Optimized Digital Image Watermarking for Uncorrelated Color Space A Thesis submitted in partial fulfillment of the requirements for the award of the degree of Doctor of Philosophy by Manish Gupta Enrolment No.: 11E7UCPEM4XP900 Supervisor(s): Dr. Rajeev Gupta Dr. Girish Parmar Department of Electronics Engineering Rajasthan Technical University, Kota Rajasthan, India December, 2015
137
Embed
Optimized Digital Image Watermarking for Uncorrelated ... · Optimized Digital Image Watermarking for Uncorrelated Color Space ... Iwouldliketoexpressheremysinceregratitudetomysupervisors,Dr.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Optimized Digital Image Watermarking for
Uncorrelated Color Space
A Thesis
submitted in partial fulfillment of the requirements for the award of
the degree of
Doctor of Philosophy
by
Manish GuptaEnrolment No.: 11E7UCPEM4XP900
Supervisor(s):
Dr. Rajeev Gupta
Dr. Girish Parmar
Department of Electronics Engineering
Rajasthan Technical University, KotaRajasthan, India
communication, medical safety, etc. [10, 11, 12, 13] as discussed in the following
sections and summarized in Table 1.1.
• Copyright Protection: One of the motivations of developing the water-
marking methods is copyright protection. In this application, a copyright
data/ information is embedded into host object without loss of quality [14].
The embedded data prevents other parties from claiming the ownership of
2
Table 1.1: Applications of watermarking methods.
S.No. Application Description
1. Copyright Protection The watermark must be known only to the author and also mostimportantly it must be robust against the various attacks.
2. Covert Communication Since various offices or governments put a ceiling on the useof encryption. In this scenario people may send their secretmessages by using the watermarking method.
3. Copy Control This application restrict the illegally copying of copyrighted ma-terials by embedding a never-copy watermark or limiting thenumber of times of copying.
4. Content Authentication Fragile watermark could be embedded into the host image tocheck the authenticity of the data.
5. Fingerprinting Fingerprinting method used by the owner is to trace the sourceof illegal copies.
6. Broadcast Monitoring Owners of copyrighted programs needs to know about illegalbroadcast, aired by the broadcasters, at the time and locationthat they want according to the contracts terms and conditions.
7. Medical Safety Embedding the date and the patients name in medical imagescould be a useful safety measure.
8. Indexing Indexing of multimedia contents like movies, news items, videomail, images, etc., to help search engines to search those contentsover the internet.
that data. Moreover, the watermark must be known only to the author and
must be robust against the various attacks.
• Covert Communication: Watermarking methods can also be used for
the covert information transmission, as various offices or governments put
a ceiling on the use of encryption. In this scenario people may send their
secret messages by using the watermarking methods.
• Copy Control: This application restricts the illegal copying of copyrighted
materials by embedding a never-copy watermark or limiting the number of
times of copying. For example, today many documents are available on the
internet which could not be saved and printed to control the illegal copying.
• Content Authentication: Fragile watermark could be embedded into the
host image to check the authenticity of the data. A fragile watermark indi-
cates whether the data has been altered and also delivers the information as
to where the data was altered. Therefore, this application does not demand
the robust watermark, since we have to detect the changes only.
• Fingerprinting: Fingerprinting method, used by the owner, is to trace the
source of illegal copies. To achieve this, owner can embed different water-
marks into each copy that distributed to a different customer. For example,
3
unique serial numbers are assigned to customers and used to identify the
customer.
• Broadcast Monitoring: Owners of copyrighted programs needs to know
about illegal broadcast or the commercials, aired by the broadcasters, at
the time and location that they want according to the contracts terms and
conditions. Watermarks can be embed in any type of data to broadcast on
the network by automated systems, which are able to monitor distribution
channels to track the content in the time and the place that they appear.
• Medical Safety: Recently, telemedicine facilitates medical diagnosis by
sending patient medical data/ report over the public network for further
analysis where the modern medical equipments are available. These equip-
ments produce large amount of data every day. Hence, it is necessary to
protect these crucial data. Medical image watermarking is a suitable method
used for enhancing security and authentication of medical data, which is used
for further diagnosis and reference. Embedding the date and the patients
name in medical images could be a useful safety measure.
• Indexing: One of the well-known application of the watermarking is index-
ing of multimedia contents like movies, news items, video mail, images, etc.
In which a comments or any tag/ level is embedded on the contents, so that
these comments or tags are utilized by any search engine to search those
contents over the internet.
The following section listed various requirements and design parameters related
to watermarking methods.
1.3 Requirements and Design Issues of Digital
Watermarking Method
There are various design issues and requirements associated with any watermark-
ing method like transparency, robustness, capacity, security, etc. as summarized
in Table 1.2. The objectives of researchers in the field of watermarking is to maxi-
mize all these parameters for a particular method. Furthermore, these parameters
are mutually dependant on each other as shown in Figure 1.1. Three parameters
namely; transparency, robustness, and capacity are inversely related to each other
i.e. the transparency of a watermarking method increases then its robustness
suffers and vice-versa. This relationship has been depicted in Figure 1.2.
4
Table 1.2: Requirements and design issues of watermarking methods.
S.No. Requirements and DesignIssues
Description
1. Transparency No visual or audio effect should be noticed by the user.
2. Robustness Watermark can be robust against one operation on host dataand may be fragile against another operation.
3. Capacity If the capacity is higher than the better robustness is achievedand at the same time transparency suffers or vice-versa.
4. Security The watermark must resist against the attacks on the hostdata.
5. Complexity The computational cost must be as low as possible to makethe applications real time.
6. Reliability Watermark data embedded into the host, must be recoverablewith the acceptable errors.
Figure 1.1: Mutual dependency between the design parameters.
Figure 1.2: Three main conflicting issues of watermarking.
Hence, the relative importance of these parameters depends on the application-
to-application as listed in the previous section. Moreover, certain applications
demand for more robustness compared to the transparency of the method viz.
copyright protection. Therefore, watermarking method design process involves
trade-off between the conflicting requirements parameters. The most important
requirement for digital watermarking are summarized below:
• Transparency : Transparency or imperceptibility refers to the correlation/
similarity between the watermarked data and the original data. The wa-
5
termark should be invisible. In other words, there is no visual or audio
effect should be noticed by the user. The watermark should not be disgrace
the quality of the host data. However, for a particular application minute
degradation in the host data is permissible to achieve better robustness or
to optimize the cost.
• Robustness: If a watermark can stay alive after common signal processing
operations (such as compression, filtering, translation, rotation operations,
analog-to-digital conversion, scaling, etc.) on host data, then such type of
watermark is called the robust. Moreover, watermark can be robust against
one operation on host data and may be fragile against another operation. For
certain applications, there is a need to embed a robust watermark into the
host data, while some applications demand the fragile watermark. Hence,
it also depends on the application. If the watermark data is embedded in
significant area of a host image then the better robustness is achieved. This
is because those area do not alter so much after common image processing
operations [15]. Contrary to robust watermark, a fragile watermark is not
designed to be robust.
• Capacity : A capacity or payload refers the amount of watermark data that
can be embedded into host. For example, the capacity in case of image
watermarking means the number of bits embedded within the host image.
If the payload is higher than the better robustness is achieved and at the
same time transparency suffers or vice-versa. Therefore, the payload of the
embedded watermark must be in sufficient amount to enable the envisioned
application.
• Security : The watermark must resist against the attacks on the host data.
It must be impossible for an attacker to delete or modify the watermark
without rendering the multimedia data unusable. From this point of view, a
secret watermark key is also used in watermarking, so that it is not possible
to retrieve or even modify the watermark without knowledge of the key.
• Complexity : Depending on the application, the watermark detection is to
be done at different speeds and complexity. For example in broadcast mon-
itoring application, the detection of the watermarking is done in real time.
The computational cost must be as low as possible to make the applications
real time. To keep above facts, the complexity of the watermarking methods
should be low.
6
• Reliability : Watermark data, embedded into the host, must be recoverable
with the acceptable errors.
The performance of watermarking methods for achieving above mentioned re-
quirements are tested after applying various attacks. Therefore, the following
section describes the classification of attacks applicable to image watermarking.
1.4 Taxonomy of Watermarking Attacks
Any procedure that can decrease the performance of watermarking method may be
termed as attack. Testing the robustness and security of a watermarking method
against attacks is as important as the design process. The attacks do not always
remove or destroy the watermark but, also disable its detection. The distortions
done by any attacks degrade the performance of the watermarking method.
In general, different attacks on watermarking can be divided into two classes
namely; unintentional and intentional attacks. To achieve the high reliability
of watermark detection, the watermark detection process has to be robust to the
modifications in the host data caused from both unintentional and intentional
attacks. Unintentional attacks take place using signal processing operations on
watermarked data namely; compression, printing, scanning, filtering, noise, geo-
metric transforms, cropping, etc. For example, multimedia data is generally stored
in lossy compressed format in order to use less storage capacities. These compres-
sion algorithms discard the unimportant parts of data. This distortion may cause
damage of inserted watermark data too. This means that a simple attack is com-
pressing multimedia data in a lossy way. In addition, a rotation or scaling can
change pixel values and destroy the watermark data. Signal processing operations
such as quantization, decompression, re-sampling, and color reduction can dam-
age the watermark. For intentional attacks, a person on purpose can attack on
inserted watermark data in order to copy the multimedia data. In both cases,
any watermarking method should be able to detect and extract the watermark
after attacks. The taxonomy of various intentional and unintentional attacks in
watermarking methods [16, 17, 18], are presented in Table 1.3 and their details
are summarized below:
• Noise: Any random unwanted signal with a given distribution namely; Gaus-
sian, salt & pepper, Poisson, etc., is added to the image unintentionally.
This type of noise may be added during the Analog-to-Digital conversion
and vice-versa, or as a result of transmission errors. However, an attacker
may introduce perceptually shaped noise with the maximum un-noticeable
7
Table 1.3: Taxonomy of watermarking attacks.
S.No. Attack Details
1. Noise Any random unwanted signal with a given distribution namely; Gaus-sian, salt & pepper, Poisson, etc., is added to the image unintention-ally.
2. Filtering Filtering attacks are linear filtering namely; low pass/ mean filtering,Gaussian, and sharpening filtering, etc.
3. Compression If the watermark is required to resist different levels of compression, itis usually advisable to perform the watermark embedding in the samedomain where the compression takes places.
4. Multiple Watermarking The one of the solution of such type of problem is to embedding thetime information by a certification authority.
5. Geometrical Attacks Geometrical attacks distort the watermark through spatial alterationsof the watermarked image. Common geometrical attacks are rotation,scaling, etc.
6. Cropping This is a very common attack which crops the region of interest fromthe watermarked object.
7. Watermark Removaland Interference At-tacks
The objective of such attacks is to forecast or estimate the watermark.
8. Statistical Averaging The objective of such attacks is to recover the host image and/orwatermark data by statistical investigation of multiple marked datasets.
power. This will characteristically force to increase the threshold at which
the correlation detector operates.
• Filtering : Filtering attacks are linear filtering namely; low pass/ mean filter-
ing, Gaussian, and sharpening filtering, etc. Mean or average filtering does
not introduce considerable degradation in watermarked images but, can dra-
matically affect the performance. Therefore, to design a watermark, robust
to a known group of filters that might be applied to the watermarked image,
the watermark data should be designed in such a way that it have most of
its energy in the frequencies which filter transfer functions changes the least.
• Compression: Compression belongs to an unintentional attack class, which
appears very often in various multimedia applications. In practice, most of
the images, audio, and video are being transmitted/ distributed via internet
after the compression in order to reduce the time and data usage. If the
watermark is required to resist different levels of compression, it is usually
advisable to perform the watermark embedding in the same domain where
the compression takes places.
• Multiple Watermarking : An attacker may watermark an already water-
marked data and later claims of ownership. One solution to such type of
8
problem is to embed the time information by a certification authority.
• Geometrical Attacks : Geometrical attacks do not pretend to remove the wa-
termark by itself, but to distort it through spatial alterations of the water-
marked image. With such attacks watermarking detector loses the synchro-
nization with the embedded information. These attacks can be subdivided
into attacks applying general affine transformations and attacks based on
projective transformation. Common geometrical attacks are rotation, scal-
ing, change of aspect ratio, translation and shearing, etc.
• Cropping : This is a very common attack since in many cases the attacker is
interested in a small portion of the watermarked object, such as parts of a
certain picture or frames of video sequence. With this in mind, in order to
survive, the watermark needs to be spread over the dimensions where this
attack takes place.
• Watermark Removal and Interference Attacks: The objective of such attacks
is to forecast or estimate the watermark and then use the predicted water-
mark either to eradicate watermark or to damage its unique extraction at the
destination side. Some known efficient removal attacks are; the median wa-
termark prediction followed by subtraction [19], the Wiener prediction and
subtraction [18] and perceptual re-modulation [20], which combines both
removal and interference attacks.
• Statistical Averaging : The objective of such attacks is to recover the host
image and/or watermark data by statistical investigation of multiple marked
data sets. An attacker may attempt to predict the watermark and then to
remove the watermark by subtracting the estimate. This is very hazardous
if the watermark does not rely significantly on data. That is why the percep-
tual masks are used to create a watermark. Averaging or smoothing attack
is belonging to this class of attack. Averaging attack consists of averaging
many instances of a given data set (e.g. N) each time marked with a different
watermark. In this pattern a prediction of host data is calculated and each
of the watermarks is weakened by a factor N.
In the above section, various classification of watermarking attacks have been
presented. The following section gives the brief details of classification of the
different watermarking methods.
9
1.5 Classification of Watermarking Methods
Mohanty [21] presented a state-of-the-art categorization of digital watermarking
methods. Digital watermarking methods may be categorized on the basis of host
multimedia data, human perception, embedding domain, robustness, data extrac-
tion, application area, etc. [13]. Figure 1.3 shows the classifications of digital
watermarking methods. There are different way of classification of digital water-
Figure 1.3: Classification of watermarking methods.
mark methods [22].
• First, watermarking methods may be divided into four groups according to
the type of host multimedia data to be watermarked;
– Image watermarking method
– Audio watermarking method
– Video watermarking method
– 3-dimension (3-D) watermarking method
• Second, watermarking methods may be grouped on the basis of the data for
extraction;
– Private or non-blind watermarking method : This class of method re-
quired the original data (watermark or/ and host) during the extraction
of watermark from the watermarked data.
10
– Semi-private or semi-blind watermarking method : This group of method
needs extra information other than the original data during the detec-
tion.
– Public or blind watermarking method : In public or blind watermarking
method the detection of watermark from the watermarked data needs
only the watermark.
• Third, watermarking methods may be categorized on the basis of human
perception;
– Visible watermarking method : If the watermark data is noticeable to
the user, then such class of watermarking is known as visible water-
marking method. Examples of visible watermarks are logos that are
used in papers and video.
– In-visible watermarking method : If the watermark data is imperceptible
to the user, then such class of watermarking is known as invisible wa-
termarking method. For example, images distribute over the internet
and watermarked invisible for copy protection.
• Fourth, watermarking methods may also be classified on the basis of robust-
ness;
– Fragile: A fragile watermark will be changed if the host data is modi-
fied.
– Semi-fragile: Semi-fragile watermark is sensitive to some degree of the
change to a watermarked image.
– Robust : Watermark in robust method cannot be removed by common
signal processing operations.
• Fifth, watermarking methods may also be classified on the basis of embed-
ding watermark data;
– Text : If the embed watermark data is in the nature of text.
– Image format : If the embedded watermark belongs to any image shape
like logo, binary image, gray scale image, color image or logo, stamp,
etc.
– Noise sequence: In this class of method the embedding watermark is
in terms of random noise sequence like pseudo random or Gaussian
random sequence etc.
11
• Sixth, watermarking methods may be categorized on the basis of application
of method;
– Source-based watermarking method : In source based, all copies of a
particular data have a unique watermark, which identifies the owner of
that data.
– Destination-based watermarking method : In this method, each dis-
tributed copy is embedded using a unique watermark data, which iden-
tifies a particular destination.
• Finally, watermarking methods may be classified into two major classes ac-
mid Transform (SPT), etc. Each of these transform method has its own specific
characteristics and representation of an image.
Digital image watermarking is a process of imperceptibly hiding a watermark
(in the form of signature, random sequence, or some image) into an image (host
or cover) which may be used to verify the genuineness of its owner. The resultant
image of this process is termed as watermarked image. The watermarking meth-
ods can be performed either in spatial domain or in the transform domain. In
spatial domain-based watermarking method, watermark may be embedded within
an image by modifying the pixel values [9] or the Least Significant Bit (LSB) val-
ues. While, in transform domain-based watermarking method, watermark may be
embedded by modifying the transform domain coefficients. However, more robust
watermark could be embedded in the transform domain of images by modifying
the transform domain coefficients as compared with the spatial domain-based im-
age watermarking method. In the following section, the state-of-the-art review of
digital image watermarking methods have been presented based on spatial domain
and transform domain [1, 2, 3, 22, 24, 25, 26].
1.6.1 Spatial Domain-based Methods
A watermarking method based on the spatial domain approach, hides watermark
data in the pixel values of the host image. Such class of methods make minor
changes in the intensity of pixel value of host image [1, 27, 28, 29, 30, 31, 32, 33, 34].
One of the most common examples of this method is to embed the watermark in
the LSB’s of image pixels [28, 30, 35]. In other words, significant portions of
low frequency components of images should be modified in order to insert the
13
watermark data in a reliable and robust way. As another example, an image is
divided into the same size of blocks and a certain watermark data is added with
the sub-blocks [28]. The imperceptibility of the watermark data is achieved on the
postulation that the LSB bits are visually insignificant. Although, spatial domain-
based watermarking method can be easily implemented and very fast, they have
the many disadvantages. These methods are highly susceptible to common signal
processing operations and can be easily impaired and tempered. For example,
lossy compression could completely crush the watermark data. In summary, wa-
termarking method based on spatial is very easy to destroy using some attacks like
low-pass filtering, additive noise, etc. In other words, the spatial domain-based
image watermarking methods are not robust against the common signal process-
ing operation on the host image. The brief summary of spatial domain-based
methods presented in Table 1.4. Researchers [15] suggested that the transform
domain-based image watermarking methods are more robust as compared to spa-
tial domain-based image watermarking methods against the various watermarking
attacks as mentioned in the Section 1.4. Therefore, further in this thesis our focus
is only on the transform domain-based image watermarking methods.
Table 1.4: Categories of spatial domain-based image watermarking method.
S.No. Category Property Description
1. Pixel-based[27, 28, 33]
Such class of methods makes mi-nor changes in the intensity ofpixel value of host image to em-beds a watermark. One of themost common example of thismethod is to embed the water-mark in the LSB’s of image pix-els.
Spatial domain-based water-marking method can be easilyimplemented and very fast.
2. Block-based[28, 29, 30]
In this class of watermarkingmethod an image is divided intothe same size of blocks and acertain watermark data is addedwith the sub-blocks.
These methods are highly sus-ceptible to common signal pro-cessing operations and canbe easily impaired and tem-pered. For example, compres-sion could completely crush thewatermark data.
1.6.2 Transform Domain-based Methods
The transform of an image is just another form of representation. It does not
change the content present in the image. Transform domain-based image wa-
termarking methods have many advantages over spatial domain-based methods
[15]. As presented in literature, transformed domain-based image watermarking
methods are more robust against the various watermarking attacks and signal
processing operations because the transform domain does not utilize the original
14
host image for casting the watermark data. In addition, the transform domain-
based image watermarking distributes the watermark data over all part of the
host image. Moreover, transform domain-based methods are capable enough to
embed more watermark bits into the host image and are more robust to attack.
However, they are difficult to implement and are computationally more expensive
as compared with the methods given in Section 1.6.1. In literature, various re-
versible transform methods namely; DFT, DCT, DWT, SPT, etc. are used by the
researcher to improve the robustness of the image watermarking methods. This
class of watermarking methods insert the watermark data into the host image by
Figure 1.4 shows the general block diagram of transform domain-based digital wa-
termarking system. Summary of the transform domain-based methods are given
in Tables 1.5. A detailed sate-of-the-art survey of transform domain-based image
watermarking methods has been presented by Potdar et al. [3].
Discrete Fourier Transform (DFT)-based Method
Ruanaidh et al. [49] presented a DFT-based image watermarking method in which
the watermark data is inserted into the host image by manipulating the phase
information. Wolfgang et al. [30] later concluded in their work that image water-
marking using the phase manipulation is robust against image contrast operation.
Further, Ruanaidh and Pun [50] presented a method of image watermarking using
the Fourier transform and concluded that method is robust against the geometric
attacks. Lin at al. [51] presented a novel image watermarking method which is ro-
bust against the rotation, scaling, and translation attacks. However, this method
is not robust enough against the cropping and compression attacks. In litera-
ture, few methods are available where the watermark is casted by modifying the
mid frequency band of DFT magnitude component [52, 53]. They concluded that
the proposed methods are robust against the Joint Photographic Expert Group
(JPEG) and Set Partitioning In Hierarchical Trees (SPIHT) compression attacks.
Moreover, Solachidis and Pitas [54] presented a novel method of image watermark-
ing in which a circularly symmetric watermark is embedded in the DFT domain.
In addition, the proposed method is robust against the geometric rotation attacks
because the watermark is circular in shape with its center at image center.
15
(a) Embedder (b) Extraction
Figure 1.4: General block diagram of transform domain-based imagewatermarking system.
Discrete Cosine Transform (DCT)-based Method
Transform domain-based image watermarking methods possess a number of de-
sirable properties as compared to spatial domain-based methods. Moreover, these
methods make it difficult for any intruder or unauthorized user to read or change
the watermark data. Since in the transform domain, embedded watermark is dis-
tributed over the area of the image after the inverse transformation. The DCT
domain-based method is divided into two groups namely; global-based and block-
based DCT image watermarking methods. The image watermarking methods that
rely on the global DCT approach, spread the watermark over the entire image.
On the other hand, block-based approach embeds the watermark as follows:
1. Divide the host image into non-overlapping blocks of 8× 8.
2. Take the DCT to each of the block as mentioned in step 1.
3. Choose a specific block for watermark embedding by using certain criteria
like Human Visual System (HVS).
4. Choose coefficients for watermark embedding by using certain selection cri-
teria like highest or lowest magnitude.
5. Cast the watermark data by modifying the selected coefficients; and
16
Table 1.5: Categories of transform domain-based image watermarking method.
S.No. Category Description
1. DFT-based Robust against image contrast operation. [49]
Robust against the geometric attacks. [50]
Robust against the rotation, scaling, and translation attacks.However, this method is not robust enough against the crop-ping and compression attacks. [51]
Watermark embedded by modifying the mid frequency band ofDFT coefficients. Furthermore this method is robust againstJPEG and SPIHT compression attacks. [52], [53]
Circularly symmetric watermark is embed in the DFT domainand concluded that method is robust against the geometricrotation attacks. [54]
2. DCT-based Global DCT approach by exploiting the HVS and shows thatthe method is robust against the geometric attacks like rota-tion, scaling, etc. [8], [15]
Block-based DCT method. DCT-based methods missing thetime and frequency information at the same time. [55], [56]
3. DWT-based Based on principle of “toral automorphism”. [31]
Known as “cocktail watermarking”and concluded that themethod is robust against the all possible watermarking attacks.[57]
Presented a novel method for any size of images, which hideswatermark into the high-frequency sub-bands of DWT coeffi-cients. [58]
Uses Daubechies-2 filter bank for transformation of host imageand show that the proposed method is robust to geometric,filtering, and StirMark attacks. [59]
Uses Symlet-8 filter bank and shows that the method is robustagainst various attacks. [60]
Uses Har wavelet for host image transformation. [61]
Uses Symlet-4 filter bank. [42]
4. SPT-based Robust against the various geometric attacks in comparisonwith DWT-based method. [46]
Hybrid watermarking method using the SPT and SVD, thismethod have good visual quality and resistance against severalattacks. [62]
Robust against the common signal processing and geometricattacks like rotation. [63]
6. Take inverse DCT (IDCT) transform on each block.
In the DCT-based image watermarking method, most of the research is dedicated
to design the specific criteria for selecting the particular block and coefficients.
Cox et al. [15] proposed a robust image watermarking method using the global
DCT approach, which embeds the imperceptible watermark data into the host
image by exploiting the HVS. Koch et al. [8] reported a method for watermark
embedding having following steps:
1. Divide the host image into non-overlapping blocks of 8× 8.
2. Take the DCT to each of the block as mentioned in step 1.
17
3. Choose the specific block by using the pseudo-random subset criteria.
4. A triplet of frequencies is selected from 1 of 18 predetermined triplets.
5. Cast the watermark data by modifying the selected coefficients, so that their
relative strengths encode a 1 or 0 value.
6. Take inverse DCT transform on each block.
In literature, various image watermarking methods have been proposed by using
DCT [55, 56, 64, 65, 66, 67, 68, 69]. Out of the existing DCT-based watermarking
methods, the block-based DCT method is widely used by researchers in the area of
watermarking. Lin et al. [69] find that the DCT-based methods are robust against
JPEG compression, but as robustness increases the quality of watermarked image
decreases. Moreover, DCT-based methods are not robust against the geometric
attacks like rotation, scaling, etc.
Discrete Wavelet Transform (DWT)-based Method
To achieve the robustness of image watermarking methods, discrete wavelet trans-
form utilizes the spatial and frequency information of the transform data in multi-
ple resolution. Recently, many image watermarking methods have been reported
which exploit the advantage of DWT over the DFT and DCT [40, 41, 42, 58, 59,
60, 70, 71, 72, 73, 74]. However, the performance of DWT-based method can be
further enhanced by exploiting the characteristics of HVS during the watermark
embedding stage. If a watermarking method can utilize the characteristics of the
HVS, then it is possible to embed watermark with more energy in a host image,
which makes watermark more robust. Although, the HVS model enhances the
imperceptibility and robustness of watermarking method, it suffers with the com-
putational cost and complexity point of view. According to the HVS, the human
eye is less sensitive to noise in high resolution DWT bands having an orientation
of 450. From this point of view, the DWT is a very useful transform as compared
to DFT and DCT, since it can be used as a computationally efficient version of
the frequency model for the HVS [70]. One of the reasons for the popularity of
DWT-based image watermarking method is that various multimedia standards
like JPEG2000, MPEG-4, etc. are based on the DWT. Hence, DWT decompo-
sition can be exploited to make a real-time watermark application. DWT-based
image watermarking methods are capable to embed a fairly good quality of wa-
termark and it can recover the watermark from watermarked image effectively.
The quality and robustness of DWT-based methods depend on the selection of
18
particular filter bank and decomposition level [61]. Voyatzis and Pitas [31] devel-
oped a robust method on the principle of “toral automorphism”, in which they
embed the binary logo watermark. Lu et al. [57] presented a method in which
they embedded dual watermarks with complement to each other. This method
is popularly known as “cocktail watermarking”. Furthermore, the result shows
that the proposed method is robust against all possible watermarking attacks.
Zhao et al. [75] reported a dual domain-based method for image authentica-
tion. In their method, authors utilized the DCT for watermark generation and
DWT for watermark casting. Agreste and Andaloro [58] presented a novel DWT-
based watermarking method for any size of image which hides watermark into the
high-frequency sub-bands of DWT coefficient of host image. Later, Agreste and
Andaloro [59] reported another method which was the modified version of their
previous method by changing the filer bank by Daubechies-2, and concluded that
it is more robust to geometric, filtering, and StirMark attacks. Ghouti et al. [41]
selected balanced multi-wavelets filter bank for the data hiding and found that the
method is more robust against the various watermarking attacks. Furthermore,
Khelifi et al. [60] utilized the advantage of symlet-8 filter bank in their method.
Vahedi et al. [61] uses the Haar wavelet for message hiding. Moreover, Vahedi
et al. [42] exploited the advantage of symlet-4 filter bank to increase the quality
and robustness of watermarking method as compared to existing methods. In his
paper, they proposed a novel DWT-based method for color images by embedding
the binary watermark. For embedding the watermark, DWT-based methods use
three or higher level decompositions [39, 40, 42].
Steerable Pyramid Transform (SPT)-based Method
During the recent years, various image watermarking methods have been devel-
oped by using the DWT because it possess a number of desirable properties as
compared with the other transform-based methods. However, there are still rooms
for improvements in the field of watermarking. Although, DWT-based image wa-
termarking methods have various advantages, it suffers in terms of recording the
directional information which is very important components for any digital image
processing operations [44, 47]. Therefore, scholars in this area looking for an-
other reversible transform which possess all the properties of DWT. As a result,
researchers in this area propose various scale and directional image illustrations
during the years and results show that out of other representation, SPT having all
the advantages of DWT. Further, SPT is also capable in capturing the directional
information. Moreover, the results show that the SPT-based image watermarking
methods are more robust against the various geometric attacks in comparison with
19
DWT-based method [46]. Literature shows that the SPT keeps most of the ad-
vantages of DWT as its basis functions are confined to a small area in both space
and spatial-frequency. However, this recursive multi-scale & multi-directional de-
composition improve the drawbacks of DWT like; it is aliasing free and capable to
generate any number of orientation bands as it is based on a category of random
orientation filters produced by linear grouping of a set of basis filters [46, 76].
Invariance, multi-resolution, and capture of multi-scale and multi-resolution con-
structions in the images are some of main properties of SPT which make it superior
in watermarking methods. Moreover, researchers concluded on the basis of the re-
sults that SPT-based methods of image watermarking are more robust against
the common signal processing and geometric attacks [45, 46, 47, 63]. There are
lot of scope of the research using SPT-based watermarking method because very
few SPT-based image watermarking methods [46, 62] have been reported till date.
Drira et al. [46] developed a SPT-based method and shows that it is resistant
to JPEG compression, additive noise, and median filtering. Hossaini et al. [62]
presented a novel hybrid watermarking method using the SPT and singular value
decomposition (SVD) and concluded that proposed method has good visual qual-
ity and resistance against several attacks.
Following section reviews the various optimization methods used to improve
the performance of image watermarking along with their constraints.
1.6.3 Optimization Methods in Image Watermarking
Previous section presented a state-of-the-art survey of different image watermark-
ing methods using various domain like spatial and transform domain. Literature
survey reveals that the transform domain methods are more robust in compari-
son with spatial domain. In any watermarking method the aim is to maximize
the various parameters like transparency, robustness, capacity, etc. as given in
Section 1.3. Furthermore, the above mentioned parameters are the function of
strength factor of watermark to be added into the host image. However, all these
parameters are inversely related to each other. Therefore, to enhance the per-
formance of the method, researchers must select the optimum value of strength
factor of watermark, so that all above mentioned parameters are maximized simul-
taneously. From this point of view, there is a room for the optimization methods
in the watermarking area to enhance the performance by optimally selecting the
value of strength factor. There are various optimization methods namely; genetic
swarm optimization (PSO), etc. which have been used to increase the quality and
20
robustness of watermarking methods [77]. During the recent years, many water-
marking methods have exploited the advantages of various available optimization
methods [42, 55, 71, 78, 79, 80, 81, 82, 83, 84, 85]. The detailed survey of different
optimization methods has been presented by Karaboga and Akay [82] and Darwish
and Abraham [13]. A brief summary of the optimization methods used in image
watermarking methods is presented in Table 1.6.
Table 1.6: Categories of optimization methods in image watermarking.
S.No. Category Details
1. Genetic Algorithm Optimize 24 strength factor of watermark. [42]
Uses GA for search the optimum values of parameters. [81, 86]
Use GA to embed two bits of watermark data within each pixelof host image. [87]
2. Differential EvolutionMethod
Scale factors optimized using DE. [88]
Used a modified DE (self-adaptive differential evolution) algo-rithm for optimizing the scaling factors. [85]
Applied a DE optimization method, to search optimal scalingfactors. [89]
Use the DE to optimally design the quantization steps (QSs) forcalculating the strength of the watermark for achieving goodrobustness and quality. [90]
3. Artificial Bee ColonyMethod
Uses the ABC method to optimize pixel by pixel embeddingat different frequency sub-band with DWT, to improves theperformance of proposed image watermarking method. [91]
Genetic Algorithm
Many researchers used GA in their method for optimizing the watermark strength
to enhance the overall performance. GA have been used by Vahedi et al. [42, 80] to
compute the optimum strength of watermark and result shows that it improves the
robustness of existing methods. Kumsawat et al. [81, 86] used GA in their methods
for searching the optimum values of parameters and the embedding strength to
improve the transparency and robustness of proposed method. Mohammed et
al. [87] proposed a novel method in which they used GA to embed two bits of
watermark data within each pixel of host image.
Differential Evolution Method
A detailed survey of DE method has been presented by Das et al. [92]. Vester-
strom and Thomsen [93] compared the performance of DE with PSO, and other
21
evolutionary algorithms (EAs), and reported that DE outperforms other consid-
ered methods. Aslantas [88] reported a method in which the singular values (SVs)
of the host image are manipulated to embed the watermark by using multiple
scaling factors. Further, scale factors are optimized using DE to enhance the
performance of proposed method. Later, Aslantas [94] exploits the DE method
for optimizing the scaling factor parameters to achieve maximum robustness and
transparency. Ali and Ahn [85] used a modified DE (self-adaptive differential evo-
lution) algorithm for optimizing the scaling factors of watermark data to achieve
better robustness and quality in DWT-SVD based watermarking method. Fur-
ther, Ali et al. [89] applied a DE optimization method to search optimal scaling
factors to improve the quality of watermarked image and robustness of the wa-
termark. Lei et al. [90] proposed a method in which they used DE to optimally
design the quantization steps (QSs) for calculating the strength of the watermark
for achieving good robustness and quality.
Artificial Bee Colony Method
Artificial Bee Colony (ABC) optimization method is one of the most recently de-
veloped swarm-based algorithms. A very few watermarking methods have been
reported which utilizes the powerfulness of ABC for optimization in their methods
till date. Karaboga and Akay [82] compared ABC optimization method with GA,
PSO, etc. and concluded that the performance of ABC is better than other opti-
mization methods. Recently, Akay and Karaboga [95] presented a state-of-the-art
survey and applications in the field of image processing. Sha et al. [91] uses the
ABC method to optimize pixel by pixel embedding at different frequency sub-band
with DWT to improve the performance of proposed image watermarking method.
Following section reviews the various color spaces used in image watermarking
method in order to enhance the performance parameters along with their con-
straints.
1.6.4 Color Spaces in Image Watermarking Method
At the outset, early image watermarking methods embed the watermark message
within the gray scale or color host image in the form of bits or bit stream and
further these bit streams are replaced by some pictorial shape representations
[39, 40]. In various multimedia applications namely; MPEG-1, MPEG-2, and
other MPEGs, the color images are the basic component. Hence, it is vital to
develop a watermarking method for color images. Although digital images are
22
available in color format, most of the watermarking methods embed the gray
scale or binary image watermark while very few work has been reported for color
watermark [96, 97, 98]. In color image processing, we cannot ignore the effects of
different color spaces used on the performance of the method. Furthermore, Vahedi
et al. [99] demonstrated the effects of different color spaces on the performance
of an image watermarking method. The color spaces are dividing into two classes
namely; correlated and uncorrelated color space. The summary of color spaces
used in image watermarking is given in Table 1.7.
Table 1.7: Categories of color spaces in image watermarking method.
S.No. Category Comments
1. Correlated ColorSpaces
Uses HSV, RGB, YUV and HSI color models respectivelyin image watermarking method. [34, 58, 100, 101]
They uses HSI color space in DWT-based watermarkingmethod. [42]
2. Uncorrelated ColorSpaces
Uncorrelated color models Lab and Lαβ are used.[102]
Embeds the color watermark in Lab color space. [96]
Correlated Color Spaces
In a correlated color spaces, a color image is decomposed into three semi-independent
images in which change in one component may affect the other two components
of the images. In literature, various correlated color spaces namely; RGB, Y CbCr,
YIQ, HSI, HSV, etc. are being used in image watermarking methods by the re-
searchers. Methods proposed by researcher in [34, 58, 100, 101] uses HSV, RGB,
YUV, and HSI color models respectively. Vahedi et al. [99] reported a work
dedicated to the performance of DWT-based watermarking method on different
color spaces like RGB, Y CbCr, YIQ, HSI, and HSV. In which they concluded
that HSI color space outperforms other considered spaces. Golea et al. [97] pre-
sented the SVD-based RGB color image watermarking for embedding the color
RGB watermark into the RGB host image. Recently, Su et al. [98] presented QR
decomposition method for embedding the RGB color watermark into RGB host
image. Vahedi et al. [42] presented a DWT-based watermarking method in which
they used HSI color space to enhance the robustness and transparency.
Uncorrelated Color Spaces
Most of the image watermarking methods reported so far have used the correlated
color spaces. However, correlated color spaces impose the constraints to use only
23
one color channel at a time for casting the watermark data [103]. As a consequence,
researchers in this area proposed various color co-ordinate illustrations during the
recent years to removes the dependency of three decomposed images. Such class
of color space is called as uncorrelated color space. There exist some uncorrelated
color models such as Lab, Lαβ, uncorrelated color space (UCS), etc. which may
be used in color image watermarking to increase the robustness and quality by
using all the color image components of host and watermark images [43, 102,
104]. There is a lot of scope of the research using uncorrelated color spaces-
based watermarking method because very few work has been reported till date.
Chou and Wu [96] embedded the color watermark in Lab color space using less
computationally complex spatial-domain color image watermarking method.
1.7 Challenges in Image Watermarking
Though many image watermarking methods have been proposed and demon-
strated significant contribution, there are still some challenges which need to be
addressed. One of the main challenges of the watermarking problem is to achieve a
better trade-off between robustness, transparency, capacity, and security. In order
to address above mentioned issue (i.e. the trade-off) to achieve a better perfor-
mance, many researchers presented solutions for this issue in their work. However,
improvements are required to fulfill the expectation of the industry. This section
overviews some of the crucial challenges of image watermarking.
1.7.1 Use of Color Watermark
Most of the work during the past decade in this area have been reported for
protection of the color or gray scale image (host) by embedding the gray scale
or binary image watermark. To embed a binary or gray scale image, one has to
convert it from color image because in nature color images are available. However,
very few methods have been reported which embed color watermark for protection
of images [96, 97, 98]. From this point of view, there is still a lot of scope for
improvements in the field of image watermarking to embed a color watermark
into the color host image.
1.7.2 Color Spaces
In color image watermarking, we are dealing with color images for host and water-
mark. To read a color image there are various color spaces/ coordinates available
in the literature as given in Section 1.6.4. Therefore, researchers are use different
24
color spaces to improve the performance of the method. Robustness and trans-
parency of any method depend on how an input host is decomposed into three
color channels and also the decomposed images are independent with each other
or not. In other words, performance parameters of a color image watermarking
depend on the color space used in the method [99]. To consider above facts, there
is a lot scope to improve the performance of method by using a suitable color
space which decompose an image into three independent images. Fortunately, a
class of color spaces known as uncorrelated color space, generate three indepen-
dent images. However, a very few method has been reported which exploits this
class of color coordinate.
1.7.3 Transform Method
Though various reversible transform methods, as given in Section 1.6.2, have been
used in the reported image watermarking methods, still there is scope for improve-
ments of image watermarking methods by using some newly introduced transforms
(like SPT and other) which show better performance as compare to existing meth-
ods.
1.7.4 Optimization Method
A better trade-off is required between the parameters given in Section 1.3 to en-
hance the performance of the watermarking method. Furthermore, better trade-off
between the parameters depend on how optimally select the embedding param-
eters (like strength factor of watermark). To achieve this objective, there are
various optimization methods which have been used by the researchers as given
in Section 1.6.3. From this point of view, there is a room for improvements in
the performance of image watermarking method by using some newly developed
optimization methods like ABC, DE, etc.
1.7.5 3-D Watermarking
A recent challenge in watermarking is for protection of the 3-D objects or models
against the illegally utilizations which was introduced by Ohbuchi [105]. This
type watermarking is called “3-D watermarking”. Recently, 3-D watermarking
have been widely used in virtual reality, medical imaging, video games, computer
aided design, etc. This is considered as a new kind of multimedia that has scored
an increasing success. A very few work has been reported to protect the 3-D
models against the illegal utilizations till date, though it has been widely utilized
25
in the entertainment industry.
1.8 Scope of the Thesis
The challenges discussed above have not been fully resolved. Therefore, there is
a need to design and develop a robust image watermarking method. The main
contributions in this thesis are five-fold. It first aims to pre-process the input color
images (host and watermark) by transforming it into UCS color space. Second,
four efficient transform-based image watermarking methods for color images have
been proposed to enhance the transparency and robustness of method. Third,
three optimization methods have been used to improve the performance of pro-
posed methods by optimally selecting the strength factors of watermarks and then
post-processed the watermarked coefficients to reconstruct the watermarked im-
age. Fourth, applying various watermarking attacks in order to test the proposed
method on various benchmark/ validation parameters like composite-peak-signal-
to-noise ratio (CPSNR), structural similarity (SSIM), and normalized correlation
(NC). Finally, an image watermarking method is designed and developed using
the proposed methods for protection of color images.
Including this introductory chapter, the rest of the thesis is organized in the
following five chapters.
Chapter 2 discusses literature survey of the existing methods for image water-
marking using DWT which embed gray-scale watermark image for the protection
of color host images. It then briefly describes the various preliminaries like DWT
method, UCS, and GA followed by the mathematical concept of pre-processing of
input images, embedding, post-processing, and extraction of watermark images.
Moreover, GA has been used to optimize the watermark strength during the em-
bedding phase. Further, the testing of the proposed method is done against the
various watermarking attacks. The results of the proposed method are compared
with existing methods and finally discussion is presented.
Chapter 3 reviews of various available SPT-based image watermarking meth-
ods which used gray-scale watermark image for the protection of color host im-
ages. Then, brief descriptions of various preface such as SPT method, UCS, and
GA are presented followed by the mathematical concept of pre-processing of in-
put images, embedding, post-processing, and extraction watermark images. The
proposed method also uses GA to optimize the watermark strength during the em-
bedding phase. The results of the proposed method are compared with existing
methods and the method proposed in Chapter 2.
Chapter 4 deals with the protection of color images by embedding the color
26
watermark by using the DWT and ABC methods. This chapter presents a survey
of the existing methods for color image watermarking using DWT followed by
preliminaries. It then describes the pre-processing of input images, embedding,
post-processing, and extraction of watermark images. Then, three optimization
methods have been used to optimize the watermark strength during the embedding
phase followed by the testing of the method. The results of the proposed method
are compared with existing methods.
Chapter 5 proposes a color image watermarking method by casting the color
image watermark into the color host image by using the SPT and DE. A brief
discussion about the SPT and DE methods have been presented followed by the
procedure for embedding and extraction phase. In the proposed method, DE
has been used to optimize the watermark strength during the embedding phase
followed by the testing under various attacks. The results of the proposed method
are compared with existing methods and proposed method in Chapter 4.
The last chapter summarizes the key findings, main contributions of the
thesis, and possible scope for future research in this area.
27
Chapter 2
Digital Image Watermarking using Dis-
crete Wavelet Transform on Gray-Scale
Watermark Image
2.1 Introduction
Two embedding-domain have been used for digital image watermarking methods,
namely; spatial and transform domain as mentioned in Section 1.3. The detailed
survey related to the image watermarking methods is given in the Section 1.6. An
image watermarking method based on the transform domain is more robust than
the spatial domain [15]. Researchers suggested that DWT-based methods of image
watermarking are more robust against the common signal processing and malicious
attacks [51, 106, 107] as compared with the DFT and DCT-based methods. DWT-
based methods are capable in embedding a better quality of watermark and also
can recovered it from the watermarked image effectively. Therefore, this chapter
focuses on the protection of color images from its illegal utilization using a DWT-
based digital image watermarking method.
During past decade, many image watermarking methods [40, 41, 42, 58, 59,
60, 70, 71, 72, 73] have been reported which utilized the advantages of DWT over
the DFT and DCT [74]. DWT-based methods are more robust because it is more
close to the frequency model for the HVS [70]. The characteristics of HVS model
are used by various researcher in image watermarking to enhance the transparency
and robustness. Kundur et al. [39] exploit HVS model to generate a visual cover
for multi-resolution-based image watermarking method. Later, Reddy et al. [40]
and Ghouti et al. [41] uses the advantages of HVS in their methods. Furthermore,
the performance of the DWT-based methods depend on the parameters namely;
selection of filter bank, decomposition level, and selection of embedding decom-
posed coefficients [42]. Therefore, all the work reported in this area using DWT
are based on the changing the above mentioned parameters to show the better
robustness and transparency of watermarking method. Vahedi et al. [42] exploit
the advantage of symlet-4 filter bank to increase the quality and robustness of wa-
termarking method as compared to existing methods. In his paper, they proposed
a novel DWT-based method for color images by embedding the binary water-
mark. Vahedi et al. [42] showed that three level decomposition with all the four
sub-spaces namely approximation, horizontal, vertical, and diagonal along with
symlet-4 filter bank provides better results for image watermarking. Therefore, in
this method DWT has been used with three level of decomposition and symlet-4
filter bank for embedding the watermark. Moreover, to increase the quality and
robustness of watermarking methods, this method utilizes the capabilities of GA
to optimize the watermark strength.
Generally, the researchers used RGB, Y CbCr, YIQ, HSI, HSV, etc., color space
models for host image in their digital watermarking method. The above mentioned
color models are correlated i.e. the image components are not independent and
change in one component may affect the other components of the image. This
imposes the constraints for the researchers who used correlated color host image
to use only one color component at a time for embedding the watermark data.
However, there exist some uncorrelated color models such as Lab, Lαβ, UCS,
etc. [102, 108], which may be used in color image watermarking to increase the
robustness and quality by using all the color image components of host images.
Saraswat and Arya [108] used UCS for color transfer of images and observed that
UCS outperforms the other uncorrelated color spaces.
Therefore, due to the limitations of correlated color spaces and powerfulness of
DWT, this chapter proposes a DWT-based image watermarking method using un-
correlated color space (UCS). Further, GA has been used to optimize the strength
factor of the proposed watermarking method. In this chapter Section 2.2 presents
the various preliminaries used. Proposed method and performance improvements
using GA to the proposed method have been discussed in Section 2.3. Further,
experimental results and the conclusion of the chapter are presented in Section
2.4 and 2.5 respectively.
2.2 Preliminaries
In this chapter, the proposed method uses DWT, UCS and GA methods for digital
image watermarking. Therefore, following section describe the functions of these
methods in brief:
30
2.2.1 Discrete Wavelet Transform (DWT)
DWT is frequently used in various image processing applications namely; compres-
sion, watermarking, etc. DWT is a sampled version of continuous wavelet trans-
form. The main advantage of DWT is that it maintains both frequency and time
information at the same time which was missing in DFT. The transform based on
small sinusoidal waves of varying frequency and limited duration is called wavelet.
DWT is used to decompose the input image into sub-images of diverse spatial
direction (i.e. horizontal, vertical, and diagonal) and independent frequency area
[40, 41]. Transformation of an image from the spatial domain to DWT domain
by one level decomposes the input image into four different frequency bands in
which one is the low frequency and remaining three are the high frequency bands.
These bands are represented as LL (approximation detail of image), HL (hori-
zontal detail of image), LH (vertical detail of image), and HH (diagonal detail of
image) respectively. Magnitude of DWT coefficients is larger in the lowest bands
(LL) at each level of decomposition and is smaller for other bands (HH, LH, and
HL). Therefore, after one level decomposition, the further decomposition of given
image is done using only LL sub-space which is also decomposed into four distinct
frequency bands as mention above. Figure 2.1 shows the two dimensional image
of size 512×512 before and after the three level of DWT decomposition with their
sub-spaces size. Since, a two dimensional image has been used, it needed a 2-D
wavelet transform. The 2-D DWT is implemented as a 1-D row transform followed
by a 1-D column transform. The 2-DWT transform coefficients for input image
function f(n1, n2) of size N1 ×N2 are calculated using Eq. (2.1) and (2.2).
Wϕ(j0, k1, k2) =1√N1N2
N1−1∑n1=0
N2−1∑n2=0
f(n1, n2)ϕj0,k1,k2(n1, n2) (2.1)
W iψ(j0, k1, k2) =
1√N1N2
N1−1∑n1=0
N2−1∑n2=0
f(n1, n2)ψij0,k1,k2
(n1, n2) (2.2)
here, j0 is the starting scale and i = {H, V,D} indicates the directional index of
wavelet function. Eq. (2.1) calculates the approximation coefficient while (2.2)
calculates the other detail coefficients. 2-D scaling function ϕ and wavelet function
ψi, used in Eq. (2.1) and (2.2), may be calculated through the separable 1-D filter
having the impulse response hϕ(−n) and hφ(−n) respectively. Sub-spaces LL, HL,LH, and HH are calculated by putting the values of ϕ(n1, n2) and ψ
i(n1, n2) from
31
the Eq. (2.3)-(2.6) into Eq. (2.1) and (2.2).
ϕ(n1, n2) = ϕ(n1)ϕ(n2) (2.3)
ψH(n1, n2) = ψ(n1)ϕ(n2) (2.4)
ψV (n1, n2) = ϕ(n1)ψ(n2) (2.5)
ψD(n1, n2) = ψ(n1)ψ(n2) (2.6)
Figure 2.1: Three level decomposition layout of an image.
To reconstruct the image from the DWT coefficients, inverse DWT (IDWT) is
calculated as follows :
f(n1, n2) =1√N1N2
∑k1
∑k2
Wϕ(j0, k1, k2)ϕj0,k1,k2(n1, n2)
+1√N1N2
∑i=H,V,D
∞∑j=0
∑k1
∑k2
W iψ(j0, k1, k2)ψ
ij0,k1,k2(n1, n2)
(2.7)
To embed the watermark in DWT-based decomposed image, the transform coeffi-
cients of DWT are modified by watermark. Since, the low frequency band (LL) of
DWT decomposed image is similar to the original image, most of the information
or energy of original image lies in this frequency band. In order to maintain the
quality of watermarked image, this low frequency or approximation detail must be
preserved and maintained the robustness of embed watermark. Therefore, it is the
trade-off between the robustness and quality that at which extent the transform
coefficients are to be modified in order to optimize the overall method.
32
2.2.2 Uncorrelated Color Space (UCS)
The quality of the color image watermarking methods depends on how images
are split into three color channels, i.e. which color space was chosen. For better
quality, color space must be uncorrelated which makes the three color channels
semi-independent [103, 108] and may be used for embedding the watermark. This
method uses the recently developed UCS proposed by Liu [109]. UCS is derived
from RGB color space using principal component analysis (PCA). UCS uses a
linear transformation, WU ∈ R3×3, of the RGB color space to uncorrelate the
component images as shown in Eq. (2.8) [109];
⎡⎢⎣U(x, y)
C(x, y)
S(x, y)
⎤⎥⎦ = WU
⎡⎢⎣R(x, y)
G(x, y)
B(x, y)
⎤⎥⎦ (2.8)
The WU is calculated by factorizing the covariance matrix C using PCA in the
following form [109]:
C =W tUΛWU (2.9)
hereW tU and Λ are the orthogonal eigenvector matrix and diagonal eigenvalue ma-
trix with diagonal elements in a decreasing order, respectively. Saraswat and Arya
[108] used UCS for color transfer of images and observed that UCS outperforms
the other uncorrelated color spaces.
2.2.3 Genetic Algorithm (GA)
Genetic algorithm (GA) belongs to the class of evolutionary algorithms and is
an ingredient of artificial intelligence. These algorithms are capable to encode a
solution to the different engineering problems. Furthermore, to achieve a better
solution to the problem these algorithms use methods which are stimulated by
nature namely; inheritance, reproduction, mutation, and crossover. GAs were
reported by the Holland [110, 111].
Terminology
The term used in GA are as follows:
• Search space : the space of all possible solutions.
• Chromosome : it contains the solution in form of genes.
• Population : a set of individuals/chromosomes.
33
• Generation : the procedure of evaluation, reproduction, crossover and mu-
tation.
• Fitness : the value assigned to an individual based on how far or close it is
from the desired solution.
The pseudo-code of the GA is shown in Algorithm 2.1 and the flow chart is
depicted in Figure 2.2.
Algorithm 2.1 Genetic Algorithm
Genetic representation of solution to the problemCreate and initialize the population individuals;Evaluate the fitness of each individual in populationwhile Termination criteria is not satisfied do
1. Reproduction: Select the individuals with greater fitness for reproduction.2. Crossover: Reproduce new individuals through crossover.3. Mutation: Apply probabilistic alteration or modification on new individuals.4. Form a new population with these offsprings.
end while
Figure 2.2: Genetic Algorithm Flow Chart.
In GA, there are three operators: reproduction, crossover, and mutation. Ini-
tially, a population is generated randomly with uniform distribution followed by
reproduction, crossover, and mutation operators to generate a new population.
Offspring vector generation is a crucial step in GA process. The two operators,
34
crossover and mutation are used to generate the offspring vectors. The reproduc-
tion operator is used to select the best vector between offspring and parent for the
next generation. GA operators are explained briefly in the following sections.
Reproduction
Reproduction or selection operator selects the individuals or chromosomes to
crossover and to generate offsprings. The selection is crucial and important step
and it is based on the principle of the best one to be survive and generates new
offsprings. To select the best individual for the optimum solution of given prob-
lem, the fitness function is formulated. This function shows the closeness of a
current result to the desired result. There are various techniques for reproduction
here F ∈ [0, 1] is the mutation scale factor which is used for controlling the
amplification of the differential variation [115].
Crossover
Offspring x′i(G) is generated using the crossover of parent vector (xi(G)) and the
trial vector (ui(G)) as follows:
x′ij(G) =
⎧⎨⎩uij(G), if j ∈ J
xij(G), otherwise.(4.6)
here J is the set of crossover points or the points that will go under perturbation
and xij(G) is the jth element of the vector xi(G).
62
Different methods may be used to determine the set J of crossover points in
which binomial crossover and exponential crossover are the most frequently used
[115]. In this chapter, the DE and its variants are implemented using binomial
crossover whereas for a D dimensional problem, the crossover points are randomly
selected from the set of possible points, {1, 2, . . . , D}. Algorithm 4.2 shows the
steps of binomial crossover to generate crossover points [115].
Algorithm 4.2 Binomial CrossoverLet CR represents the probability with which the considered crossover points will be included.U(1,D) is a uniformly distributed random integer between 1 and D.J = φj∗ ∼ U(1, D);J ← J ∪ j∗;for each j ∈ 1...D do
if U(0, 1) < CR and j �= j∗ thenJ ← J ∪ j;
end ifend for
Selection
There are two functions for the selection operator: First, it selects the individual
for the mutation operation to generate the trial vector and second, it selects the
best between the parent and the offspring based on their fitness value for the next
generation. If fitness of the parent is greater than the offspring then parent is
selected otherwise offspring is selected:
xi(G+ 1) =
⎧⎨⎩x′i(G), if f(x′i(G)) > f(xi(G)).
xi(G), otherwise.(4.7)
This ensures that the population’s average fitness does not deteriorate. The
Pseudo-code for DE method is described in Algorithm 4.3 [115].
Algorithm 4.3 Differential Evolutionary AlgorithmLet F and CR are the control parameters termed as scale factor and crossover probability respectively.Let P is the population vector.Initialize the control parameters F and CR;Create and initialize the population P (0) of NP individuals;while termination condition do
for each individual xi(G) ∈ P (G) doEvaluate the fitness f(xi(G));Create the trial vector ui(G) by applying the mutation operator;Create an offspring x′i(G) by applying the crossover operator;if f(x′i(G)) is better than f(xi(G)) then
Add x′i(G) to P (G+ 1);else
Add xi(G) to P (G+ 1);end if
end forend whileReturn the fittest individual as the solution.
63
4.4 Proposed Method
The proposed method explores the advantages of uncorrelated color space over
the correlated color space to improve the performance of watermarking methods
in terms of quality and robustness. It implants the color watermark image into
the color host image by modifying the decomposed wavelet coefficients of host
image. The each channel of color image is embedded in the corresponding channel
of host image to increase the reliability during the recovery process and protect
against the common signal processing attacks. The proposed method consists
of five phases namely; pre-processing of host and watermark image, watermark
embedding, image post-processing, extraction of watermark, and performance im-
provement using GA, ABC, and DE. The basic structural design of the proposed
method is shown in Figure 4.1. Since each color image has three channels, the
structural design and methodology represented in Figure 4.1 are repeated for all
the three channels. The details of each phase of the proposed method is described
in the following sections.
Image Pre-processing
In image pre-processing, both the host RGB color image (H) and watermark image
(W ) are transformed into UCS color space using Eq. (2.8) which produces six
independent image components (three for host and three for watermark) namely;
HU , HC , and HS for host image and WU , WC , and WS for watermark image.
After transformation, each component of watermark image is divided into 16 non-
overlapping sub-areas as shown in Figure 4.2. These sub-areas of watermark image
are then re-arranged or scrambled by some pre-defined sequence or key to introduce
one more level of security and enforced the user to use the key for the extraction
of the image from watermarked image.
Further, third level of decomposition is applied on each component of host im-
age using symlet-4 wavelet function. The resultant third level wavelet coefficients,
(Hk3, k = {LL,HL, LH,HH}) is chosen for embedding process.
Watermark Embedding
After pre-processing of the images, the scrambled watermark is embedded into
the DWT coefficients of host image. The third level decomposition coefficients of
host image are HLL3(x, y), HHL3(x, y), HLH3(x, y), and HHH3(x, y) representing
the approximation, horizontal, vertical, and diagonal details of 3-DWT decom-
posed host image respectively. In order to increase the reliability and robustness
64
Figure 4.1: Structural design of proposed method.
Figure 4.2: Watermark division.
65
of proposed method against the malicious attacks, it is desired to hide each color
component of the watermark image in corresponding component of host coeffi-
cients. Therefore, approximation details of the third level coefficients are divided
into 16 non-overlapping areas as depicted in Figure 4.3.
Figure 4.3: Partitioning of 3-DWT coefficients.
Now, the scrambled watermark is embedded into approximation details of the
third level decomposed host image coefficients by using Eq. (4.8).
Now, the watermark images are embedded into each of the considered host
images and the resultant watermarked images are shown in Figure 4.9.
To compare the objective test for quality measures of resultant watermarked
images using proposed and considered methods, CPSNR and SSIM performance
parameters are calculated and shown in Table 4.2. From Table 4.2, it is validated
that all the methods including proposed method maintain the quality of the wa-
termarked image in terms of CPSNR (> 35dB) and SSIM (> 0.96). However,
the proposed method shows lower values of CPSNR and SSIM as compared to
other considered methods due to the hiding of complete watermark into all three
channels of host image while existing methods hide in only one channel of host
image. Moreover, subjective test on the resultant watermarked images is per-
formed and presented in Table 4.3. The results of subjective test show that the
proposed method effectively embeds the watermarks which is imperceptible by
human beings.
71
(a) (b) (c) (d)
(e) (f) (g) (h)
Figure 4.9: RGB watermarked images embedded by using DE-based proposedmethod (a)-(d) RTU logo and (e)-(h) Aeroplane image
Table 4.2: Comparison of CPSNR and SSIM values of watermarked imagesresultant from proposed and considered methods.
S.No. QualityParameters
Watermarkedimages
Su et al.
[98]Chouand Wu[96]
ProposedMethodusingGA
ProposedMethodusingABC
ProposedMethodusingDE
1. CPSNR Lena 36.57 37.79 35.81 35.92 36.01
Mandrill 36.42 37.71 35.13 35.67 35.89
Pepper 36.61 37.01 35.07 35.23 35.41
Sailboat 36.52 37.32 35.00 35.06 35.21
2. SSIM Lena 0.98 0.98 0.98 0.98 0.98
Mandrill 0.98 0.98 0.97 0.96 0.97
Pepper 0.96 0.97 0.96 0.96 0.96
Sailboat 0.98 0.97 0.97 0.97 0.97
Table 4.3: Average subjective quality comparison of original and watermarkedimages by 10 human beings in the scale of 0 to 5.
Average Score of 10 Human beings
S.No. Watermarkimage
Watermarkedimage
Su et al.
[98]Chouand Wu[96]
ProposedMethodusingGA
ProposedMethodusingABC
ProposedMethodusingDE
1. RTU Logo Lena 5 5 5 5 5
Mandrill 5 5 5 5 5
Pepper 5 5 5 5 5
Sailboat 5 5 5 5 5
2. Aeroplane Lena 5 5 5 5 5
Mandrill 5 5 5 5 5
Pepper 5 5 5 5 5
Sailboat 5 5 5 5 5
72
To show the effectiveness of the proposed watermarking method, the extracted
watermarks from the watermarked images and their NC values are shown in Figure
4.10 for each of the considered method. From Figure 4.10, it is visualized that the
proposed method and Su et al. [98] have highest NC values (1.0) for all extracted
watermarks and hence outperforms the method of Chou and Wu [96]. The columns
of Figure 4.10 show extracted watermarks from considered watermarked images
and first five rows show the extraction of RTU logo while last five show extraction
of aeroplane image using the proposed and considered methods of watermarking.
The robustness of the proposed method has been tested by applying different
attacks on the watermarked images. In this chapter, attacks have been catego-
rized into two classes namely; common signal processing attacks and geometric
attacks. The considered common signal processing attacks consists of filtering at-
tacks (mean, median, wiener), noise attacks (gaussian, poisson, salt and pepper),
and JPEG compression attacks while rotation, scaling, and cropping are the con-
sidered geometric attacks. The attacks are applied on all the eight watermarked
images embedded with RTU logo and aeroplane images. The comparison of NC
values after applying the common signal processing attacks on watermarked im-
ages embedded with RTU logo are depicted in Table 4.4 while Table 4.5 shows
the NC values for watermarked images embedded with aeroplane image. After
applying the geometric attacks, the measured NC values for both the watermark
images are compared in Table 4.6. From Tables 4.4, 4.5, and 4.6, it is observed
that the robustness of watermarked images embedded with RTU logo have higher
values of NC as compared to aeroplane image due to the coarseness of aeroplane
image. The comparative results show that the proposed method using DE out-
performs other methods for all the considered attacks except JPEG compression
and rotation where the method of Su et al. shows slightly better robustness. The
similar performance of the proposed method can be observed from Figure 4.11 and
Figure 4.12 where the extracted RTU logo and Aeroplane watermarks by proposed
method using DE respectively, for all the considered attacks have been depicted.
Therefore, it is validated from the results that the proposed method using DE
produces high quality and better robust watermarked images and can be utilized
for content authentication to protect the copyrighted images.
4.7 Results and Discussions
This chapter proposes a novel DWT-based color image watermarking method us-
ing UCS. The use of uncorrelated color space increases the effective utilization of
all color channels of host image which is not feasible in correlated color spaces.
73
Used Watermarked Image
Method a. Lena b. Mandrill c. Pepper d. Sailboat
Chou and Wu [96]
(NC) 1.00 0.99 0.99 0.98
Su et al. [98]
(NC) 1.00 1.00 1.00 1.00
Proposed Method using GA
(NC) 1.00 1.00 1.00 1.00
Proposed Method using ABC
(NC) 1.00 1.00 1.00 1.00
Proposed Method using DE
(NC) 1.00 1.00 1.00 1.00
Chou and Wu [96]
(NC) 0.99 0.98 0.99 0.99
Su et al. [98]
(NC) 1.00 1.00 1.00 1.00
Proposed Method using GA
(NC) 1.00 1.00 1.00 1.00
Proposed Method using ABC
(NC) 1.00 1.00 1.00 1.00
Proposed Method using DE
(NC) 1.00 1.00 1.00 1.00
Figure 4.10: Comparison of extracted watermarks by considered and proposedmethods along with their corresponding NC values. Columns shows extractedwatermarks from watermarked image namely (a). Lena, (b). Mandrill, (c).
Pepper, and (d). Sailboat using the considered and proposed methods mentionedin first column. First five rows shows the extraction of RTU logo while last five
The performance of the proposed method is compared with method proposed
in the chapter 4, in which DWT and DE have been used to increases the quality
and robustness of image watermarking method. To compare the objective test for
quality measures of resultant watermarked images using proposed and considered
methods, CPSNR and SSIM performance parameters are calculated and depicted
in Table 5.1. From Table 5.1 it is validated that all the methods including proposed
method maintain the quality of the watermarked image in terms of CPSNR (>
35dB) and SSIM (> 0.95). However, the proposed method shows lower values of
CPSNR and SSIM as compared to other considered methods due to the hiding
of complete watermark into all the three channels of host image, while existing
methods hide in only one channel of the host image. Further, the results using
SPT is more promising than DWT-based method. Moreover, subjective test on
the resultant watermarked images is performed and presented in Table 5.2. The
results of subjective test show that the proposed method effectively embeds the
88
(a) (b) (c) (d)
(e) (f) (g) (h)
Figure 5.7: RGB watermarked images embedded by (a)-(d) RTU logo and(e)-(h) Aeroplane image
Figure 5.8: A comparison of before and after optimization of fitness values for 30runs.
watermarks which is imperceptible by human beings.
To show the effectiveness of the proposed watermarking method, the extracted
watermarks from the watermarked images and their NC values are shown in Figure
5.9 for each of the considered method. From Figure 5.9, it is visualized that the
proposed method using UCS color space and Su et al. [98] have highest NC values
(1.0) for all extracted watermarks and hence outperforms the method of Chou and
Wu [96]. The columns of Figure 5.9 show extracted watermarks from considered
watermarked images and first four rows shows the extraction of RTU logo while
last four shows extraction of aeroplane image using the proposed and considered
89
Table 5.1: Comparison of CPSNR and SSIM values of watermarked imagesresultant from proposed and considered methods.
S.No. QualityParameters
Watermarkedimages
Su et al.
[98]Chou andWu [96]
DWT-basedProposedMethod
SPT-basedProposedMethod
1. CPSNR Lena 36.94 38.17 36.01 36.28
Mandrill 36.78 38.09 35.89 36.03
Pepper 36.98 37.38 35.41 35.58
Sailboat 36.89 37.69 35.21 35.41
2. SSIM Lena 0.98 0.97 0.98 0.98
Mandrill 0.97 0.96 0.97 0.96
Pepper 0.97 0.97 0.96 0.96
Sailboat 0.97 0.97 0.97 0.97
Table 5.2: Average subjective quality comparison of original and watermarkedimages by 10 human beings in the scale of 0 to 5.
Average Score of 10 Human beings
S.No. Watermarkimage
Watermarkedimage
Su et al.
[98]Chou andWu [96]
DWT-basedProposedMethod
SPT-basedProposedMethod
1. RTU Logo Lena 5 5 5 5
Mandrill 5 5 5 5
Pepper 5 5 5 5
Sailboat 5 5 5 5
2. Aeroplane Lena 5 5 5 5
Mandrill 5 5 5 5
Pepper 5 5 5 5
Sailboat 5 5 5 5
methods of watermarking.
The robustness of the proposed method has been tested by applying different
attacks on the watermarked images. In this chapter, attacks have been catego-
rized into two classes namely; common signal processing attacks and geometric
attacks. The considered common signal processing attacks consists of filtering at-
tacks (mean, median, wiener), noise attacks (gaussian, poisson, salt and pepper),
and JPEG compression attacks while rotation, scaling, and cropping are the con-
sidered geometric attacks. The attacks are applied on all the eight watermarked
images embedded with RTU logo and aeroplane images. The comparison of NC
values after applying the common signal processing attacks on watermarked im-
ages embedded with RTU logo are depicted in Table 5.3 while Table 5.4 shows
the NC values for watermarked images embedded with aeroplane image by all
considered methods and proposed method. After applying the geometric attacks,
the measured NC values for both the watermark images are compared in Table
5.5. From Tables 5.3 – 5.5, it is observed that the robustness of watermarked
images embedded with RTU logo have higher values of NC as compared to aero-
90
Used Watermarked ImageMethod a. Lena b. Mandrill c. Pepper d. Sailboat
Chou and Wu [96]
(NC) 0.98 0.99 0.99 0.98
Su et al. [98]
(NC) 1.00 1.00 1.00 1.00
DWT-based Proposed Method
(NC) 1.00 0.99 1.00 1.00
SPT-based Proposed Method
(NC) 1.00 1.00 1.00 1.00
Chou and Wu [96]
(NC) 0.99 0.98 0.99 0.99
Su et al. [98]
(NC) 1.00 1.00 1.00 1.00
DWT-based Proposed Method
(NC) 1.00 0.99 1.00 0.99
SPT-based Proposed Method
(NC) 1.00 1.00 1.00 1.00
Figure 5.9: Comparison of extracted watermarks by considered and proposedmethods along with their corresponding NC values. Columns shows extractedwatermarks from watermarked image namely (a). Lena, (b). Mandrill, (c).
Pepper, and (d). Sailboat using the considered and proposed methods mentionedin first column. First four rows shows the extraction of RTU logo while last four