[intelligent systems reference library] bio-inspiring cyber security and cloud services: trends and...

Chapter 4Bio-inspiring Techniques in WatermarkingMedical Images: A Review

Mona M. Soliman, Aboul Ella Hassanien and Hoda M. Onsi

Abstract Bio-inspiring (BI) is a well-established paradigm with current systemshaving many of the characteristics of biological computers and capable of perform-ing a variety of tasks that are difficult to do using conventional techniques. BI isconsidered as one of the most important increasing fields, which attract a large num-ber of researchers and scientists working in areas such as neuro-computing, globaloptimization, swarms and evolutionary computing. On the other hand, digital radio-logical modalities in modern hospitals have led to the producing a variety of a vastamount of digital medical files. Therefore, for the medical imaging, the authentic-ity needs to ensure the image belongs to the correct patient, the integrity check toensure the image has not been modified, and safe transfer are very big challenges.The integrity of the images must be protected by using watermarking, which is calledintegrity watermark. At the same time the copyright and intellectual property of themedical images should be also protected, which is called copyright watermark. Thischapter presents a brief overview of well known Bio-inspiring techniques includingneural networks, genetic algorithm, swarms and evolutionary algorithms and showhow BI techniques could be successfully employed to solve watermarking problem.Challenges to be addressed and future directions of research are also presented andan extensive bibliography is included.

M. M. Soliman (B) · A. E. HassanienScientific Research Group in Egypt (SRGE), Faculty of Computers and Information,Cairo University, Cairo, Egypte-mail: [email protected]: http://www.egyptscience.net

A. E. Hassaniene-mail: [email protected]

H. M. OnsiFaculty of Computers and Information, Cairo University, Cairo, Egypt

A. E. Hassanien et al. (eds.), Bio-inspiring Cyber Security and Cloud Services: 93Trends and Innovations, Intelligent Systems Reference Library 70,DOI: 10.1007/978-3-662-43616-5_4, © Springer-Verlag Berlin Heidelberg 2014

94 M. M. Soliman et al.

4.1 Introduction

The development of multimedia and communication technologies has made med-ical images act important roles in the fields of telediagnosis and telesurgery. At thesame time such advances provide new means to share, handle and process medicalimages, it also increases security issues in terms of: confidentiality, availability andintegrity. Since any person with privilege can access to images which are containedin database and can modify them maliciously, the integrity of the images must beprotected by using watermarking, which is called integrity watermark. Meanwhile,web-based image database system contains valuable medical image resources fornot only research purpose but also commercial purpose. Therefore the copyright andintellectual property of the database should be also protected by a watermark, whichis called copyright watermark. The basic principle of watermarking methods is toadd copyright information into the original data, by embedding it into the originalimage. Then if the image is modified in any sense, it can be detected with the water-mark. Initially, the watermark could be simply a unique number, such as the patientinsurance code but as research moves into new paths, a new role has been given tothe watermark: to include (apart from hospital digital signatures or copyright infor-mation), the electronic patient record, digital documents with diagnosis, blood testprofiles or an electrocardiogram signal.

For the medical image, the authenticity needs to ensure the image belongs tothe correct patient, the integrity check to ensure the image has not been modified,and safe transfer are very big challenges. Also, when a digital medical image isopened for diagnosis, it is important that an automated framework exists to verifythe authenticity and integrity of the image itself. Hospital Information System (HIS)and Picture Archiving and Communication System (PACS) have been establishedto provide security solutions to ensure confidentiality, integrity and authentication[1]. Digital image watermarking provides copyright protection to digital image byhiding appropriate information in original image to declare rightful ownership [2].The primary applications of watermarking are to protect copyrights and integrityverification. The main reason for protecting copyrights is to prevent image piracywhen the transmitter sends it on the internet. For integrity verification, it is importantto ensure that the medical image originated from a specific source and that it has notbeen changed, manipulated or falsified [3].

There are several types of algorithms for watermarking. Each type of algorithmshas its own advantages and limitations. Nomethod can provide fully perfect solution.Each type of solution has robustness to some type of attacks but is less resilient tosome other types of attacks. In medical applications, because of their diagnosticvalue, it is very important to maintain the quality of images. For this matter, thedevelopment of a new algorithm that can satisfy both invisibility and robustness isneeded. Improvements in performance of watermarking schemes can be obtainedby several methods. One way is to make use of Intelligent computing techniques.The objective of this chapter is to present to the medical watermarking researchcommunities some of the state-of-the-art in BI applications to medical watermarking

4 Bio-inspiring Techniques in Watermarking Medical Images 95

problem andmotivate research in new trend-setting directions. Hence, we review anddiscuss some representative methods to provide inspiring examples to illustrate howBI techniques can be applied to solve medical watermarking problems. We wantto stress that the literature in this domain is of course huge and that therefore thework that we include here is only exemplatory and focussing on recent advances,while many other interesting research approaches had to be omitted due to spacelimitations [4].

The rest of the chapter is organized as follows. Section4.2 introduces the funda-mental aspects of the key components of modern BI techniques including: neuralnetworks, evolutionary algorithms (EA), genetic algorithms (GA) and swarm intelli-gence. Of course there are many other methods of BI but, we only focusing on thoseBI methods used in medical image watermarking. Section4.3 review some detailsof medical image watermarking categories and classification with brief illustrationof the generic watermarking procedure. Section4.4 reviews and discusses some suc-cessful approaches to illustrate howBI can be applied tomedical imagewatermarkingproblems. Section4.5 explore different watermarking assessment measures used inthe evaluation of watermarking algorithms. Challenges and future trends are pre-sented in Sect. 4.6 and an extensive bibliography is provided.

4.2 Bio-inspiring Computing

In this section, we explore an overview ofmodern BI techniques applied successfullyon medical image watermarking problem including, neural networks, EAs, GAs andswarm intelligence.

4.2.1 Artificial Neural Networks

Artificial Neural Network (ANN) is a computational structure paradigm modeled onthe biological process that is inspired by the way biological nervous systems, suchas the brain, processes information. The key element of this paradigm is the novelstructure of the information processing system [5]. Neural computing is an alternativeto programmed computing which is a mathematical model inspired by biologicalmodels. This computing system is made up of a number of artificial neurons anda huge number of interconnections between them. There are six major componentsmake up an artificial neuron [6]. These components are valid whether the neuron isused for input, output, or is in the hidden layers:

• Weighting Factors: A neuron usually receives many simultaneous inputs. Eachinput has its own relative weight, which gives the input the impact that it needs onthe processing element’s summation function. Some inputs are made more impor-tant than others to have a greater effect on the processing element as they combineto produce a neural response. Weights are adaptive coefficients that determine theintensity of the input signal as registered by the artificial neuron.


• The summation function: The input and weighting coefficients can be combinedin many different ways before passing on to the transfer function. In addition tosumming, the summation function can select the minimum, maximum, majority,product or several normalizing algorithms. The specific algorithm for combiningneural inputs is determined by the chosen network architecture and paradigm.Some summation functions have an additional activation function applied to theresult before it is passed on to the transfer function for the purpose of allowing thesummation output to vary with respect to time.

• Transfer Function: The result of the summation function is transformed to a work-ing output through an algorithmic process known as the transfer function. In thetransfer function the summation can be compared with some threshold to deter-mine the neural output. If the sum is greater than the threshold value, the processingelement generates a signal and if it is less than the threshold, no signal (or someinhibitory signal) is generated.

• Scaling and Limiting: After the transfer function, the result can pass through addi-tional processes, which scale and limit. This scaling simply multiplies a scalefactor times the transfer value and then adds an offset. Limiting is the mechanismwhich insures that the scaled result does not exceed an upper, or lower bound. Thislimiting is in addition to the hard limits that the original transfer function may haveperformed.

• Output Function “Competition”: Each processing element is allowed one outputsignal, which it may give to hundreds of other neurons. Normally, the output isdirectly equivalent to the transfer function’s result. Some network topologies mod-ify the transfer result to incorporate competition among neighboring processingelements.

In most cases ANN is an adaptive system that changes its structure based onexternal or internal information that flows through the network during the learningphase. The learning capability of an artificial neuron is achieved by adjusting theweights in accordance with a chosen learning algorithm. Learning algorithms can besupervised, unsupervised, or reinforced [7]. The reader may refer to [8–10] for anextensive overview of the artificial neural networks.

4.2.2 Evolutionary Algorithms

Evolution in nature is responsible for the design of all living beings on earth, and forthe strategies theyuse to interactwith eachother. Evolutionary algorithms employ thispowerful design philosophy to find solutions to hard problems. The idea of applyingthe biological principle of natural evolution to artificial systems, introduced morethan four decades ago, has seen impressive growth in the past few years. Known asevolutionary algorithms or evolutionary computation, these techniques are commonnowadays, having been successfully applied to numerous problems from differentdomains, including optimization, automatic programming, circuit design, machinelearning, economics, ecology and population genetics [11].


Over many generations, natural populations evolve according to the principles ofnatural selection and “survival of the fittest”. Bymimicking this process, evolutionaryalgorithms are able to “evolve” solutions to real world problems, if they have beensuitably encoded [12].

Evolutionary computation makes use of a metaphor of natural evolution [13].According to this metaphor, a problem plays the role of an environment wherein livesa population of individuals, each representing a possible solution to the problem. Thedegree of adaptation of each individual (i.e. candidate solution) to its environmentis expressed by an adequacy measure known as the fitness function. The phenotypeof each individual, i.e. the candidate solution itself, is generally encoded in somemanner into its genome (genotype). Like evolution in nature, evolutionary algorithmspotentially produce progressively better solutions to the problem. This is possiblethanks to the constant introduction of new genetic material into the population, byapplying so-called genetic operators that are the computational equivalents of naturalevolutionary mechanisms.

There are several types of evolutionary algorithms, among which the best knownare genetic algorithms [14], genetic programming [15], evolution strategies [16],evolutionary programming [17] and learning classifier systems [18]; though differentin the specifics they are all based on the same general principles. All share a commonconceptual base of simulating the evolution of individual structures via processes ofselection, mutation and reproduction.

These processes depend on the perceived performance of the individual structuresas defined by the environment. EAs deal with parameters of finite length, whichare coded using a finite alphabet, rather than directly manipulating the parametersthemselves. This means that the search is unconstrained neither by the continuity ofthe function under investigation, nor the existence of a derivative function.

The application of an evolutionary algorithm involves a number of importantconsiderations. The first decision to take when applying such an algorithm is how toencode candidate solutions within the genome. The representation must allow for theencoding of all possible solutions while being sufficiently simple to be searched in areasonable amount of time. Next, an appropriate fitness function must be defined forevaluating the individuals. The (usually scalar) fitness must reflect the criteria to beoptimized and their relative importance. Representation and fitness are thus clearlyproblem-dependent, in contrast to selection, crossover and mutation, which seemprima facie more problem-independent. Practice has shown, however, that whilestandard genetic operators can be used, one often needs to tailor these to the problemas well.

4.2.3 Genetics Algorithms

Genetic algorithms (GAs), introduced byHollard in his seminal work, are commonlyused as adaptive approaches that provide a randomized, parallel and global searchmethod based on the mechanics of natural selection and genetics in order to find


solutions. In biology, the gene is the basic unit of genetic storage [19]. Within cells,genes are strung together to form chromosomes. The simplest possible sexual repro-duction is between single-cell organisms. The two cells fuse to produce a cell withtwo sets of chromosomes, called a diploid cell. The diploid cell immediately under-goes meiosis. In meiosis, each of the chromosomes in the diploid cell makes an exactcopy of itself. Then the chromosome groups (original and copy) undergo crossoverwith the corresponding groups, mixing the genes somewhat. Finally the chromo-somes separate twice, giving four haploid cells. Mutation can occur at any stage, andany mutation in the chromosomes will be inheritable. Mutation is essential for evo-lution. There are three types relevant to genetic algorithms: point mutations wherea single gene is changed, chromosomal mutations where some number of genes arelost completely, and inversion where a segment of the chromosome becomes flipped.

The power of GAs as mentioned in [20] comes from the fact that the techniqueis robust and can deal successfully with a wide range of problem areas includingthose which are difficult for other methods to solve. GAs are not guaranteed to findthe global optimum solution to a problem but they are generally good at findingacceptably good solutions to problems.

In GA, a candidate solution for a specific problem is called an individual or achromosome and consists of a linear list of genes. Each individual represents a pointin the search space, and hence a possible solution to the problem. A population con-sists of a finite number of individuals. Each individual is decided by an evaluatingmechanism to obtain its fitness value. Based on this fitness value and undergoinggenetic operators, a new population is generated iteratively with each successivepopulation referred to as a generation. The GAs use three basic operators (reproduc-tion, crossover and mutation) to manipulate the genetic composition of a population[21]. A population is created with a group of randomly individuals. The individualsin the population are then evaluated by fitness function. Two individuals (off-spring)are selected for the next generation based on their fitness. Crossover is a processyielding recombination of bit strings via an exchange of segments between pairs ofchromosomes to create the new individuals. Finally, mutation has the effect of ensur-ing that all possible chromosomes are reachable or a certain number of generationshave passed.. The reader may refer to [22–26] for an extensive overview of the GAs.

4.2.4 Swarm Intelligence

The expression swarm intelligence was introduced by Beni and Wang in 1989,in the context of cellular robotic systems, SI systems are typically made up of apopulation of simple agents interacting locally with one another and with their envi-ronment. Although there is normally no centralized control structure dictating howindividual agents should behave, local interactions between such agents often leadto the emergence of global behavior. Examples of systems like this can be found innature, including ant colonies, bird flocking, animal herding, bacteria molding andfish schooling [27]. Optimization techniques inspired by swarm intelligence have


become increasingly popular during the last decade [28]. They are characterized bya decentralizedway ofworking thatmimics the behavior of swarms. The advantage ofthese approaches over traditional techniques is their robustness and flexibility. Theseproperties make swarm intelligence a successful design paradigm for algorithms thatdeal with increasingly complex problems. Two successful examples of optimizationtechniques inspired by swarm intelligence are : ant colony optimization and particleswarm optimization [29]. Ant Colony Optimization (ACO) has been successfullyapplied to solve various engineering optimization problems. ACO algorithms canalso be used for clustering datasets and to optimize rule bases.

Dorigo and Blumb in [30] introduced the first ACO algorithms in the early 1990s.The development of these algorithms was inspired by the observation of ant colonies.Ants are social insects. They live in colonies and their behavior is governed by thegoal of colony survival rather than being focused on the survival of individuals. Thebehavior that provided the inspiration for ACO is the ants foraging behavior, andin particular, how ants can find shortest paths between food sources and their nest.While moving, ants leave a chemical pheromone trail on the ground [31]. Ants cansmell pheromone.When choosing their way, they tend to choose, in probability, pathsmarked by strong pheromone concentrations. As soon as an ant finds a food source,it evaluates the quantity and the quality of the food and carries some of it back tothe nest. During the return trip, the quantity of pheromone that an ant leaves on theground may depend on the quantity and quality of the food. The pheromone trailswill guide other ants to the food source. The concept of particle swarms, althoughinitially introduced for simulating human social behaviors, has become very pop-ular these days as an efficient search and optimization technique. PSO has gainedincreasing popularity among researchers and practitioners as a robust and efficienttechnique for solving difficult optimization problems. In PSO, individual particlesof a swarm represent potential solutions, which move through the problem searchspace seeking an optimal, or good enough, solution. The particles broadcast theircurrent positions to neighboring particles. The position of each particle is adjustedaccording to its velocity (i.e. rate of change) and the difference between its currentposition, respectively the best position found by its neighbors, and the best positionit has found so far. As the model is iterated, the swarm focuses more and more onan area of the search space containing high-quality solutions [32]. We have to notethat PSO is mainly used for continuous optimization while ACO is mainly used forcombinatorial optimization. The reader may refer to [33–38].

4.3 Medical Image Watermarking: Classificationand Generic Model

Digital radiological modalities in modern hospitals have led to the producing avariety of a vast amount of digital medical files. It is very common and usualfor physicians to participate in technical group communication existed betweenphysicians and hospitals in order to diagnosing and spotting the patient’s problem.Hospital Information System (HIS) comprising radiology information system (RIS)


and picture archiving and communication system (PACS) based on Digital Imagingand Communication in Medicine standard (DICOM) has facilitated offering vari-ous e-Health services. These e-Health services are introducing new practices forthe profession as well as for the patients by enabling remote access, transmissionand interpretation of the medical images for diagnosis purposes [39]. The health-care professionals use the Internet for transmitting or receiving Electronic PatientRecords (EPR) via e-mail. An EPR typically contains the health history of a patient,including X-ray images, CT-Scan images, physical examinations report, laboratorytests, treatment procedures, prescriptions, radiology examinations etc. An EPR canbe represented in various forms such as diagnostic reports, images, vital signals,etc. An EPR transmitted through the Internet is very important since it contains themedical information of a person in digital format [40].

4.3.1 Medical Watermarking Classification

As ordinary watermarking techniques, Medical image watermarking techniques canbe categorized in different ways. They can be classified according to the type ofwatermark being used, i.e. the watermark may be visible or invisible. Visible water-marking is the process of embedding the watermark into some visible parts of thedigital media [41]. Examples of visible watermarking can be seen as TV logos andcompany emblems. As for invisible watermarking, it is the process of embeddingthe watermark confidentially into the parts of the digital media known only by theowner [42].

For different purposes, digital watermarking has been branched into two classi-fications: robust watermarking technique and fragile watermarking technique [43].Robust watermarks are designed to resist intentional or unintentional image mod-ifications for instance filtering, geometric transformations, noise addition, etc. Incontrast, fragile watermarking is used to determine the modifications on the digitalmedia that is, to ensure the integrity of the digital media [41]. The design goal ofrobust watermarking is to make the embedded watermarks remain detectable afterbeing attacked. In contrast, the requirements of fragile watermarking are to detectthe slightest unauthorized modifications and locate the changed regions.

Another classification is based on domain which the watermark is applied i.e.the spatial domain or the frequency domain. The earlier watermarking techniqueswere almost in spatial domain. Spatial-domain schemes embed the watermark bydirectly modifying the pixel values of the original image [3]. Spatial domainmethodsare less complex and not robust against various attacks as no transform is used inthem. The frequency domain techniques based on transforming the original mediato frequency coefficient by using some transformations such as discrete Fouriertransformation, discrete cosine transformation and discrete wavelet transformation.Then, the watermark is embedded bymodifying coefficients [44]. Frequency domaintechniques are robust as compared to spatial domain methods [45].

There are three possible categories for medical image watermarking have beenidentified in the literature survey [46]. The first category is based on region of


non-interest. A medical image in case of clinical outcome can be divided in twoparts, the region of interest (ROI) where the diagnosis focuses, or it can be definedas an area including important information and must be stored without any distor-tion, while the region of non-interest (RONI), is considering the remaining area. Thedefinition of the ROI space depends on the existence of a clinical finding and itsfeatures [47]. The second category is based on reversibility. Here the watermarkingscheme should able retrieve the original image from the watermarked image withoutany loss of information at the extraction process. The reversible watermarking notonly provides authentication and tamper proofing but also can recover the originalimage from the suspected image. After the verification process if the transmittedimage is deemed to be authentic the doctor reconstitutes the original image (with-out any degradation) and uses it in its diagnosis avoiding all risk of modification[48]. The third category is based on non-reversible. Here, tolerable information lossis accepted as in lossy compression. Any watermarking scheme can be classifiedinto either reversible or irreversible. It can use region of non-interest for embedding.However, the selection of region of images for embedding is application specific.

4.3.2 Generic Medical Image Watermarking System

“Watermarking” is the process of hiding digital information in a carrier signal; thehidden information should, but does not need to contain a relation to the carriersignal. The information to be embedded in a signal is called a digital watermark.The signal where the watermark is to be embedded is called the host signal. Allwatermarking systems as shown in Fig. 4.1 including medical image watermarkingsystem are usually divided into three distinct steps, embedding, attack and detection.

In embedding, an algorithm accepts the host and the data to be embedded, andproduces a watermarked signal. For a blind watermark, the goal is that the digitaldata appears to be the same as before the embedding process, and the casual useris unaware of the watermark’s presence. Then, the watermarked digital signal istransmitted or stored, usually transmitted to another person. If this person makes amodification, this is called an attack. While the modification may not be malicious,the term attack arises from copyright protection application, where third partiesmay attempt to remove the digital watermark through modification. We can classifyattacks into the Simple attack, Detection-disabling or Geometric attacks, Ambiguityattacks and Removal attacks.

Detection (often called extraction) is an algorithmwhich is applied to the attackedsignal to attempt to extract the watermark from it. If the signal was unmodifiedduring transmission, then the watermark still is present and it may be extracted. Inrobust digital watermarking applications, the extraction algorithm should be able toproduce the watermark correctly, even if the modifications were strong. In fragiledigital watermarking, the extraction algorithm should fail if any change is made tothe signal. The watermark detector decides whether a watermark is present or not(refer to [49]).

According to [50] watermark algorithms can be classified into generations basedon the watermark embedding domain. The first generation include those algorithms


Fig. 4.1 General medical watermarking system

that embed watermark signal into spatial domain. There have been several enhance-ments since the first generation to improve the performance in terms of capacity,invisibility, and robustness of the watermark. The second generation algorithms usevarious transformations to insert the watermark in the transform domain coefficientsto enhance robustness. Third generation techniques build on existing first and secondgeneration algorithms and also include hybrid domains to allow information fusionfrom different domains. The proposed third generation algorithms explore the useof computational intelligence techniques to insert a high capacity watermark in boththe spatial and transform domains.

4.4 BI in Medical Image Watermarking

4.4.1 Artificial Neural Networks in Medical Image Watermarking

Only few works have been reported to use neural network to embed watermarkinto medical images [51] watermark mammograms images. These images are water-marked in order to proof its integrity; not modified by unauthorized person, and toascertain the authenticity; ensuring that the image belong to the correct patient and


source. Mammograms contain diagnostic information which can be used for earlydetection of breast cancer diseases and breast abnormality.

Olanr in [51] have been exploited using the Complex version of ANN (CVNN),trained by Complex backpropagation (CBP) algorithm. This technique was usedto embed and detect forge watermark in Fast Fourier Transform FFT domain. Theperformance of the algorithm has been evaluated using mammogram images. Theimperceptibility and detection accuracy was appraised with objective performancemeasure. Results indicate that watermarked mammogram were perceptually indis-tinguishable from the host mammogram, hence the application of the developedCVNN-based watermarking technique in medical images can improve correct diag-noses. Ability of the algorithm to localize modification undergone makes it a uniqueand efficient algorithm for authentication and tamper detection as well as blind detec-tion applications. CVNN is use to process complex valued data (CVD) such as inimage with real and imaginary component. CVNN is made up of Complex-ValuedFeedforward (CVFF) and Complex Back-Propagation (CBP) algorithm. CVNN hasbeen studied and developed by authors in solving various problems. The CVNNconsists of an interconnection of Complex-Valued (CV) neurons and complex val-ued synaptic weights. It processes information using a connectionist approach tocomputation in complex domain.

Oues in [52], attempts to define an adaptive watermarking scheme based onMulti-Layer Feed forward (MLF) neural networks. Neural network applied to digitalwatermark embedded process simulates human visual characteristic to determine themaximum watermark embedded intensity endured by middle frequency coefficientsin every one of (8*8) image block DCT coefficients. The watermarking schemein this work has been tested on the medical images with size of 512 × 512 pixels.The experimental results demonstrated that the proposed method significantlyimprove the perceptual quality of a watermarked image while preserving the robust-ness against various attacks such as of compression, cropping and filtering. Thetrade-off between the imperceptibility and robustness is one of most serious chal-lenges in digital watermarking system, in particular the medical imaging.

4.4.2 Genetic Algorithms in Medical Image Watermarking

The learning capabilities of GA make an effective trade off between the watermarkpayload and imperceptibility, through effective selection of threshold. This trade offhas much more importance in case o sensitive imagery such as medical imagery. In[21] author developed an intelligent reversible watermarking approach for medicalimages. Companding technique is effectively used to achieve higher PSNR valuefor the images and is controlled using threshold. We have observed that the thresh-old value has a pronounce effect on the actual payload available for watermarkembedding. The higher the threshold, the lower is the companding, and the cor-responding companding error, and the higher is the effective payload. Thus, withchange in the threshold, the effective payload and PSNR values are also changedbut in reciprocating manner. The characteristics of each wavelet subband are used to


evolve a threshold matrix. An optimum threshold value is computed for each of thewavelet subband.

Reference [49] presents a robust technique embedding the watermark of signa-ture information or textual data around the ROI of a medical image based on geneticalgorithms. A fragile watermark is adopted to detect any unauthorized modification.The embedding of watermark in the frequency domain is more difficult to be piratedthan in spatial domain. The embedded watermark in the coefficients of the trans-formed image will be somewhat disturbed in the process of transforming the imagefrom its frequency domain to spatial domain because of deviations in convertingreal numbers into integers. This work developed a technique to correct the errorsby using genetic algorithm. In order to protect the copyright of medical images, thewatermark is embedded surrounding their ROI parts. A high compression ratio ispreferable for reducing transmission time. But, it is difficult to attain the same com-pression ratio for an image with low PSNR, because it might degrade the originalimage, making it difficult for clinical reading. Hence it is necessary to find an opti-mal trade-off between the image quality and compression ratio. In addition, it is alsoessential to ensure that the compressed image has sufficient bandwidth to accommo-date the watermark payload. Meanwhile, the embedded watermark is pre-processedby SPIHT (set partitioning in hierarchical trees) Near-Lossless Compression, whosemain requirement is that of ensuring that the maximum error between the originaland the compressed image does not exceed a fixed threshold. In the same line, theconcept of near-lossless watermarking has been introduced recently to satisfy thestrict requirements for medical image watermarking. Moreover, these techniques donot adaptively arrive at an optimal compression ratio. A single compression tech-nique might not be suitable for all medical images because of their differing noisecharacteristics. Reference [53] attempts to investigate, for the first time, the appli-cation of GA in achieving an optimal compression ratio for dual watermarking inwavelet domain without degrading the image.

Reference [54] presents a lossless data hiding method using integer wavelet trans-form and Genetic Programming (GP) based intelligent coefficient selection scheme.By exploiting information about the amplitude of the wavelet coefficient and thetype of the sub band, GP is used to evolve a mathematical function in view of thepayload size and imperceptibility of the marked image. The evolved mathematicalfunction acts like a compact but robust coefficient map for the reversible watermark-ing approach. Information is embedded into the least significant bit-plane of thosehigh frequency wavelet coefficients that are intelligently selected by the GeneticProgramming module.

4.4.3 Swarms Intelligent in Medical Image Watermarking

Image watermarking can be viewed as an optimization problem. Reference [55]proposed an adaptive and optimal watermark method for brain magnetic resonanceimages. First it have used segmentation to extract region of interests (ROI). Patient


information and hospital logo has been used as a watermark and embeded in thenon-region of interest part so that ROI remain same after embedding watermark.Watermark has been embedded in the Discrete Wavelet Transform (DWT) domain.Particle swarm optimization (PSO) has been used to optimize the strength of water-mark intelligently. This work proposed an adaptive watermark method for brainmagnetic resonance images that can be used to secure medical information. Firstsegmentation to extract region of interests (ROI) is performed. Patient informationand hospital logo has been used as a watermark and embedded in the non-region ofinterest part so that ROI remain same after embedding watermark. Watermark hasbeen embedded in the Discrete Wavelet Transform (DWT) domain. Particle swarmoptimization (PSO) has been used to optimize the strength ofwatermark intelligently.

Soliman et al. in [56–59] discuss the subject of the optimization of medical imagewatermarking and provide different solutions. Figure4.2 shows the model designedby Soliman et al. [56] to use PSO in building robust system of medical image water-mark.

Soliman et al. in [56] introduce an adaptive watermarking scheme in medicalimaging based on swarm intelligence. The watermark bits are embedded on singularvalue vector of each embedding block within low frequency sub-band in the hybridDWT-DCT domain. In adaptive watermark scheme the quantization step that usedin determining the embedding watermark is changed for every image. For a singleimage the quantization step is not fixed but it change its value through PSO trainingprocedure till reaching the best quantization step and hence best locations for water-mark bits to be embedded. The embedding strength is more or less proportional tothe perceptual sensitivity to distortions for using adaptive quantization step size. Inorder to resist the normal signal processing and other different attacks, the quanti-zation step has to be as high as possible. However, because the watermark directlyaffects the host image, it is obvious that the higher the quantization step, the lowerthe quality of the watermarked image will be. In other words, the robustness and theimperceptibility of the watermark are contradictory to each other. The experimentsis performed on different MRI medical images, and the proposed approach perfor-mance compared to other scheme with scheme built on ordinary methods and itsshown its superior.

Soliman et al. in [57] present a novel application of Quantum Particle SwarmOptimization (QPSO) in the field of medical image watermarking for copyright pro-tection and authentication. The global convergence of PSO cannot always be guar-anteed because the diversity of population is decreased with evolution developed. Todeal with this problem, concept of a global convergence guaranteed method calledas Quantum behaved Particle Swarm Optimization (QPSO) was developed [60]. Itprovides a good scheme for improving the convergence performance of PSO becauseit is a theoretically global convergence algorithm. It utilizing human visual system(HVS) characteristics and QPSO algorithm in adaptive quantization index modula-tion and singular value decomposition in conjunctionwith discrete wavelet transform(DWT) and discrete cosine transform (DCT). This work provides an enhanced ver-sion of PSO using quantum theory. Quantum computing is a new class of computingbased on the concepts and principles of quantum theory, such as superposition of


Fig. 4.2 Proposed model of medical image watermark


quantum states, entanglement and intervention. This research area was first proposedby Benioff in the early 1980s. Since Benioff in [61] introduced it, quantum comput-ing has developed rapidly, and it also has been turning out that quantum computinghas significant potential to be applied to various difficult problems including opti-mization. In terms of classical mechanics, a particle is depicted by its position vectorxi and velocity vector vi, which determine the trajectory of the particle. The particlemoves along a determined trajectory in Newtonian mechanics, but this is not thecase in quantum mechanics. Experimental results prove the effectiveness of the pro-posed algorithm that yields a medical watermarked image with good visual fidelity,at the same time watermark able to withstand a variety of attacks including JPEGcompression, Gaussian noise, Salt and Pepper noises, Gaussian filter, median filter,image cropping and image scaling.

Another modification had been added to the work proposed in [56], where,a novel watermarking approach based on weighted quantum particle warm opti-mization (WQPSO) is presented [58] with the aim to balance the global and localsearching abilities, and focusing on adaptive determination of the quantization para-meters for singular value decomposition.WQPSO introduce a new control parameterin QPSO algorithm, where in QPSO the best position of the particle swarm is com-puted and the current position of each particle is updated. InWQPSO the best particleposition is replaced by aweightedmean best position. Themost important problem ishow to evaluate particle importance in calculate the value of weighted best position.the particles were ranked in descendent order according to their fitness value first,then assign each particle a weight coefficient linearly decreasing with the particle’srank, that is, the nearer the best solution, and the larger its weight coefficient.

4.4.4 Hybrid Bio-inspiring Systems in Medical Watermarking

Fakhari et al. in [1] combine recently developed stochastic and bio-inspired optimiza-tion algorithms called Particle Swarm Optimization (PSO) and Genetic Algorithm(GA) to improve the performance of data hiding In designing a digital image water-marking system, it encounter two conflicting objectives which are visual qualityand robustness. This work, propose an image watermarking method based on thediscrete wavelet transform for the application of tracing colluders in clinical envi-ronment. In their work, 24 particles are used for PSO optimization. As mentionedbefore, watermark keys is embedded in all 4 frequency parts and 4 times in each,so the dimension of our problem is 16. Each particle represents 16 parameters to besearched. At this stage, the proposed algorithm educate watermarking algorithm ina way that it can resist different kinds of attacks in addition to having a high levelof perceptual quality. Similarly and by considering the important factors explainedin the previous part, GA is an efficient approach to search for optimal places, result-ing in optimum performance of our proposed digital watermarking scheme. In thiswork, 24 chromosomes are used for GA optimization. Each chromosome represents16 places to be searched with the resolution of 20 bits per variable, resulting in


a total length of 320 bits. Training the watermarking algorithm to be resistant againstdifferent kinds of attacks and having a high level of perceptual quality are equivalentto converging in acceptable time and number of iterations. To achieve this goal, thementioned 16 factors should be set properly and this is carried out by getting benefitfrom a GA. The GA block uses information from the fitness function to optimize thewatermarking algorithm and find the best possible places like PSO.

Inspired by PSO and GA, Soliman et al. in [59] introduce a hybrid d GeneticParticle Swarm Optimization (GPSO) technique by combining the advantages ofboth PSO and GA. The algorithm starts by applying PSO procedure in the searchspace and allow particles to adjust their velocity and position according to PSOequations, then in the next stepwe select a certain number of particles according toGAselection methods. The particles are matched into couples. Each couple reproducestwo children by crossover. Then some children are adjusted by applying mutationprocess. These children are used to replace their parents of the previous particles tokeep the number of particles unchanged. By combination of PSO and GA, evolutionprocess is accelerated by flying behavior and the population diversity is enhancedby genetic mechanism. The proposed GPSO is modeled to solve such optimizationproblem of medical image watermarking.

4.5 Watermarking Assessment Measures

Although digital watermarking provides better data security and authentication, how-ever, themajor drawback is distortions/visual artifacts introduced during data embed-ding which makes it difficult in detecting forged watermarks introduced by attackers.The accuracy of diagnosis and treatment of patients greatly depends on the received(watermarked) image by clinician. So, it is very important to preserve the diagnos-tic value of images. For instance, artifacts in a patient’s diagnostic image due toimage watermarking may cause errors in diagnosis and treatment, which may lead topossible life-threatening consequences [40]. For this restrictions embedding water-marks and image compressions must not distort and degrade the quality of images.Therefore, minimum Peak Signal to Noise Ratio (PSNR) of 40–50 db is advised byprevious works. More importantly, watermarks should survive the standard imageprocessing like low pass filtering (LPF) which removes noise and improves visualquality; and High Pass Filtering (HPF) that enhances the information content [1].

Watermarking systems can be characterized by a number of defining properties.The relative importance of each property is dependent on the requirements of theapplication and the role the watermark will play. In fact,even the interpretation of awatermark property can varywith the application. Thesewatermarking properties canbe used as performance criteria to evaluate watermarking schemes and provide somefavorable guidance for the design of watermarking schemes with certain applicationbackground. In this Section, some watermarking performances are introduced [62].

According to [63] there have not a complete evaluation system about digitalwatermark. Because the uniformdescription of the performance, the testmethods, the


method of attack, the standard test procedure have not been established. How to judgethe performance of a watermarking system is very important. The evaluation of awatermarking system is associated to the watermark robustness and imperceptibility.So, watermark should with stand any alteration of the watermarked image, whetherit is unintentional or malicious. This is a key requirement. Researchers currently donot appear to have found a watermarking method that is 100% robust. If the originalwatermark(s) is (are) removed the image quality must degrade so the image will beunusable according to there requirements of the application.

4.5.1 Imperceptibility or Transparency

Imperceptibility of digital watermark is also known as transparency. The watermarksignal should be imperceptible to the end user who is listening to or viewing thehost signal. This means that the perceived quality of the host signal should not bedistorted by the presence of the watermark [64]. Ideally, a typical user should not beable to differentiate between watermarked and un-watermarked signals.

Transparent assessment is divided into the subjective evaluation and objectiveevaluation [63]:

• Subjective evaluation: refers to the human visual effect as the evaluation criterion.That is given by the observer to judge the image quality. The mathematical modelscan not be used quantitatively to describe the image quality, and it is too timeconsuming. The application of subjective evaluation is very limited, so we wantto use the objective, stable mathematical model to express the image quality.

• Objective evaluation: Image objective evaluation method used mathematicalmodel and computed similarity between the image distortion and the originalimage (or distortion) and quantized evaluation scores. MSE (mean squared error)and peak signal to noise ratio (PSNR) is widely used as an evaluation criteriaof watermark imperceptibility [65]. The mean-square error between any signalsS and S∗ is defined as:

MSE(S, S∗) = 1

m2

m∑

i=1

m∑

j=1

||Si,j − S∗i,j||2 (4.1)

when S and S∗ are identical, thenMSE (S, S)= 0. A related distortion measure is thepeak signal-to-noise ratio (PSNR), measured in decibels (dB). The PSNR is definedas follows:

PSNR = 10log10(MAX2

i

MSE) = 20log10(

MAXi√MSE

) (4.2)

whereMAXi =max(S∗i, j, l≤ i, j≤m). The higher the PSNR (S, S∗), the less distortion

between S and S∗. If the signals are identical, then PSNR (S, S∗) = ∞.


Because the mean square error and peak signal to noise ratio reflect the differenceof the original image and restore the image in whole, not reflect in the local ofimages, they can not reflect the human visual characteristics. There are other varietyof evaluation methods based on the above principle, such as: assessing the qualityof enhanced images based on human visual perception, image quality assessingmodel by using neural network and support vector machine, gradient informationbased image quality assessment, image quality assessment method based on contrastsensitivity, and so on [63].

4.5.2 Robustness

The watermarking system should be robust enough to detect and extract the water-mark similar to the original one. Different types of alteration (distortions) which areknown as attacks can be performed to degrade the image quality [66]. The distor-tions are limited to those factors which do not produce excessive degradations in theimage otherwise the transformed object would be unusable. These distortions alsointroduce degradation on the performance of the watermark extraction algorithm. Totest for the robustness of the methods or a combination of the methods, an attack isperformed intentionally on a watermarked document in order to destroy or degradethe quality of the hidden watermark. According to [67] we can classify attacks intothe following groups:

• Simple attack: Simple attack aim to modify the entire cover image without extrac-tion of the watermark. Examples of such attacks include compression, addition ofnoise and editing.

• Detection-disabling or Geometric attacks: The objective of these attacks is notremoval of watermark but to damage the watermark. Watermark still exist butnot detectable. Normally, they make some geometric distortions such as zoom-ing, rotating the object, cropping or pixel permutation, shift in spatial/temporaldirection and removal/insertion etc.

• Ambiguity attacks: These attacks try to deceive the detection process throughfalse watermarked data. In this attack many additional watermarks to discredit theoriginal owner so that it is not clear thatwhichwatermark is the originalwatermark.

• Removal attacks: The objective of these attacks is to detect and then removethe embedded watermark without harming the cover media. Examples includecollusion attack, denoising, use of the conceptual cryptographic weakness of thewatermarking system, quantization, averaging, filtering, printing and scanning.

Robustness metrics include the correlation measure and bit error rate. The correla-tion measure usually use NC (normalized correlation) coefficient as the similaritymeasure of extracted watermark and the original watermark. NC equation is definedas following:


Corr =

N−1∑n=1

(w′n − w−′

)(wn − w−)√

N−1∑n=1

(w′n − w−′

)2N−1∑n=1

(wn − w−)2(4.3)

where w−′and w− indicate respectively the averages of the watermark bit sequence

w′n and wn. This correlation value measures the similarity between two strings and

varies between 1 (orthogonal sequence) and +1 (the same sequence).The trade-off between the imperceptibility and robustness is one of most serious

challenges in digital watermarking system, in particular the medical imaging.

4.6 Conclusions and Future Directions

Bio-inspiring has increasingly gained attention in watermarking research. The mainpurpose of this chapter was to present the computational intelligence and medicalwatermarking research communities some of the state-of-the-art and recent advancesin BI applications to medical image watermarking, and to inspire further researchand development on new applications and new concepts in new trend-setting direc-tions. In medical applications, because of their diagnostic value, it is very importantto maintain the quality of images. There is always a trade-off between the impercep-tibility and robustness. It is one of most serious challenges in digital watermarkingsystem, in particular the medical imaging. For instance, artifacts in a patient’s diag-nostic image due to imagewatermarkingmay cause errors in diagnosis and treatment,which may lead to possible life-threatening consequences . For this matter, the devel-opment of a new algorithm that can satisfy both imperceptibility and robustness isneeded. Improvements in performance of watermarking schemes can be obtained byseveral methods. One way is to make use of Bio-inspiring techniques. The learningcapability of all BI techniques is used to make an optimal trade-off between imper-ceptibility and robustness through effective selection of threshold. This trade-off hasmuch more importance in case of sensitive imagery such as medical imagery. BItechniques can be used also to find which are watermarks embedding places. Theseparameters are optimally varied to achieve the most suitable places for watermarkembedding that achieve best values for both imperceptibility and robustness.

Different BI techniques such as restricted Boltzmann machine, deep belief net-work, rough sets, swarm intelligence, artificial immune systems and support vectormachines, could be successfully employed to tackle various problems in watermarkembedding and extraction procedure of medical image watermarking. Also a hybridsystem of different BI techniques can be built to combine the advantages of partici-pating techniques in the hybrid system.


References

1. Fakhari, P., Vahedi, E., Lucas, C.: Protecting patient privacy from unauthorized release of med-ical images using a bio-inspired wavelet-based watermarking approach. Digit. Signal Process.21, 433–446 (2011)

2. Gunjal, B.L., Mali, S.N.: ROI based embedded watermarking of medical images for securedcommunication in telemedicine. Int. J. Comput. Commun. Eng. 68, 293–298 (2012)

3. Kallel, M., Lapayre, J.C., Bouhlel, M.S.: A multiple watermarking scheme for medical imagein the spatial domain. GVIP J. 7(1), 37–42 (2007)

4. Horng, J.-T., Wu, L.-C., Liu, B.-J., Kuo, J.-L., Kuo, W.-H., Zhang, J.-J.: An expert system toclassify microarray gene expression data using gene selection by decision tree. Expert Syst.Appl. 36(5), 9072–9081 (2009)

5. Olanrewaju, R.F., Khalifa, O., Abdul Latif, K.N.: Computational intelligence: its applicationin digital watermarking. Middle-East J. Sci. Res. (Math. Appl. Eng.). 13, 25–30 (2013)

6. Beckenkamp, F.: A component architecture for artificial neural network systems. A dissertationsubmitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy inUniversity of Constance, Computer and Information Science (2002)

7. Moore, A.W.: Reinforcement learning: a survey. J. Artif. Intell. Res. 4, 237–285 (1996)8. Haykin, S.: NeuralNetworks:AComprehensive Foundation. PrenticeHall, Upper SaddleRiver

(1999)9. Bishop, C.M.: Neural Networks for Pattern Recognition. Oxford University Press, Oxford

(1995)10. Motsinger, Alison A., Dudek, Scott M., Hahn, Lance W., Ritchie, Marylyn D.: Comparison of

neural network optimization approaches for studies of human genetics. EvoWorkshops 2006,103–114 (2006)

11. Ghosh, A., Dehuri, S.: Evolutionary algorithms for multi-criterion optimization: a survey. Int.J. Comput. Inf. Sci. 2(1), 38–57 (2004)

12. Fogel, D.B.: Evolutionary Computation: Toward a New Philosophy of Machine Intelligence,2nd edn. IEEE Press, Piscataway (1999)

13. Andres, C., Reyes, P., Sipper, M.: Evolutionary computation in medicine: an overview. Artif.Intell. Med. 19, 1–23 (1999)

14. Holland, J.H.: Adaptation in Natural and Artificial Systems. University ofMichigan Press, AnnArbor (1975)

15. Koza, J.R.: Genetic Programming. The MIT Press, Cambridge (1992)16. Back, T.: Evolutionary Algorithms in Theory and Practice: Evolution Strategies, Evolutionary

Programming, Genetic Algorithms. Oxford University Press, New York (1996)17. Fogel, L.J., Owens, A.J., Walsh, M.J.: Artificial Intelligence Through Simulated Evolution.

Wiley, New York (1967)18. Smolinski, T.G.,Milanova,M.G., Hassanien, A.E.: Applications of Computational Intelligence

in Biology: Current Trends and Open Problems, Studies in Computational Intelligence, vol.122. Springer, Berlin (2008)

19. Purves, W., Orians, G., Heller, C.: Life, the Science of Biology, Sinauer, Sunderland (1995)20. Beasley, D., Bully, D.R., Martinz, R.R.: An overview of genetic algorithms: part 1, fundamen-

tals. Univ. Comput. 15(2), 58–69 (1993)21. Juang, C.F.: A hybrid of genetic algorithm and particle swarm optimization for recurrent net-

work design. IEEE Trans. Syst. Man Cypern. Part B: Cybern. 34(2), 997–1006 (2004)22. Imade, H., Morishita, R., Ono, I., Ono, N., Okamoto, M.: A framework of grid-oriented genetic

algorithms for large-scale optimization in bioinformatics. In: The Proceeding of the 2003Congress on Evolutionary Computation, vol. 1, pp. 623–630 (2003)

23. Jiao, C.Y., Li, D.G.: Microarray image Converted database—genetic algorithm application inbioinformatics. Int. Conf. BioMed. Eng. Inform. 1, 302–305 (2008)

24. Glen, R.C., Payne, A.W.R.: A genetic algorithm for the automated generation of moleculewithin constraints. J. Comput.-Aided Mol. Des. 9, 181–202 (1995)


25. Venkatasubramanian, V., Chan, K., Caruthers, J.M.: Evolutionary design of molecules withdesired properties using the genetic algorithm. J. Chem. Inf. Comp. Sci. 35, 188–195 (1995)

26. Goldberg, D.E.: Genetic Algorithms in Search, Optimization, andMachine Learning. Addison-Wesley Publishing Corporation Inc., Reading (1989)

27. Fleischer, M.: Foundations of Swarm Intelligence: From Principles to Practice, CSHCN; TR2003–5 (2005)

28. Blum, Ch., Merkle, D.: Swarm Intelligence Introduction and Applications, Natural ComputingSeries. Springer, Berlin (2008)

29. Felix, T.S., Kumar, M.T.: Swarm Intelligence Focus on Ant and Particle Swarm Optimization.I-Tech Education and Publishing, Vienna (2007)

30. Dorigoa, M., Blumb, Ch.: Ant colony optimization theory: a survey. Theor. Comput. Sci.344(2005), 243–278 (2005)

31. Dorigoa, M., Stutzle, T.: Ant Colony Optimization. MIT Press, Cambridge (2004)32. Pant, M., Thangaraj, R.: Particle swarm optimization: performance tuning and empirical analy-

sis. Stud. Comput. Intell. 203, 101–128 (2007)33. Blum, Christian: Ant colony optimization: introduction and recent trends. Phys. Life Rev. 2(4),

353–373 (2005)34. Parpinelli, R.S., Lopes, H.S.: New inspirations in swarm intelligence: a survey. Int. J. Bio-

Inspired Comput. 3(1), 1–16 (2011)35. Kentzoglanakis, K., Poole, M.: A swarm intelligence framework for reconstructing gene net-

works: searching for biologically plausible architectures. IEEE/ACM Trans. Comput. Biol.Bioinf. 29, 358–371 (2011)

36. Das, S. et al.: Swarm intelligence algorithms in bioinformatics. Stud. Comput. Intell. 94, 113–147 (2008)

37. Kennedy, J., Eberhart, R., Shi, Y.: Swarm Intelligence, pp. 1931–1938. Morgan KaufmannAcademic Press, San Francisco (2001)

38. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of IEEE InternationalConference on Neural Networks, pp. 1942–1948 (1995)

39. Nyeem, H., Boles, W., Boyd, C.: Review of medical image watermarking requirements forteleradiology. Digit. Imaging 26(2), 326–343 (2013)

40. Dharwadkar, N.V., Amberker, B.B., Panchannavar, P.B.: Reversible Fragile Medical ImageWatermarking with Zero Distortion. In: Computer and Communication Technology (ICCCT),pp. 248–254. Allahabad, Uttar Pradesh (2010)

41. Umaamaheshvari, A., Thanushkodi, K.: Performance analysis of watermarking medicalimages. Life Sci. J. 10(1), 2653–2660 (2013)

42. Fndk, O., Babaoglu, I., Ulker, E.: A color image watermarking scheme based on artificialimmune recognition system. Expert Syst. Appl. 38, 1942–1946 (2011)

43. Aslantas, V.: A singular-value decomposition-based image watermarking using genetic algo-rithm. Int. J. Electron. Commun. 62(5), 386–394 (2008)

44. Pal, K., Ghosh, G., Bhattacharya, M.: Biomedical image watermarking in wavelet domain fordata integrity using bit majority algorithm and multiple copies of hidden information. Am. J.Biomed. Eng. 2(2), 29–37 (2012)

45. Bhatnagar, G., Raman, B.: A new robust reference watermarking scheme based on DWT-SVD.Comput. Stand. Interfaces 31(5), 1002–1013 (2009)

46. Poonkuntran, S., Rajesh, R.S., Eswaran, P.: Analysis of difference expanding method for med-ical image watermarking. In: International Symposium on Computing, Communication, andControl (ISCCC 2009), vol. 1, pp. 31–34 (2011)

47. Fotopoulos, V., Stavrinou, M.L., Skodras, A.N.: Medical Image Authentication and Self-Correction through an Adaptive Reversible Watermarking Technique. In: 8th IEEE Interna-tional Conference on Bio Informatics and Bio Engineering, Athens, pp. 1–5 (2008)

48. Kallel, I.M., Bouhlel, M.S., Lapayre, J.C.: Improved tian method for medical image reversiblewatermarking. GVIP J. 7(2), 2–5 (2007)

49. Frank, Y.S., Wu, Y.T.: Robust watermarking and compression for medical images based ongenetic algorithms. Inf. Sci. 175, 200216 (2005)


50. Motwani, M.C.: Third generation 3D watermarking: applied computational intelligence tech-niques. A dissertation submitted in partial fulfilment of the requirements for the degree ofDoctor of Philosophy in Computer Science and Engineering. University of Nevada, Reno(2011)

51. Olanrewaju, R.F., Khalifa, O.O., Hashim, A.H., Zeki, A.M., Aburas, A.A.: Forgery detectionin medical images using complex valued neural network (CVNN). Aust. J. Basic Appl. Sci.5(7), 1251–1264 (2011)

52. Oueslati, S., Cherif, A., Solaimane, B.: Adaptive image watermarking scheme based on neuralnetwork. Int. J. Eng. Sci. Technol. 3(1), 748–756 (2011)

53. Ramya, M.M., Murugesan, R.: Joint image-adaptive compression and watermarking by GA-based wavelet localization: optimal trade-off between transmission time and security. Int. J.Image Process. 6(6), 478–487 (2012)

54. Usman, I., Khan, A., Ali, A., Choi, T.S.: Reversible watermarking based on intelligent coeffi-cient selection and integer wavelet transform. J. Innov. Comput. Inf. Control 5(12), 46754682(2009)

55. Aleisa, E.A.: A secure transmission of medical images over wireless networks using intelligentwatermarking. Life Sci. J. 10(2), 2438–2444 (2013)

56. Soliman, M.M., Hassanien, A.H., Ghali, N., Onsi, H.: An adaptive watermarking approach formedical imaging using swarm intelligent. Int. J. Smart Home 6(1), 37–50 (2012)

57. Soliman, M.M., Hassanien, A.H., Onsi, H.: An adaptive medical images watermarking usingquantum particle swarm optimization. In: 35th International Conference on Telecommunica-tions and Signal Processing (TSP), pp. 735–739, Prague, Czech Republic (2012)

58. Soliman, M.M., Hassanien, A.H., Onsi, H.: An adaptive watermarking approach based onweighted quantum particle swarm optimization. Submitted to Neural Computing and Applica-tions (2013)

59. Soliman, M.M., Hassanien, A.H., Onsi, H.: The way of improving PSO performance: medicalimaging watermarking case study. In: A Joint Conference of the 8th International ConferenceonRough Sets andCurrent Trends inComputing (RSCTC2012), pp. 237–242, Chengdu (2012)

60. Sun, J., Feng, B., Xu, W.B.: Particle swarm optimization with particles having quantum behav-iour. In: IEEE Proceedings of Congress on Evolutionary Computation, pp. 325–331 (2004)

61. Benioff, P.: The computer as a physical system: amicroscopic quantummechanical hamiltonianmodel of computers as represented by turing machines. J. Stat. Phys. 22(5), 563–591 (1980)

62. Zhou, X., Wang, S., Xiong, Sh: Attack model and performance evaluation of text digital water-marking. J. Comput. 5(12), 1933–1941 (2010)

63. Zhang, X., Zhang, F., Xu, Y.: Quality evaluation of digital image watermarking, ISNN 2011,Part II, LNCS 6676, pp. 241250 (2011)

64. Gordy, J.D.: Performance evaluation of digital watermarking algorithms.A dissertation submit-ted in partial fulfilment of the requirements for the degree of Doctor of Philosophy in ComputerScience and Engineering, faculty of graduate studies. University of Calgary, Alberta (2000)

65. Abdallah E.A.: Robust digital watermarking techniques for multimedia protection. A disserta-tion submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophyin Computer Science, Concordia University, Montreal, Quebec, Canada (2009)

66. Pal, K., Ghosh, G., Bhattacharya, M.: Biomedical image watermarking in wavelet domain fordata integrity using bit majority algorithm and multiple copies of hidden information. Am. J.Biomed. Eng. 2(2), 29–37 (2012)

67. Jan, Z.: Intelligent image watermarking using genetic programming. A dissertation submittedin partial fulfilment of the requirements for the degree of Doctor of Philosophy in ComputerScience, National University of Computer and Emerging Sciences Islamabad

[intelligent systems reference library] bio-inspiring cyber security and cloud services: trends and...

Documents