technology. In the center, the same picture is
shown but with a small modification: the numbers on the license plate have been changed.
The picture on the right shows the photo after running the watermark detection program on
the tampered photo. The tampered areas are indicated in white and it can clearly be seen that
the detected areas correspond to the modifications applied to the original photo.
Using digital watermarks for integrity verification.
--------------------------------------------------------------------------------------
------------------------------------------

45
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Other applications related to conditional access and copy-control are also possible.
For example conditional access to confidential data on CD-ROMs may be provided using
digital watermarking technology. The concept consists of inserting a watermark into the CD
label. In order to read and decrypt the data stored on the CD, the watermark has to be read
since it contains information needed for decryption. If someone copies the CD, he will not be
able to read the data in clear-text since he does not have the required watermark. The picture
below shows an example of a protected CD. To read the data on the CD, the user starts a
program on the CD. This program asks the user to put the CD on the scanner and then reads
the watermark. If the watermark is valid the program decrypts the data on the CD and gives
the user access the clear-text data.
Conditional access to confidential data stored on CD-ROMs.
Chapter 10
CONCLUSION
An efficient multipurpose watermarking algorithm based on multistage VQ has been
presented. In the this algorithm, the robust watermark is embedded in the first stage using the
robust watermarking method based on index properties [21] and the semi-fragile watermark
is embedded in the second stage using a simple index constrained method. Although the
encoded indices of the attacked watermarked image may be very different from the original
ones, the variance of neighboring indices does not vary too much. The first-stage
watermarking method is, therefore, robust. On the other hand, the second-stage watermarking
method is based on an index constrained codeword search procedure, in which the index is
--------------------------------------------------------------------------------------
------------------------------------------

46
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
modified according to the bit to be embedded. Any change in the encoded indices may
introduce the change in the extracted watermark bit. In other words, the second-stage
watermarking method can tolerate few modifications, so it is fragile to most intentional
attacks. Experimental results demonstrate that this method can be used for copyright
protection by extracting the first-stage watermark, and it can also be used for image
authentication by extracting the second stage watermark.
Compared with existing multipurpose watermarking algorithms, the
advantages of the proposed algorithm are as follows.
1) The proposed algorithm can tolerate rotation attacks with relatively
larger angles.
2) The semi-fragile and robust watermarks are extracted
independently and blindly.
3) Different codebooks are used in the embedding (the two-stage
codebooks) and the extraction (the product codebook) processes.
The final product codebook can be public for users, so the extraction
process sometimes can be performed publicly for special
applications.
4) Because the embedded position in the VQ index for the semi-fragile
watermark is secret, the two stage codebooks used in the
embedding process are secret, and the product codebook for
extraction is encrypted (not the same as that for users), it is hard for
the attacker to forge the valid semi-fragile watermark in the
tampered watermarked image.
5) The developed algorithm can be extended to meet three purposes,
e.g., digital fingerprint, copyright protection and image
authentication, by embedding three watermarks in a three-stage
VQ.
6) The VQ-based watermarking algorithm can reduce the amount of
the data transmitted.
Chapter 11
FUTURE WORK AND LIMITATIONS
--------------------------------------------------------------------------------------
------------------------------------------

47
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
1) The quality of the watermarked image obtained by existing VQ-
based watermarking algorithms is not high enough for copyright
protection. We may use other kinds of vector quantizers or
combine it with DWT or DCT to improve the image quality.
2) The semi-fragile watermark used in the proposed algorithm cannot
tell what kind of attack the watermarked image suffers from.
3) The human visual characteristics are not adopted in the VQ-based
watermarking systems.
4) The extraction is based on the product codebook, so this process is
time consuming if we use the full-search encoding algorithm.
5) From the experimental results, we can see that in the proposed
algorithm, the robust watermark should be embedded in the first
stage for obtaining high quality. How to analyze the embedding
order in general sense is a problem to solve.
--------------------------------------------------------------------------------------
------------------------------------------

48
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
Chapter 12
REFERENCES
[1] Zhe-Ming Lu, Dian-Guo Xu, and Sheng-He Sun, Multipurpose Image Watermarking
Algorithm Based on Multistage Vector Quantization, IEEE Transactions on Image
Processing, vol. 14, no. 6, June 2005
[2] J. J. K. ORuanaidh, W. J. Dowling, and F. M. Boland, Watermarking digital images for
copyright protection, Proc. Inst. Elect. Eng., vol. 143, no. 4, pp. 250256, Aug. 1996.
[3] I. J. Cox, J. Kilian, F. T. Leighton, and T. Shamoon, Secure spread spectrum
watermarking for multimedia, IEEE Trans. Image Process., vol. 6, no. 12, pp. 16731687,
Dec. 1997.
[4] M. D. Swanson, Z. Bin, and A. H. Tewfik, Multiresolution scene-based video
watermarking using perceptual models, IEEE J. Sel. Areas Commun., vol. 16, no. 5, pp.
540550, May 1998.
[5] G. Voyatzis and I. Pitas, The use of watermarks in the protection of digital multimedia
products, Proc. IEEE, vol. 87, no. 7, pp. 11971207, Jul. 1999.
[6] S. Pereira and T. Pun, An iterative template matching algorithm using the chirp-Z
transform for digital image watermarking, Pattern Recognit., vol. 33, no. 1, pp. 173175,
2000.
[7] Y. Wang, J. F. Doherty, and R. E. Van Dyck, A wavelet-based watermarking algorithm
for ownership verification of digital images, IEEE Trans. Image Process., vol. 11, no. 2, pp.
7788, Feb. 2002.
[8] P. S. L. M. Barreto, H.Y. Kim, andV. Rijmen, Toward secure public-key blockwise
fragile authentication watermarking, Proc. Inst. Elect. Eng., vol. 149, no. 2, pp. 5762, Apr.
2002.
[9] M. U. Celik, G. Sharma, E. Saber, and A. M. Tekalp, Hierarchical watermarking for
secure image authentication with localization, IEEE Trans. Image Process., vol. 11, no. 6,
pp. 585595, Jun. 2002.
[10] L. Jaejin and S. W. Chee, A watermarking sequence using parities of error control
coding for image authentication and correction, IEEE Trans. Consum. Electron., vol. 46, no.
2, pp. 313317, May 2000.
--------------------------------------------------------------------------------------
------------------------------------------

49
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
[11] D. Kundur and D. Hatzinakos, Digital watermarking for telltale tamper proofing and
authentication, Proc. IEEE, vol. 87, no. 7, pp. 11671180, Jul. 1999.
[12] C. S. Lu and H. Y. M. Liao, Multipurpose watermarking for image authentication and
protection, IEEE Trans. Image Process., vol. 10, pp. 15791592, Oct. 2001.
[13] Brad Cox. Superdistribution: objects as property on the electronic frontier. Addison-
Wesley, 1996.
[14] Fabien A. P. Petitcolas, Ross J. Anderson, and Markus G. Kuhn. Attacks on Copyright
Marking Systems. In Second International Workshop on Information Hiding, April 1998.
[15] C. Busch and S. D. Wolthusen, Tracing data diffusion in industrial research with robust
watermarking, in Proc. 4th IEEE Workshop on Multimedia Signal Processing, 2001, pp.
207212.
[16] Z. M. Lu and S. H. Sun, Digital image watermarking technique based on vector
quantization, Electron. Lett., vol. 36, no. 4, pp. 303305, Feb. 2000.
[17] Z. M. Lu, J. S. Pan, and S. H. Sun, VQ-based digital image watermarking method,
Electron. Lett., vol. 36, no. 14, pp. 12011202, Jul. 2000.
[18] Z. M. Lu, C. H. Liu, and S. H. Sun, Digital image watermarking technique based on
block truncation coding with vector quantization, Chin. J. Electron., vol. 11, no. 2, pp. 152
157, Apr. 2002.
[19] J. Minho and K. HyoungDo, A digital image watermarking scheme based on vector
quantization, IEICE Trans. Inf. Syst., vol. E85-D, no. 6, pp. 10541056, Jun. 2002.
[20] A. Makur and S. S. Selvi, Variable dimension vector quantization based image
watermarking, Signal Process., vol. 81, no. 4, pp. 889893, Apr. 2001.
[21] H. C. Huang, F. H. Wang, and J. S. Pan, A VQ-based robust multiwatermarking
algorithm, IEICE Trans. Fund., vol. E85-A, no. 7, pp. 17191726, Jul. 2002.
[22] , Efficient and robust watermarking algorithm with vector quantization, Electron.
Lett., vol. 37, no. 13, pp. 826828, Jun. 2001.
[23] Y. Linde, A. Buzo, and R. M. Gray, An algorithm for vector quantizers design, IEEE
Trans. Commun., vol. 28, no. 1, pp. 8495, Jan. 1980.
[24] Z. M. Lu, J. S. Pan, and S. H. Sun, Efficient codeword search algorithm based on
Hadamard transform, Electron. Lett., vol. 36, no. 16, pp. 13641365, Aug. 2000.
[25] Intertrust. Website. http://www.intertrust.com, 1997.
--------------------------------------------------------------------------------------
------------------------------------------

50
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
[26] Alma Whitten. Making Security Usable. PhD proposal at Carnegie Mellon University,
January 1998. Work in progress.
[27] Robert A. Cinque. Making Cyberspace Safe for Copyright: The Protection of Electronic
Works in a Protocol to the Berne Convention. 18 FORDHAM INT'L L.J., 1995. citing Berne
Convention art 7(6).
[28] Ross J. Anderson and Fabien A.P. Petitcolas. On The Limits of Steganography. In
IEEE Journal on Selected Areas in Comunications - Special Issue on Copyright & Privacy
Protection, 1998. http://www.cl.cam.ac.uk/users/fapp2/ papers/ On The Limits Of
Steganography/.
[29] J. J. K. Ruanaidh and Thierry Pun. Rotation, Scale and Translation Invariant Digital
Image Watermarking. Submitted to Signal Processing, February 1997.
[30] I. Cox, J. Killian, T. Leighton, and T. Shamoon. Secure spread spectrum watermarking
for images, audio and video. In Proceedings of the IEEE Int. Conf. on Image Processing
ICIP-96, pages 243{246, Lausanne, Switzerland, September 16-19 1996.
[31] J. J. K. Ruanaidh, W. J. Dowling, and F. M. Boland. Phase watermarking of digital
images. In Proceedings of the IEEE Int. Conf. on Image Processing ICIP-96, pages 239-242,
Lausanne, Switzerland, September 16-19 1996.
[32] Robert A. Cinque. Making Cyberspace Safe for Copyright: The Protection of Electronic
Works in a Protocol to the Berne Convention. 18 FORDHAM INT'L L.J., 1995. citing Berne
Convention art 7(6).
[33] I.J. Cox, J. Kilian, T. Leighton, and T. Shamoon. A secure, robust watermark for
multimedia. In ed. R. Anderson, editor, Information Hiding: First Int. Workshop Proc.,
volume 1174 of Lecture Notes in Computer Science, pages 185{206. Springer-Verlag, 1996.
[34] I.J. Cox, J. Kilian, T. Leighton, and T. Shamoon. Secure spread spectrum watermarking
for images audio and video. In IEEE Int. Conf. on Image Processing, volume 3, pages
243{246, Lausanne,Switzerland, September 1996.
[35] I.J. Cox, J. Kilian, T. Leighton, and T. Shamoon. Secure spread spectrum watermarking
for multimedia. IEEE Trans. on Image Processing, 6(12):1673-1687, 1997.
[36] I.J. Cox and L. Miller Matt. A review of watermarking and the importance of
perceptual modeling. In Proc. of Electronic Imageing'97, February 1997.
[37] Ingemar J. Cox and Jean-Paul M.G. Linnartz. Public watermarks and resistance to
tampering. In Proceedings of the IEEE International Conference on Image Processing, 1997.
--------------------------------------------------------------------------------------
------------------------------------------

51
Seminar Report on
------------------------------------------------------------------------------------------
--------------------------------------------
[38] Mauro Barni, Franco Bartolini, Ingemar J. Cox, Juan Hernandez, Fernando Perez
Gonzalez, Digital watermarking for copyright protection: A Communication perspective,
IEEE Communications Magazine, August 2001.
WEB SITES
http://www.ieeexplore.ieee.org
http://www.altern.com/watermark
http://www.cl.cam.ac.uk/~mgk25/stirmark.html
http://www.chez.com/pdufour/ zip/digimarc.zip
http://www.watermarkingworld.org/
http://www.cosy.acat/~pmeerw/watermarking
--------------------------------------------------------------------------------------
------------------------------------------

52

Seminar Report On Image Watermarking

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Seminar Report On Image Watermarking

Uploaded by

Copyright:

Available Formats

Seminar Report on

Multipurpose Image Watermarking

[0,L-1], can be represented by

You might also like