DIGITAL IMAGE WATERMARKING

Transcription

1 DIGITAL IMAGE WATERMARKING Thesis submitted in the fulfillment of the Degree of Doctor of Philosophy by VIKAS SAXENA Department of Computer Science and Engineering JAYPEE INSTITUE OF INFORMATION TECHNOLOGY UNIVERSITY A-10, SECTOR-62, NOIDA, INDIA October, 2008

2 JAYPEE INSTITUE OF INFORMATION TECHNOLOGY UNIVERSITY, NOIDA, INDIA October, 2008 ALL RIGHT RESERVED

3 SCHOLAR S CERTIFICATE This is to certify that the work reported in the Ph.D. thesis entitled Digital Image Watermarking submitted at Jaypee Institute of Information Technology University, Noida, India is an authentic record of my work carried out under the supervision of Prof. J.P.Gupta. I have not submitted this work elsewhere for any other degree or diploma. (Vikas Saxena) Department of Computer Science and Engineering Jaypee Institute of Information Technology University, Noida, India October 10, 2008

4

5 SUPERVISOR S CERTIFICATE This is to certify that the work reported in the Ph.D. thesis entitled Digital Image Watermarking submitted by Vikas Saxena at Jaypee Institute of Information Technology University, Noida, India is a bonafide record of his original work carried out under my supervision. This work has not been submitted elsewhere for any other degree or diploma. (Prof. J. P. Gupta) Vice Chancellor Jaypee Institute of Information Technology University, Noida, India October 10, 2008

6

7 TABLE OF CONTENTS ABSTRACT ACKNOWLEDGEMENT LIST OF ACCRONYMS LIST OF SYMBOLS LIST OF FIGURES LIST OF TABLES Page No. vii ix xi xiii xv xix CHAPTER-1 INTRODUCTION DATA HIDING BACKGROUND STEGANOGRAPHY VS. WATERMARKING CRYPTOGRAPHY VS. WATERMARKING DIGITAL SIGNATURE VS. WATERMARKING APPLICATION AREAS OF DIGITAL WATERMARKING COPYRIGHT PROTECTION COPY PROTECTION TEMPER DETECTION BROADCAST MONITORING FINGERPRINTING ANNOTATION APPLICATIONS CHARACTERISTICS OF WATERMARKING SCHEMES TYPES OF DIGITAL WATERMARKS STRUCTURE OF THE THESIS 15 i

8 CHAPTER-2 IMAGE WATERMARKING LITERATURE SURVEY SPATIAL DOMAIN BASED WATERMARKING SCHEMES LSB BASED SCHEMES PATCH WORK BASED SCHEME CORRELATION BASED WATERMARKING SCHEMES CORRELATION BASED SCHEMES WITH 1 PN SEQUENCE CORRELATION-BASED IMAGE WATERMARKING SCHEMES 19 WITH 2PN SEQUENCES IMAGE WATERMARKING USING PRE-FILTERING CDMA BASED IMAGE WATERMARKING SCHEME OTHER SPATIAL DOMAIN BASED WATERMARKING SCHEMES TRANSFORMED DOMAIN BASED SCHEMES DFT BASED WATERMARKING SCHEMES DCT BASED WATERMARKING SCHEMES THE MIDDLE-BAND COEFFICIENT EXCHANGE SCHEME DCT-CDMA BASED IMAGE WATERMARKING DWT BASED WATERMARKING SCHEMES CDMA-DWT BASED WATERMARKING SCHEME DWT BASED BLIND WATERMARK DETECTION DWT BASED NON-BLIND WATERMARK DETECTION RECENT METHODOLOGIES PROBLEM STATEMENT FORMULATION JUSTIFICATIONS OF THE PROBLEM STATEMENT CHOSEN 40 CHPATER-3 PRELIMINARIES IMAGE ENCODING STANDARDS JPEG ENCODING 45 ii

9 3.1.2 JPEG2000 ENCODING IMAGE QUALITY MEASURES PEAK SIGNAL TO NOISE RATIO CORRELATION COEFFICIENT TEST DATA 58 CHAPTER-4 WATERMARKING OF GRAY IMAGES INTRODUCTION INCREASING THE ROBUSTNESS OF IMAGE WATERMARKING SCHEMES 62 AGAINST JPEG COMPRESSION 4.3 INCREASING THE ROBUSTNESS OF IMAGE WATERMARKING SCHEME 64 AGAINST HISTOGRAM EQUALIZATION ATTACK 4.4 DEVISING A COLLUSION ATTACK RESISTANT WATERMARKING 68 SCHEME FOR IMAGES USING DCT G, THE POLICY GENERATOR ALGORITHM E, THE WATERMARK EMBEDDING ALGORITHM D, THE WATERMARK DETECTION ALGORITHM PERFORMANCE OF THE PROPOSED SCHEME PERFORMANCE AGAINST JPEG COMPRESSION PERFORMANCE AGAINST COMMON IMAGE MANIPULATIONS COMPARATIVE STUDY WITH OTHER MECHANISMS CONCLUSION 79 CHAPTER-5 WATERMARKING of COLOR IMAGES INTRODUCTION PERFORMANCE ANALYSIS OF COLOR CHANNEL FOR DCT BASED IMAGE WATERMARKING SCHEME 5.3 DEVISING AN ICAR WATERMARKING SCHEME FOR COLORED BMP IMAGES iii

10 5.3.1 G, THE POLICY GENERATOR ALGORITHM COLOR CHANNEL SELECTION E, THE WATERMARK EMBEDDING ALGORITHM D, THE WATERMARK DETECTION ALGORITHM PERFORMANCE OF THE PROPOSED SCHEME PERFORMANCE AGAINST JPEG COMPRESSION PERFORMANCE AGAINST COMMON IMAGE MANIPULATIONS COMPARATIVE STUDY RESULTS WITH OTHER SCHEMES CONCLUSION 96 CHAPTER-6 WATERMARKING OF JPEG IMAGES INTRODUCTION DEVELOPMG AN ICAR WATERMARKING ALGORITHM FOR JPEG 97 IMAGES G, THE POLICY GENERATOR ALGORITHM COLOR CHANNEL SELECTION E, THE WATERMARK EMBEDDING ALGORITHM D, THE WATERMARK DETECTION ALGORITHM PERFORMANCE OF THE PROPOSED SCHEME COLOR CHANNEL SELECTION AND PERFORMANCE AGAINST 105 JPEG COMPRESSION PERFORMANCE AGAINST IMAGE MANIPULATIONS COMPARATIVE STUDY WITH SIMILAR, STATE-OF-THE-ART SCHEMES A DWT BASED WATERMARKING SCHEME FOR JPEG IMAGES EXPLORATION OF DWT DOMAIN ISSUES IN USING DWT BACKGROUND OF THE PROPOSED SCHEME DUAL WATERMARKING 115 iv

11 6.3.4 THE DWT BASED WATERMARKING P, THE POLICY G, THE POLICY GENERATOR ALGORITHM E, THE WATERMARK EMBEDDING ALGORITHM D, THE WATERMARK DETECTION ALGORITHM THE DCT BASED WATERMARKING RESULTS THE VALUE OF T PERFORMANCE AGAINST JPEG COMPRESSION PERFORMANCE AGAINST COMMON ATTACKS AND IMAGE 127 MANIPULATIONS COMPARATIVE STUDY WITH DCT BASED SCHEMES COMPARATIVE STUDY WITH DWT BASED SCHEMES CONCLUSION 130 CHAPTER-7 RESULTS AND CONCLUSION SUMMARY MAIN CONTRIBUTIONS AND HIGHLIGHTS OF THE RESULTS FUTURE WORK 132 REFERENCES 135 LIST OF AUTHOR S PUBLICATION 147 SYNOPSIS v

12 vi

13 ABSTRACT Watermarking has been invoked as a tool for the protection of Intellectual Property Rights (IPR) of multimedia contents. Because of their digital nature, multimedia documents can be duplicated, modified, transformed, and diffused very easily. In this context, it is important to develop a system for copyright protection, protection against duplication, and authentication of contents. For this, a watermark is embedded into the digital data in such a way that it is indissolubly tied to the data itself. Later on, such watermark can be extracted to prove ownership to trace the dissemination of the marked work through the network, or simply to inform users about the identity of the rights-holder or about the allowed use of data. This thesis deals the developing the watermarking schemes for digital images stored in both, spatial and transformed domain. In this thesis we mainly focus on the Discrete Cosine Transform (DCT) based development. To prove its commercial usability, we take special care so that at least one attack, having huge financial implications, can be sustained due to the in-built capacity of the watermarking scheme. Apart from this, since JPEG is the most commonly used image format over WWW, we pay special attention to robustness against JPEG compression attack. Apart from developing watermarking schemes, we also discuss the selection of color channel to be used to carry the watermark data based on the attack that may occur most commonly on the watermarked images. We propose to increase the robustness against some attacks by preprocessing the images. In this thesis, we also present a correlation between the performance of the watermarking scheme against some attacks and the original image characteristics. All presented watermarking schemes are robust against common image manipulations and attacks. vii

14 viii

15 ACKNOWLEDGEMENT I am greatly indebted to my supervisor Prof. J. P. Gupta for his valuable technical guidance and moral support through out this work. Without his support this thesis would have not been completed. I would also like to thank to Prof S.L Maskara, Prof Sanjay Goel and faculty members of the department who always enlightened me by sharing their research experiences to accomplish the quality work. My mother provided me all support I needed to complete this thesis and other family members specially my wife also helped me a lot in getting me this far. Vikas Saxena Department of Computer Science Engineering and Information Technology Jaypee Institute of Information Technoogy University Noida, India ix

16 x

17 LIST OF ACCRONYMS CC CDMA DCT DFT DWT EBCOT EZW FFT HH HL HVS ICAR IPR JND JPEG LH LL LSB MBCE MSE PN PSNR PSW REL RGB SPIHT SVD VQ Correlation Coefficient Code Division Multiple Access Discrete Cosine Transform Discrete Fourier Transform Discrete Wavelet Transform Embedded Block Coding with Optimized Truncation Embedded Zero-tree Wavelet Fast Fourier Transform High-High Band of DWT High-Low Band of DWT Human Visual System Inherently Collusion Attack Resistant Intellectual Property Right Just Noticeable Distortion Joint Photographic Expert Group Low-High Band of DWT Low-Low Band of DWT Least Significant Bit Middle Band Coefficient Exchange Mean Square Error Pseudo-random noise Peak Signal to Noise ration Perceptually Shaped Watermarking Run Length Encoding Red Green Blue Set Partitioning In Hierarchical Trees Singular Value Decomposition Vector Quantization xi

18 xii

19 LIST OF SYMBOLS D E FH FL FM G K P Pi Q S Sr T W Wi X Xi Watermark detection algorithm Watermark embedding algorithm High frequency region in an 8 x 8 DCT Low frequency region in an 8 x 8 DCT Middle frequency region in an 8 x 8 DCT Policy generator algorithm Watermark strength parameter Policy An instance of a policy JPEG quantization factor Watermark logo converted into string of 0 s and 1 s A single bit of S Watermark strength parameter Watermark logo A single bit of the watermark data Original cover image An instance of the cover image xiii

20 xiv

21 LIST OF FIGURES Figure No. Caption Page No. 1.1 Watermark on the bank currency note Various classifications of watermarking Image watermark embedding scheme Image watermark detection scheme FIR Edge Enhancement Pre-Filter A General Frequency domain based watermarking model as presented by Cox Frequency regions in 8 x 8 DCT JPEG Quantization matrix Scale and 2-Scale 2-Dimensional Discrete Wavelet Transform The Targeted types of to be developed watermarking schemes JPEG Compression Scheme An example sub image Example sub image after subtracting 128 from each pixel DCT of sub image shown in Figure JPEG Quantization matrix DCT values after quantization JPEG Decompression Scheme DCT values regenerated in decompression (a) Sub image pixel values (still shifted down by 128) (b) Decompressed sub image pixel values Error matrix for example sub image Test images of Lena, Mandrill, Pepper and Barbara (Gray) Test images of Lena, Mandrill, Pepper and Goldhill (Colored) Watermark logo used in the proposed schemes 59 xv

22 4.1 (a) Extracted watermark logos from test images of Lena, Mandrill and Pepper by applying DCT based scheme (b) Extracted watermark logos from test images of Lena, Mandrill and Pepper by applying DWT based scheme (a) Extracted watermark logos from test images of Lena, Mandrill, Pepper and Barbara by applying DCT based scheme (b) Extracted watermark logos from test images of Lena, Mandrill, Pepper and Barbara by applying DWT based scheme Extracted logos from original image (left) and transformed image (right) of Lena, Mandrill, Pepper and Barbara s (Top to Bottom) histogram equalized images (By applying DCT based scheme) Swapping of 4 pairs to hide 0 or 1 in conjunction with low frequency values Extracted watermark logos after JPEG compression at Q = 20 from watermarked Lena, Mandrill and Pepper images Extracted watermark logos from Lena s image after Horizontal flipped, scaled, brightness /contrast adjusted and Noising (Left to Right, Top to bottom) Percentage decrease in quality of extracted watermark with respect to JPEG quality factor Recovered watermarks for Lena.bmp after jpeg attack at Q = Watermarked test images keeping T = Extracted watermark from watermarked Lena, Mandrill and Pepper images respectively at T = Recovered logos from attacked images Extracted logos using proposed scheme from highly compressed watermarked test images Watermarked test images generated by keeping T = Extracted watermark logos from watermarked Lena, Mandrill, Pepper and Goldhill test images respectively at T = xvi

23 6.3 Goldhill test image after hiding the watermark logo and the recovered logo at T = Extracted logos from attacked watermarked images Comparison of correlation coefficients at Q = Comparison of correlation coefficients at Q = D Haar DWT An example of 2 consecutive DWT blocks An example of 2 consecutive DWT blocks Watermark embedding approach The watermark logo Graph of the values shown in Table Extracted logos from Lena, Mandrill and Pepper s test images The extracted logos using DWT based method The extracted logos using DCT based method Extracted logos from highly compressed JPEG images Extracted watermark logos after applying common attacks 128 xvii

24 xviii

25 LIST OF TABLES Table No. Caption Page No. 4.1 PSNR (in decibel) of extracted watermark logo from JPEG compressed (Q = 20) watermarked image PSNR of extracted logos from attacked test images PSNR of extracted watermarks after JPEG compression PSNR of Extracted watermark from JPEG compressed watermark test images PSNR of extracted watermark from attacked watermarked test images PSNR of extracted watermark logos after JPEG compression PSNR of extracted watermark logo from watermarked test images after attacks PSNR values of extracted logos from highly compressed watermarked test images using various schemes SD values of color channels for test images PSNR and CC of extracted logo by using BLUE channel for all images PSNR and CC of extracted logo by using BLUE and GREEN channels for images CC of the extracted logos PSNR of watermarked image and CC of extracted logo for various values of T Revised Table CC of extracted logos from JPEG2000 attacked images Decrement in the PSNR values after the application of DCT based scheme CC values of the extracted watermark logos recovered by both recovery methods 125 xix

26 6.10 CC of extracted logo from highly compressed jpeg image using DCT based recovery CC of the extracted watermark logos Comparison of CC of Extracted logos from JPEG compressed (Q = 10) watermarked images Comparison of CC of Extracted logos from JPEG compressed (Q = 5) watermarked images 129 xx

27 CHAPTER 1 INTRODUCTION The growth of high speed computer networks and World Wide Web (WWW) have explored means of new business, scientific, entertainment and social opportunities in the form of electronic publishing and advertising, massaging, real-time information delivery, data sharing, collaboration among computers, product ordering, transaction processing, digital repositories and libraries, web newspapers and magazines, network video and audio, personal communication and lots more. The cost effectiveness of selling softwares in the form of digital images and video sequences by transmission over WWW is greatly enhanced due to the improvement in technology. We know that one of the biggest technological events of the last two decades was the invasion of digital media in an entire range of everyday life aspects. Digital data can be stored efficiently and with a very high quality, and it can be manipulated very easily using computers. Furthermore, digital data can be transmitted in a fast and inexpensive way through data communication networks without losing quality. Digital media offer several distinct advantages over analog media. The quality of digital audio, images and video signals are higher than that of their analog counterparts. Editing is easy because one can access the exact discrete locations that need to be changed. Copying is simple with no loss of fidelity. A copy of a digital media is identical to the original. With digital multimedia distribution over World Wide Web, authentications are more threatened than ever due to the possibility of unlimited copying. The easy transmission and manipulation of digital data constitutes a real threat for information creators, and copyright owners want to be compensated every time their work is used. Furthermore, they want to be sure that their work is not used in an improper way (e. g. modified without their permission). For digital data, copyright enforcement and content verification are very difficult tasks. One solution would be to restrict access to the data using some encryption techniques. However, 1

28 encryption does not provide overall protection. Once the encrypted data are decrypted, they can be freely distributed or manipulated. Unauthorized use of data creates several problems. For example, if we visit wallpaper.com, we observe that all the wallpaper images are created by the owners, which are their Intellectual Property Right (IPR). Any user can download the wallpapers. Now, consider that a user downloads the images and posts those images (either after modifying or original) on his/her website. Three issues may arise in this situation: 1) How will the owner of wallpaper.com know that there is one more web server on WWW posting their wallpapers? 2) If the owner knows about this fact, where shall he go to make a complaint? 3) The last but very important issue is that even if first two problems are resolved, how the owner will prove the ownership on the wallpaper images posted on another server? The first issue is related to network technologies and involves issues like web crawler and pattern matching etc. Second issue is related to the international copyright laws and is another very tricky issue. This thesis does not deal with these 2 issues. This thesis covers the third issue, the authentication i.e. how to prove the ownership? The above problem can be solved by hiding some ownership data into the multimedia data, which can be extracted later to prove the ownership. This idea is implemented in bank currency notes embedded with the watermark which is used to check the originality of the note. The same watermarking concept may be used in multimedia digital contents for checking the authenticity of the original content. To begin with a quick background of watermarking, first we present the history of data hiding and related terminologies. Then, we will move on to a discussion on the 2

29 watermarking, requirements that watermarking system must meet, types of the watermarking, applications and then various attacks on a watermarking system. 1.1 DATA HIDING BACKGROUND The solution of the problem discussed above seems to lie in a technique that dates back to ancient Egypt and Greece: data hiding or steganography. Steganography deals with the methods of embedding data within a medium (host or cover medium) in an imperceptible way. All forms of digital data (still images, audio, video, text documents and multimedia documents) can be used as a cover medium for information hiding. The history of steganography goes all the way back to the 5 th Century. The earliest known writings about steganography were by the Greek historian Herodotus. The historian relates how a slave had a message tattooed on his head by Histiaeus who was trying to get a message to his son-in-law Aristagoras. Once the slaves hair was long enough to cover the message he was sent to Aristagoras in the city of Miletus [92]. Stegnography has been used in many different ways. The simplest was the use of invisible inks that a person could use to send a message to another person without anyone else knowing. Different forms of invisible ink were used to conceal messages. Some of the more common forms of invisible ink have been lemon juice, milk, and urine to name a few. If someone wanted to conceal a message, he would simply write a message, using one of these inks, on a sheet of paper that already had something written on it. The person receiving the message would then hold the paper over a flame and the transparent message would appear. Image stegnography was done during the early twentieth century. During the Boer War in South Africa, the British were using Lord Robert Baden-Powell as a scout. He was scouting the Boer artillery bases mapping their positions. He took his maps and converted them into pictures of butterflies with certain markings on the wings that were actually the enemies positions [92]. 3

30 During World War II, Nazis introduced a new concept in espionage, which was called the microdot. This simple device could conceal a full typewritten page within the size of a common period. A microdot could hold valuable information such as charts, diagrams and drawings. Watermark symbol is added here to prove the originality Figure 1.1: Watermark on the bank currency note Thus, stegnography is an area which is, more or less, a Hide-&-Seek game. Some important data or information is hidden in another medium. The cover medium has no relationship with the data or information hidden. Data or information which is hidden is not encrypted also. The key issue in a stegnography system becomes that no one should suspect that a particular medium is carrying any hidden data or information. We can extend the stegnography concept for the authentication of digital multimedia data. Digital multimedia data which has to be protected is now the cover medium and then we can hide the copyright data into it. In this case, there will be two major requirements as follows: 1) Imperceptibility: After hiding the copyright data, cover medium should not be affected, and 2) Robustness: No body should be able to remove the data without affecting the cover medium. 4

31 The copyright data may be termed as digital watermark data. This area of application of stegnography is known as Digital Watermarking. Therefore, digital watermark is a message/data/information which is embedded into digital content (audio, video, images or text) that can be detected or extracted later. Such message/data/information mostly carries the copyright or ownership information of the content. The process of embedding digital watermark information into digital content is known digital watermarking. Before moving further in this discussion, we must first understand the difference of the digital watermarking with other related terms like stegnography, cryptography and digital signature STEGANOGRAPHY VS WATERMARKING Watermarking is the subset of Stegnography. In Stegnography, data which is hidden has no relationship with the cover medium and the requirement from such a system is that no suspicion should arise that a medium is carrying any hidden data. In watermarking, unlike stegnography, the data which is hidden has relationship with the cover medium data. Data hidden is the ownership data of the cover medium and there is no issue like suspecting that a particular medium is carrying some copyright data. As the purpose of stegnography is to have a covert communication between two parties i.e. existence of the communication is unknown to a possible attacker, and a successful attack shall detect the existence of this communication. On the contrary, watermarking, as opposed to stegnography, requires a system to be robust against possible attacks. Other requirements of watermarking are entirely different from stegnography and these are discussed in detail in Section CRYPTOGRAPHY VS. WATERMARKING Cryptography can be defined as the processing of information into an unintelligible form known as encryption, for the purpose of secure transmission. Through the use of a key, the receiver can decode the encrypted message (the process known as decryption) to retrieve 5

32 the original message. So, cryptography is about protecting the contents of the message. But as soon as the data is decrypted, all the in-built security and data is ready to use. Cryptography "scrambles" a message so that it can not be understood by unauthorized user. This does not happen in watermarking. Neither the cover medium nor the copyright data changes its meaning. Rather, copyright data is hidden to give the ownership information of the medium in which it is hidden DIGITAL SIGNATURE VS. WATERMARKING Digital signatures, like written signatures, are used to provide authentication of the associated input, usually called a "message. Digital signature is an electronic signature that can be used to authenticate the identity of the sender of a message or the signer of a document, and possibly to ensure that the original content of the message or document that has been sent is unchanged. Digital signatures are easily transportable, cannot be imitated by someone else, and can be automatically time-stamped. The ability to ensure that the original signed message arrived means that the sender cannot easily repudiate it later. A digital signature can be used with any kind of message, whether it is encrypted or not, simply so that the receiver can be sure of the sender's identity and that the message arrived intact. A digital signature is apart from the protected message, whereas a digital watermark is inside a multimedia message. Both, digital signature and watermarking protect integrity and authenticity of a document. Digital signature system is vulnerable to distortion but a watermark system has to tolerate a limited distortion level. So, to conclude, Watermarking is adding ownership information in multimedia contents to prove the authenticity. This technology embeds a data, an unperceivable digital code, namely the watermark, carrying information about the copyright status of the work to be protected. Continuous efforts are being made to device efficient watermarking schema but techniques proposed so far do not seem to be robust to all possible attacks and multimedia data processing operations. The sudden increase in watermarking interest is most likely due to the increase in concern over IPR. Today, digital data security covers such topics as access control, authentication, and copyright protection for still images, audio, video, and 6

33 multimedia products. A pirate tries either to remove a watermark to violate a copyright or to cast the same watermark, after altering the data, to forge the proof of authenticity. Generally, the watermarking of still images, video, and audio demonstrate certain common fundamental concepts APPLICATION AREAS OF DIGITAL WATERMARKING Watermarking techniques may be relevant in the following application areas [26]: COPYRIGHT PROTECTION The primary use of watermarking is where an organization wishes to assert its ownership of copyright for digital objects. This application is of great interest to big media organizations, and of some interest to other vendors of digital information, such as news and photo agencies. These applications require a minimal amount of information to be embedded, coupled with a high degree of resistance to signal modification (since they may be subjected to deliberate attack). For example, now a days, a news channel AAJ-TAK is showing the animal s clips (which are already shown on Discovery Channel) by hiding the Discovery channel s logo on the video clips. As per the law, The AAJ-TAK should show the curtsey-sign and should pay the copyright fee to the Discovery channel. In such cases, There is a strong need of watermarking as once the digital data is broadcasted, any body else can start selling it without paying the IPR value to its owner COPY PROTECTION Watermarking can be used as a strong tool to prevent illegal copying. For example, if an audio CD has a watermark embedded into it, then any of the system (Hardware like DVD, or software) can not make a copy of it, and even if it copies, the watermark data will not get copied to new duplicate audio CD. Now the duplicate CD can be easily found because it does not have watermark data. Some schemes have attempted to satisfy more complex copy protection requirements. An early example is the Serial Copy Management System (SCMS), introduced in the 1980s, which enabled a user to make a single digital audio tape of a recording they had purchased but prevented the recording of further copies (i.e. second 7

34 generation) from that first copy. The scheme failed ultimately because not all manufacturers of consumer equipment were prepared to implement the scheme in their products TEMPER DETECTION In this application area, it is necessary to assure that the origin of a data object is demonstrated and its integrity is proved. One example of temper detection is photographic forensic information which may be presented as evidence in the court. Given the ease with which digital images can be manipulated, there is a need to provide proof that an image has not been altered. Such a mechanism could be built into a digital camera [29]. For example, if a cop s camera catches an over speeding vehicle then when proving the driver guilty in front of the judge, the accused may claim that the video presented in the court is tempered and the car shown in the video does not belong to him. A watermarking system which is embedded in digital cameras may help to resolve the issue. If somebody tries to temper the data, the watermark will get destroyed indicating that the data is tempered. In our country, a well-known example is the Tahalka-Scam BROADCAST MONITORING There are several types of organizations and individuals interested in monitoring the broadcast of their interest. For example, advertisers want to ensure that they receive the exact airtime that they have purchased from broadcasting firms. Musicians and actors want to ensure that they receive accurate royalty payments for broadcasts of their performances and copyright owners want to ensure that their property is not illegally rebroadcast by pirate stations. In 1997, a scandal broke out in Japan regarding television advertising. At least two stations had been routinely overbooking air time. Advertisers were paying for thousands of commercials that were never aired [16]. The practice had remained largely undetected for over twenty years because there were no systems in place to monitor the actual broadcast of advertisements. 8

35 This broadcast monitoring can be implemented by putting a unique watermark in each video or sound clip prior to broadcast. Automated monitoring stations can then receive broadcasts and look for these watermarks identifying when and where each clip appears FINGERPRINTING If monitoring and owner identification applications place the same watermark in all copies of the same content, it may create a problem. If out of n number of legal buyers of a content, one starts selling the contents illegally, it may be very difficult to catch who is redistributing the contents without permission. Allowing each copy distributed to be customized for each legal recipient can solve this problem. This capability allows a unique watermark to be embedded in each individual copy. Now, if the owner finds an illegal copy, he can find out who is selling his contents by finding the watermark which belongs to only singly legal buyer. This particular application area is known as fingerprinting. This is potentially valuable both as a deterrent to illegal use and as a technological aid to investigation ANNOTATION APPLICATIONS In this applications area, watermarks convey object-specific information ( feature tags or captions ) to users of the object. For example, patient identification data can be embedded into medical images. These applications require relatively large quantities of embedded data. While there is no need to protect against deliberate tampering. Normal use of the data object may involve such transformations as image cropping or scaling and will require the use of a technique that is resistant to those types of modification. For more details of various watermarking applications, one may refer [20]. 9

36 1. 3 CHARACTERISTICS OF WATERMARKING SCHEMES An effective watermarking scheme should have the following characteristics: 1) Imperceptibility: In terms of watermarking, imperceptibility means that after inserting the watermark data, cover medium should not alter much. In other words, the presence of the watermark data should not affect the cover medium being protected. If a watermarking scheme does not ensure this requirement, it may happen that after inserting a watermark data in a cover medium (say an image), image quality may alter which the owner of the image will never like that a protecting mechanism modifies his work. 2) Robustness: Robustness of the watermark data means that the watermark data should not be destroyed if someone performs the common manipulations as well as malicious attacks. It is more of a property and also a requirement of watermarking and its applicability depends on the application area. 3) Fragility: Fragility means that the watermark data is altered or disturbed up to a certain extent when someone performs the common manipulations & malicious attacks. Some application areas like temper detection may require a fragile watermark to know that some tempering is done with his work. Some application may require semi-fragility too. The semi-fragile watermark comprises a fragile watermark component and a robust watermark component i.e. semi-fragile watermarks are robust to some attacks but fragile to others attacks. 4) Resilient to common signal processing: The watermark should be retrievable even if common signal processing operations are applied to the watermarked cover medium data. These operations include digital-to-analog and analog-to-digital conversion (i.e. taking the printout of an image and then scan it to create another digital copy of the image), re-sampling, re-quantization (including dithering and recompression), and common signal enhancements such as image contrast, brightness and color adjustment, or audio bass and treble adjustment, high pass and low pass filtering, histogram equalization of an image and format conversion (BMP image to JPEG image, MPEG movie to WMV movie, mp3 song to mp4 etc.) 10

37 5) Resilient to common geometric distortions (image and video data): Watermarks in image and video data should also be immune from geometric image operations such as rotation, translation, cropping and scaling. This property is not required for audio watermarking. 6) Robust to subterfuge attacks (collusion and forgery): In addition, the watermark should be robust to collusion attack. Multiple individuals, who possess a watermarked copy of the data, may collude their watermark copies to destroy the watermark presence and can generate a duplicate of the original copy. Further, if a digital watermark is to be used in litigation, it must be impossible for colluders to combine their images to generate a different valid watermark. 7) Unambiguousness: Retrieval of the watermark should unambiguously identify the owner. Furthermore, the accuracy of owner identification should not degrade much in the case of an attack. The Unzign and Stirmark [97] have shown remarkable success in removing data embedded by commercially available programs. Watermarking of watermarked image (re-watermarking) is also a major threat [97]. 1.4 TYPES OF DIGITAL WATERMARKS Prof. S. Mohanty presents a very good classification of watermarking areas in his paper [62]. We can classify the types of watermarking based on the cover medium, embedding domain, perception and application domain. Figure 1.2 shows the various classifications of watermarking. Based on their embedding domain, watermarking schemes can be classified as follows: 1) Spatial Domain: The watermarking system directly alters the main data elements (like pixels in an image) to hide the watermark data. 2) Transformed Domain: The watermarking system alters the frequency transforms of data elements to hide the watermark data. This has proved to be more robust than the spatial domain watermarking. 11

38 3) Feature Domain: The watermarking system takes into account the region, boundary and object characteristics. It presents better detection and recovery from attacks. Figure 1.2: Various classifications of watermarking Watermarking techniques can also be divided into four categories, according to the type of document to be watermarked, as follows. 1) Image Watermarking: Figure 1.3 and 1.4 represent the general scheme of an image watermarking, embedding and decoding (specifically key based, invisible and fragile) 12

39 system respectively. E represents the watermarking embedding algorithm and D represents the watermarking decoding algorithm. 2) Other types of watermarking, according to the type of document to be watermarked are Video Watermarking, Audio Watermarking and Text Watermarking. Figure 1.3: Image watermark embedding scheme Figure 1.4: Image watermark detection scheme According to the human perception, the digital watermarks can be divided into 4 different types: Visible watermark, Invisible-Robust watermark, Invisible-Fragile watermark, and dual watermark. Visible watermark is a secondary translucent overlaid into the primary image. The watermark appears visible to a casual viewer on a careful inspection. The invisiblerobust watermark is embedded in such a way that alternations made to the pixel value are perceptually not noticed and it can be recovered only with appropriate decoding mechanism. The fragile watermark is embedded in such a way that any manipulation or modification of the image would alter or destroy the watermark. Dual watermark is a combination of a visible and an invisible watermark [8]. 13

40 According to application domain, Source-based watermarks are desirable for ownership identification or authentication where a unique watermark identifies the owner. A sourcebased watermark could be used for authentication and to determine whether a received image or other electronic data has been tampered. The watermark could also be destination based where each distributed copy gets a unique watermark identifying the particular buyer. The destination based watermark could be used to trace the buyer in the case of illegal reselling. This is used in fingerprinting. A watermark is said private if only authorized readers can detect it. In other words, in private watermarking, a mechanism is envisaged that makes it impossible for unauthorized people to extract the watermark. A watermarking algorithm is said blind if it does not resort to the comparison between the original non-marked and the marked document to recover the watermark. Conversely, a watermarking algorithm is said non-blind if it needs the original data to extract the information contained in the watermark. The definition of invertible and quasi-invertible is more technical and can be given as follows [2]: If E is the Embedding algorithm, D is detection algorithm, C δ is Comparator function, I is original cover image, Î is watermarked image, J is recovered attacked image, S is watermark signal and S is extracted watermark data, then: 1) E (I, S) = Î 2) D (J, I) = S or D (J) = S 3) Comparator C δ : A watermarking scheme (E, D, C δ ) is invertible if: 1) Inverse mapping E -1 does exist such that E -1 (Î) = (Î, S ) &E (Î, S ) = Î; 2) E -1 is computational feasible; 14

41 3) S is an allowed watermark; 4) Î and Î are perceptually similar; and 5) Comparator output C δ (D (Î, Î ), S ) = 1 Otherwise the watermarking scheme is non-invertible. A watermarking scheme (E, D, C δ ) is quasi-invertible if: 1) Properties for invertible watermarking schemes apply; 2) Only difference E (Î, S ) = Î Î; and 3) Î and Î perceptually similar. Otherwise the watermarking scheme is non-quasi-invertible. A Non-invertible scheme can be quasi-invertible and Non-quasi-invertibility implies non-invertibility. 1.5 STRUCTURE OF THE THESIS This thesis comprises of the following chapters: Chapter 2 describes the image watermarking literature survey and problem statement. Chapter 3 describes the preliminaries (like background of JPEG compression, 2D DCT and DWT, image quality parameter, some standard watermarking techniques which are used to compare the performances of the proposed techniques etc and test images data). The watermarking techniques for gray images have been proposed in Chapter 4. Chapter 5 describes the proposed watermarking techniques and issues related to colored BMP images. In Chapter 6, the proposed watermarking techniques for JPEG images have been given. Finally the summary of results, conclusions and future work is given in Chapter 7 followed by references, author s publications and synopsis at the end. 15

42 16

43 CHAPTER-2 IMAGE WATERMARKING LITERATURE SURVEY Within the field of watermarking, image watermarking particularly has attracted lot of attention in the research community. Most of the research work is dedicated to image watermarking as compared to audio and video. There may be 3 reasons for it. Firstly, because of ready availability of the test images, secondly because it carries enough redundant information to provide an opportunity to embed watermarks easily, and lastly, it may be assumed that any successful image watermarking algorithm may be upgraded for the video also. Images are represented/stored in spatial domain as well as in transform domain. The transform domain image is represented in terms of its frequencies; whereas, in spatial domain it is represented by pixels. In simple terms, transform domain means the image is segmented into multiple frequency bands. To transfer an image to its frequency representation, we can use several reversible transforms like Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), or Discrete Fourier Transform (DFT). Each of these transforms has its own characteristics and represents the image in different ways. Watermarks can be embedded within images by modifying these values, i.e. the transform domain coefficients. In case of spatial domain, simple watermarks could be embedded in the images by modifying the pixel values or the Least Significant Bit (LSB) values. However, more robust watermarks could be embedded in the transform domain of images by modifying the transform domain coefficients. In 1997 Cox et al. presented a paper Secure Spread Spectrum Watermarking for Multimedia [19], one of the most cited paper (cited 2985 times till April 2008 as per Google Scholar search), and after that most of the research work is based on this work. Even though spatial domain based techniques can not sustain most of the common attacks like compression, high pass or low pass filtering etc., researchers present spatial domain based schemes. Firstly, brief 17

44 introductions of some classical well-known spatial domain based schemes are being given as follows [19]: 2.1 SPATIAL DOMAIN BASED WATERMARKING SCHEMES LSB BASED SCHEMES In their paper, Macq and Quisquater [60] briefly discussed the issue of watermarking digital images as part of a general survey on cryptography and digital television. The authors provided a description of a procedure to insert a watermark into the least significant bits of pixels located in the vicinity of image contours. Since it relies on modifications of the least significant bits, the watermark is easily destroyed. Further, their method is restricted to images, in that it seeks to insert the watermark into image regions that lie on the edge of contours. Rhoads [79] described a method that adds or subtracts small random quantities from each pixel. Addition or subtraction is determined by comparing a binary mask of bits with the LSB of each pixel. If the LSB is equal to the corresponding mask bit, then the random quantity is added, otherwise it is subtracted. The watermark is subtracted by first computing the difference between the original and watermarked images and then by examining the sign of the difference, pixel by pixel, to determine if it corresponds to the original sequence of additions and subtractions. This method does not make use of perceptual relevance, but it is proposed that the high frequency noise be prefiltered to provide some robustness to lowpass filtering. This scheme does not consider the problem of collusion attacks PATCH WORK BASED SCHEMES Another, well known spatial domain based scheme is patchwork-based technique given by Bender et al. [7]. They described two watermarking schemes. The first is a statistical method called patchwork. Patchwork randomly chooses pairs of image points, and increases the brightness at one point by one unit while correspondingly decreasing the brightness of another point. The second method is called texture block coding wherein 18

45 a region of random texture pattern found in the image is copied to an area of the image with similar texture. Autocorrelation is then used to recover each texture region. The most significant problem with this scheme is that it is only appropriate for images that possess large areas of random texture. The scheme could not be used on images of text. Other Patchwork based algorithm can be found in [110, 124] CORRELATION BASED WATERMARKING SCHEMES The most straightforward way to add a watermark to an image in the spatial domain is to add a pseudorandom noise pattern to the luminance values of its pixels. Many methods are based on this principle [6, 11, 27, 33-34, 53, 68, 70, 91, 95, ] CORRELATION BASED SCHEMES WITH 1 PN SEQUENCE: A well known technique for watermark embedding is to exploit the correlation properties of additive pseudo-random noise patterns as applied to an image [42, 52]. A Pseudo-random Noise (PN) pattern W (x, y) is added to the cover image I (x, y), according to the Equation 2.1 given below: I w ( x, y) = I( x, y) + k * W ( x, y) (2.1) In Equation 2.1, k denotes a gain factor and I W the resulting watermarked image. Increasing k increases the robustness of the watermark at the expense of the quality of the watermarked image. To retrieve the watermark, the same pseudo-random noise generator algorithm is seeded with the same key, and the correlation between the noise pattern and possibly watermarked image is computed. If the correlation exceeds a certain threshold T, the watermark is detected, and a single bit is set. This method can easily be extended to a multiple-bit watermark by dividing the image into blocks and performing the above procedure independently on each block CORRELATION-BASED IMAGE WATERMARKING SCHEMES WITH 2PN SEQUENCES: This basic algorithm, as given in previous section, can be improved in a number of ways. First, the notion of a threshold being used for determining a logical 19

46 1 or 0 can be eliminated by using two separate pseudo-random noise patterns. One pattern is designated a logical 1 and the other a logical 0. The above procedure is then performed once for each pattern, and the pattern with the higher resulting correlation is used. This increases the probability of correct detection, even after the image has been subject to attack [42, 52] IMAGE WATERMARKING USING PRE-FILTERING: We can further improve the basic algorithm by pre-filtering the image before applying the watermark. If we can reduce the correlation between the cover image and the PN sequence, we can increase the immunity of the watermark to additional noise. By applying the edge enhancement filter shown below in Figure 2.1, the robustness of the watermark can be improved with no loss of capacity and very little reduction of image quality [42, 52]. F edge = Figure 2.1: FIR Edge Enhancement Pre-Filter CDMA BASED IMAGE WATERMARKING SCHEME Rather than determining the values of the watermark from blocks in the spatial domain, we can employ CDMA spread-spectrum schemes to scatter each of the bits randomly throughout the cover image, thus increasing capacity and improving resistance to cropping. The watermark is first formatted as a long string rather than a 2D image. For each value of the watermark, a PN sequence is generated using an independent seed. These seeds could either be stored or themselves generated through PN methods. The summation of all of these PN sequences represents the watermark, which is then scaled and added to the cover image [42, 52]. To detect the watermark, each seed is used to generate its PN sequence which is then correlated with the entire image. If the correlation is high, that bit in the watermark is set to 1, otherwise a 0. The process is then repeated for all the values of the watermark. 20

47 CDMA improves on the robustness of the watermark significantly but it requires more computation OTHER SPATIAL DOMAIN BASED WATERMARKING SCHEMES In [104], a method that embeds a binary watermark image in the spatial domain is proposed. A spatial transform that maps each pixel of the watermark image to a pixel of the host image, is used. Chaotic spread of watermark image pixels in the host image is achieved by toral automorphisms. For watermark embedding, the intensity of the selected pixels is modified by an appropriate function that takes into account neighborhood information in order to achieve watermark robustness to modifications. For detection, a suitable function is applied on each of the watermarked pixels to determine the binary digit (0 or 1) that has been embedded. The inverse spatial transform is then used to reconstruct the binary watermark image. In the method proposed in [69], the image is split into two random subsets A and B and the intensity of pixels in A is increased by a constant embedding factor k. Watermark detection is performed by evaluating the difference of the mean values of the pixels in subsets A and B. This difference is expected to be equal to k for a watermarked image and equal to zero for an image that is not watermarked. Hypothesis testing can be used to decide for the existence of the watermark. The above algorithm is vulnerable to lowpass operations. Extensions to above algorithm are proposed in [64]. According to this paper, the robustness of the method can be increased by grouping pixels so as to form blocks of certain dimensions to enhance the low pass characteristics of the watermark signal. Alternatively, one can take advantage of the fact that different embedding factor can be used for each pixel, to shape appropriately the watermark signal. An optimization procedure that calculates the appropriate embedding value for each pixel so that the energy of the watermark signal is concentrated at low frequencies is proposed. Constraints that ensure that the watermark signal is invisible can be incorporated in the optimization procedure. 21

48 In [45] the authors derived analytical expressions for the probabilities P-, P+ of false negative and false positive watermark detection. Their model assumes an additive watermark and a correlator-based detection stage. Both, the white watermarks and watermarks having low pass characteristics, are considered. The host image is treated as noise, assuming a first order separable autocorrelation function. The probabilities P-, P+ are expressed in terms of the watermark to image power ratio. The authors conclude that detection error rates are higher for watermarks with low pass characteristics. In last 12 years, number of publications in this area is increasing very rapidly and no survey can cover all the presented schemes, but there are some very good survey papers and interested reader may explore the papers [3, 13, 54, 76]. We are limiting the discussion of the spatial domain based schemes here. 2.2 TRANSFORMED DOMAIN BASED SCHEMES As presented in literature, transformed domain based watermarking schemes are more robust as compared to simple spatial domain watermarking schemes. Such algorithms are robust against simple image processing operations like low pass filtering, brightness and contrast adjustment, blurring etc. However, they are difficult to implement and are computationally more expensive. We can use either of Discrete Fourier Transform (DFT), Discrete Cosine Transform (DCT) or Discrete Wavelet Transform (DWT) but DCT is the most exploited one. A General transformed domain based scheme, as presented by Cox, is shown in Figure 2.2. A very good discussion on DCT/DWT/DFT based watermarking schemes is given in [76] DFT BASED WATERMARKING SCHEMES We start from DFT. There are few algorithms that modify these DFT magnitude and phase coefficients to embed watermarks. Ruanaidh et al. proposed a DFT watermarking scheme in which watermark is embedded by modifying the phase information within the DFT. It has been shown that phase based watermarking is robust against image contrast operation [114]. Later Ruanaidh and Pun showed how Fourier Mellin transform could be 22

49 used for digital watermarking. Fourier Mellin transform is similar to applying Fourier Transform to log-polar coordinate system for an image. This scheme is robust against geometrical attacks [116]. De Rosa et al. proposed a scheme to insert watermark by directly modifying the mid frequency bands of the DFT magnitude component [115]. Ram kumar et al. also presented a data hiding scheme based on DFT, where they modified the magnitude component of the DFT coefficients. Their simulations suggest that magnitude DFT survives practical compression which can be attributed to the fact that most practical compression schemes try to maximize the PSNR. Hence using magnitude DFT is a way to exploit the hole in most practical compression schemes. Figure 2.2: A General Frequency domain based watermarking model as presented by Cox [19] 23

50 The proposed scheme is shown to be resistant to Joint Photographic Expert Group (JPEG) and (Set Partitioning In Hierarchical Trees) SPIHT compression [68]. Lin et al. presented a RST resilient watermarking algorithm. In their algorithm, the watermark is embedded in the magnitude coefficients of the Fourier transform re-sampled by log-polar mapping. The scheme is, however, not robust against cropping and shows weak robustness against JPEG compression (Q = 70) [53]. Solachidis and Pitas presented a novel watermarking scheme. They embed a circularly symmetric watermark in the magnitude of the DFT domain [8]. Since the watermark is circular in shape with its centre at image center, it is robust against geometric rotation attacks. The watermark is centered around the mid frequency region of the DFT magnitude. Neighborhood pixel variance masking is employed to reduce any visible artifacts. The scheme is computationally not expensive to recover from rotation. Robustness against cropping, scaling, JPEG compression, filtering, noise addition and histogram equalization is demonstrated. A semi-blind watermarking scheme has been proposed by Ganic and Eskicioglu [30]. They embed circular watermarks with one in the lower frequency while the other is in the higher frequency DCT BASED WATERMARKING SCHEMES DCT domain watermarking can be classified into Global DCT watermarking and Block based DCT watermarking. One of the first algorithms presented by Cox et al. [19] used global DCT approach to embed a robust watermark in the perceptually significant portion of the Human Visual System (HVS). Embedding in the perceptually significant portion of the image has its own advantages because most compression schemes remove the perceptually insignificant portion of the image. In spatial domain it represents the LSB. However in the frequency domain it represents the high frequency components. As described in [76], steps in DCT Block Based Watermarking Algorithm are: 1) Segment the image into non-overlapping blocks of 8x8; 2) Apply forward DCT to each of these blocks; 3) Apply some block selection criteria (e.g. HVS); 24

51 4) Apply coefficient selection criteria (e.g. highest); 5) Embed watermark by modifying the selected coefficients; and 6) Apply inverse DCT transform on each block. Most DCT based algorithms differ with each other on account of step 3 and 4 i.e. they differ either in the block selection criteria or coefficient selection criteria. Initially, Koch, Rindfrey, and Zhao [7] proposed a method for watermarking images. In that method, they break up an image into 8x8 blocks and compute discrete cosine transform (DCT) of each of these blocks. A pseudorandom subset of the blocks is chosen and then in each such block, a triplet of frequencies is selected from one of 18 predetermined triplets and modified so that their relative strengths encode a 1 or 0 value. The 18 possible triplets are composed by selection of three out of eight predetermined frequencies within the 8x8 DCT block. The choice of the eight frequencies to be altered within the DCT block is based on a belief that the middle frequencies have moderate variance, i.e. they have similar magnitude. This property is used to allow the relative strength of the frequency triplets to be altered without requiring a modification that would be perceptually noticeable. Several DCT based schemes are presented in [8, 17-19, 21, 37, 71, 74, 81, 99, 118]. Using the DCT, an image can easily be split up in pseudo frequency bands so that the watermark can conveniently be embedded in the most important middle band frequencies. Furthermore, the sensitivity of the HVS to the DCT based images has been extensively studied, which resulted in the recommended JPEG quantization Table [112]. These results can be used for predicting and minimizing the visual impact of the distortion caused by the watermark. Finally, the block-based DCT is widely used for image and video compression. By embedding a watermark in the same domain as the compression scheme used to process the image (in this case in the DCT domain), we can anticipate lossy compression because we are able to anticipate which DCT coefficients will be discarded by the compression scheme. Furthermore, we can exploit the DCT decomposition to make real-time watermark applications. 25

52 Further improvements for DCT-domain correlation-based watermarking systems' performance could be achieved by using watermark detectors based on generalized Gaussian model instead of the widely used pure Gaussian assumption [35]. By performing a theoretical analysis for DCT-domain watermarking methods for images, the authors in [35] provided analytical expressions which could be used to measure beforehand the performance expected for a certain image and to analyze the influence of the image characteristics and system parameters (e.g. watermark length) on the final performance. Furthermore, the result of this analysis may help in determining the proper detection threshold T to obtain a certain false positive rate. The authors in [35] claimed that by abandoning the pure Gaussian noise assumption, some substantial performance improvements could be obtained. In [4], the authors embedded a watermark signal domain by modifying a number of predefined DCT coefficients. They used a weighting factor to weight the watermark signal in the spatial domain according to HVS characteristics. In [75] authors embedded watermark data in DCT Difference (JND) as predicted domain in perceptually meaningful way and used the Just Noticeable by model reported in [108] THE MIDDLE-BAND COEFFICIENT EXCHANGE SCHEME [42, 52]: The middle-band frequencies (FM) of an 8x8 DCT block are shown in Figure 2.3. In this Figure, FL is used to denote the lower frequency components of the block and FH is used to denote the higher frequency components. FM is chosen as embedding region to provide additional resistance to lossy compression techniques, while avoiding significant modification of the cover image. First, 8x8 DCT of an original image is taken. Then, two locations DCT (u 1, v 1 ) and DCT (u 2, v 2 ) are chosen from the F M region for comparison of each 8 x 8 block. These locations are selected based on the recommended JPEG quantization table shown in Figure 2.4. If two locations are chosen such that they have identical quantization values, then any scaling of one coefficient will scale the other by the same factor to preserve their relative strength. It may be observed from Figure 2.4, that coefficients at location (4, 1) and (3, 2) or (1, 2) and (3, 0) are more suitable candidates for comparison because their quantization values are equal. The DCT block 26

53 will encode a 1 if DCT (u 1, v1) > DCT (u 2, v2); otherwise it will encode a 0. The coefficients are swapped if the relative size of coefficients does not agree with the bit that is to be encoded [42, 52]. Thus, instead of embedding any data, this scheme is hiding watermark data by means of interpreting 0 or 1 with relative values of two fixed locations in middle frequency region. FL FM FH Figure 2.3: Frequency regions in 8 x 8 DCT Swapping of such coefficients will not alter the watermarked image significantly, as it is generally believed that DCT coefficients of middle frequencies have similar magnitudes. Further, the robustness of the watermark can be improved by introducing a watermark strength constant k, such that DCT (u 1, v 1 ) DCT (u 2, v 2 ) > k. If coefficients do not meet these criteria, they are modified by the use of random noise to satisfy the relation. Increasing k thus reduces the chance of detection errors at the expense of additional image degradation. By increasing k, larger coefficients remain larger even after lot of compression and thus help in decoding because their relative values decide the decoding of the watermark data. While extracting the watermark, again the 8x8 DCT of image in taken in which 1 is decoded if DCT (u 1, v1) > DCT (u 2, v2); otherwise a 0 is decoded. 27

54 Figure 2.4: JPEG Quantization matrix Limitation of middle-band coefficient exchange scheme: Experimental results show that Middle-Band Coefficient Exchange is quite efficient against JPEG compression, Cropping, Noising and other common image manipulation operations. But above scheme has one serious drawback. If only one pair of coefficient is used (say (4, 1) and (3, 2)) to hide the watermark data, then it is vulnerable to collusion attack. By analyzing four or five watermarked copies of an image, one can easily find out that these coefficients always have a certain pattern and attacker can predict the watermark as well as destroy it DCT-CDMA BASED IMAGE WATERMARKING [42, 52]: In this technique authors embedded a PN sequence W into the middle frequencies of the DCT block. A DCT block can be modulated using the Equation 2.2. I W x I x, y ( u, v) + k * Wx, y ( u, v), u, v FM ( u, v) = y.. (2.2), Ix, y( u, v), u, v FM For each 8 x 8 block of the image, the DCT for the block is first calculated. In that block, the middle frequency components F M are added to the PN sequence W, multiplied by a gain factor k. Each block is then inverse-transformed to give the final watermarked image I W. The watermarking procedure is made somewhat more adaptive by slightly altering the embedding process to the method shown in Equation

55 I W x I x, y ( u, v) * (1 + k * Wx, y ( u, v)), u, v FM ( u, v) = y... (2.3), Ix, y( u, v), u, v FM This slight modification scales the strength of the watermarking based on the size of the particular coefficients being used. Larger values of k can thus be used for coefficients of higher magnitude; in effect strengthening the watermark in regions that can afford it; weakening it in other regions. For detection, the image is broken up into same 8x8 blocks and a DCT is taken. The same PN sequence is then compared to the middle frequency values of the transformed block. If the correlation between the sequences exceeds some threshold T, a 1 is detected for that block; otherwise a 0 is detected. Again k denotes the strength of the watermarking, where increasing k increases the robustness of the watermark at the expense of quality DWT BASED WATERMARKING SCHEMES If watermarking techniques can exploit the characteristics of the Human Visual System (HVS), it is possible to hide watermarks with more energy in an image, which makes watermarks more robust. From this point of view, the DWT is a very attractive transform, because it can be used as a computationally efficient version of the frequency models for the HVS [5]. For instance, it appears that the human eye is less sensitive to noise in high resolution DWT bands and in the DWT bands having an orientation of 45 (i.e., HH bands). Furthermore, DWT image and video coding, such as embedded zero-tree wavelet (EZW) coding, are included in the upcoming image and video compression standards, such as JPEG2000 [112]. Thus DWT decomposition can be exploited to make a real-time watermark application. Many approaches apply the basic schemes described at the beginning of this section to the high resolution DWT bands, LH, HH, and HL [35, 40]. A large number of algorithms operating in the wavelet domain have been proposed till date. 29

56 Figure 2.5: 1-Scale and 2-Scale 2-Dimensional Discrete Wavelet Transform CDMA-DWT BASED WATERMARKING SCHEME: This scheme is the most straightforward scheme which is similar to embedding scheme to that used in the DCT-CDMA scheme. The embedding of a CDMA sequence in the frequency bands is shown in Equation 2.4. I W u, v W i = + α W W i i x, i u, v HL, LH u, v LL, HH. (2.4) where Wi denotes the coefficient of the transformed image, x i the bit of the watermark to be embedded, and α a scaling factor. To detect the watermark, same pseudo-random sequence used in CDMA generation is generated and its correlation is determined with the two transformed frequency bands. If the correlation exceeds some threshold T, the watermark is detected. This can be easily extended to multiple bit messages by embedding multiple watermarks into the image. In the spatial version, a separate seed is used for each PN sequence, which are then added to the frequency coefficients. During detection, if the correlation exceeds T for a particular sequence a 1 is recovered; otherwise a 0. The recovery process then iterates through the entire PN sequence until all the bits of the watermark have been recovered. 30

57 DWT based watermarking schemes follow the same guidelines as DCT based schemes, i.e. the underlying concept is the same; however, the process to transform the image into its transform domain varies and hence the resulting coefficients are different. Wavelet transforms use wavelet filters to transform the image. There are many available filters, although the most commonly used filters for watermarking are Haar Wavelet Filter, Daubechies Orthogonal Filters and Daubechies Bi-Orthogonal Filters. Each of these filters decomposes the image into several frequencies. Single level decomposition gives four frequency representations of the images. In their paper [76], authors presented a survey of wavelet based watermarking algorithms. They classify algorithms based on decoder requirements as Blind Detection or Non-blind Detection. As mentioned earlier blind detection doesn't require the original image for detecting the watermarks; however, non-blind detection requires the original image DWT BASED BLIND WATERMARK DETECTION: Lu et al. [58] presented a novel watermarking technique called as "Cocktail Watermarking". This technique embeds dual watermarks which compliment each other. This scheme is resistant to several attacks, and no matter what type of attack is applied; one of the watermarks can be detected. Furthermore, they enhance this technique for image authentication and protection by using the wavelet based Just Noticeable Distortion (JND) values. Hence this technique achieves copyright protection as well as content authentication simultaneously. Zhu et al. [126] presented a multi-resolution watermarking scheme for watermarking video and images. The watermark is embedded in all the high pass bands in a nested manner at multiple resolutions. This scheme doesn't consider the HVS aspect; however, Kaewkamnerd and Rao [43-44] improve this scheme by adding the HVS factor in account. Voyatzis and Pitas [104], who presented the "toral automorphism" concept, provide a technique to embed binary logo as a watermark which can be detected using visual models as well as by statistical means. So, in case the image is degraded too much and the logo is not visible, it can be detected statistically using correlation. Watermark embedding is based on a chaotic (mixing) system. Original image is not required for watermark detection. However, the watermark is embedded in spatial domain by modifying the pixel or luminance values. 31

58 A similar approach is presented for the wavelet domain [121], where the authors proposed a watermarking algorithm based on chaotic encryption. Zhao et al.[125] presented a dual domain watermarking technique for image authentication and image compression. They used the DCT domain for watermark generation and DWT domain for watermark insertion. A soft authentication watermark is used for tamper detection and authentication while a chrominance watermark is added to enhance compression. They use the orthogonality of DCT-DWT domain for watermarking [125] DWT BASED NON-BLIND WATERMARK DETECTION: This technique requires the original image for detecting the watermark. Most of the schemes found in literature use a smaller image as a watermark and hence cannot use correlation based detectors for detecting the watermark; as a result they rely on the original image for informed detection. The size of the watermark image (normally a logo) normally is smaller compared to the host image. Xia et al. presented a wavelet based non-blind watermarking technique for still images where watermarks are added to all bands except the approximation band. A multi-resolution based approach with binary watermarks is presented here [37]. Here, both the watermark logo as well as the host image is decomposed into sub bands and later embedded. Watermark is subjectively detected by visual inspection; however, an objective detection is employed by using normalized correlation. Lu et al. presented another robust watermarking technique based on image fusion. They embedded a grayscale and binary watermark which is modulated using the "toral automorphism" described in [106]. Watermark is embedded additively. The novelty of this technique lies in the use of secret image instead of host image for watermark extraction and use of image dependent and image independent permutations to de-correlate the watermark logos [57]. Raval and Rege presented a multiple watermarking scheme. The authors argued that if the watermark is embedded in the low frequency components, it is robust against low pass filtering, lossy compression and geometric distortions. On the other hand, if the watermark is embedded in high frequency components, it is robust against contrast and brightness adjustment, gamma correction, histogram equalization and cropping and vice-versa. Thus, to achieve overall robustness 32

59 against a large number of attacks, the authors proposed to embed multiple watermarks in low frequency and high frequency bands of DWT [78]. Kundur and Hatzinakos [50] presented image fusion watermarking scheme. They used salient features of the image to embed the watermark. They used a saliency measure to identify the watermark strength and later embedded the watermark additively. Normalized correlation is used to evaluate the robustness of the extracted watermark. Later the authors proposed another scheme termed as FuseMark [51] which includes minimum variance fusion for watermark extraction. Here, they propose to use a watermark image whose size is a factor of the host by 2xy. Tao and Eskicioglu presented an optimal wavelet based watermarking scheme. They embedded binary logo watermark in all the four bands. But they embedded the watermarks with variable scaling factor in different bands. The scaling factor is high for the LL sub band but for the other three bands it is lower. The quality of the extracted watermark is determined by Similarity Ratio measurement for objective calculation [100]. Ganic and Eskicioglu inspired by Raval and Rege [78] proposed a multiple watermarking scheme based on DWT and Singular Value Decomposition (SVD). They argued that the watermark embedded by Raval and Rege [78] scheme was visible in some parts of the image especially in the low frequency areas, which reduced the commercial value of the image. Hence they generalized their scheme by using all the four sub bands and embedding the watermark in SVD domain. The core technique is to decompose an image into four sub bands and then applying SVD to each band. The watermark is actually embedded by modifying the singular values from SVD [30]. 2.3 RECENT METHODOLOGIES Now-a-days, researchers are focusing on mixing of spatial and transformed domains (i.e. combinations of DFT, DWT and DCT) concepts and also applying more and more mathematical and statistical model, and other interdisciplinary approaches in watermarking: for example use of chaotic theory, fractal image coding etc. In this section we are presenting the brief of few recent watermarking algorithms. 33

60 In [103], authors presented a reversible watermarking scheme for the 2D-vector data (point coordinates), which are used in geographical information related applications. This reversible watermarking scheme exploits the high correlation among points in the same polygon in a map and achieves the reversibility of the whole scheme by an 8-point integer DCT, which ensures that the original 2D-vector data can be watermarked during the watermark embedding process and then perfectly restored during the watermark extraction process. In this scheme, author used an efficient highest frequency coefficient modification technique in the integer DCT domain to modulate the watermark bit 0 or 1, which can be determined during extraction without using any additional information. To alleviate the visual distortion in the watermarked map caused by the coefficient modification, they proposed an improved reversible watermarking scheme based on the original coefficient modification technique. Combined with this improved scheme, the embedding capacity could be greatly increased while the watermarking distortion is reduced as compared to the original coefficient modification scheme presented in [103]. In [65], authors presented zero-knowledge watermark detectors. Current detectors are based on a linear correlation between the asset features and a given secret sequence. This detection function is susceptible of being attacked by sensitivity attacks for which zeroknowledge does not provide protection. In this work, a new zero-knowledge watermark detector robust to sensitivity attacks is presented, using the generalized Gaussian Maximum Likelihood (ML) detector as the basis. The inherent robustness that this detector presents against sensitivity attacks, together with the security provided by the zero-knowledge protocol that conceals the keys that could be used to remove the watermark or to produce forged assets, results in a robust and secure protocol. Additionally, two new zero-knowledge proofs for modulus and square root calculation are presented. They serve as building blocks for the zero-knowledge implementation of the Generalized Gaussian ML detector, and also open new possibilities in the design of high level protocols. 34

61 If digital watermarking is to adequately protect content in systems which provide resolution and quality scalability, then the watermarking algorithms must provide both resolution and quality scalability. Although there exists a trade off between resolution and quality scalability, it has been demonstrated that it is possible to achieve both types by taking advantage of human visual system characteristics to increase quality scalability without compromising resolution scalability. Watermarking algorithms considering this problem have been proposed; however, they tend to focus on a single type of scalability, resolution [96, 120] or quality [12, 98]. Peng et al. [66] considered both types, but their algorithm deals exclusively with authentication and is not a watermarking algorithm. In their work [67] authors focused on providing a spread spectrum watermarking algorithm which had both resolution and quality scalability demonstrated through experimental testing using the JPEG2000 compression algorithm. To alleviate this trade off, they began with a non-adaptive resolution scalable algorithm and exploited the contrast sensitivity and texture masking characteristics of the HVS to construct an HVS adaptive algorithm that has good quality scalability. Their algorithm is specifically designed to concentrate on textured regions only, avoiding the visible distortions, which may occur when strength increases are applied to edges. Furthermore, this texture algorithm is applied in the wavelet domain but uses only a single resolution for each coefficient to be watermarked. In their work [126], authors presented a new image adaptive watermarking scheme based on perceptually shaping watermark block wise. Instead of the global gain factor, a localized one is used for each block. Watson s DCT-based visual [109] model is adopted to measure the distortion of each block introduced by watermark, rather than the whole image. With the given distortion constraint, the maximum output value of linear correlation detector is derived in one block, which demonstrated the reachable maximum robustness in a sense. Meanwhile, an EXtended Perceptually Shaped Watermarking (EX- PSW) is acquired through making detection value which approaches to upper limit. It is proved mathematically that EX-PSW outputs higher detection value than Perceptually Shaped Watermarking (PSW) with the same distortion constraint. Authors used this idea and also discussed the adjustment strategies of parameters in EX-PSW, which were 35

62 helpful for improving the local image quality. Experimental results show that scheme provides very good results both in terms of image transparency and robustness. In [10], authors presented an Independent Component Analysis (ICA) [40-41] based watermarking method. This watermarking scheme is domain-independent ICA-based approach. This approach can be used on images, music or video to embed either a robust or fragile watermark. In the case of robust watermarking, the method shows high information rate and robustness against malicious and non-malicious attacks while inducing low distortion. Another version of this scheme is a fragile watermarking scheme which shows high sensitivity to tampering attempts while keeping the requirement for high information rate and low distortion. The improved performance is achieved by employing a set of statistically independent sources (the independent components) as the feature space and principled statistical decoding methods. In [90], authors presented a dual watermarking Scheme. In general, the watermark embedding process affects the fidelity of the underlying host signal. Fidelity, robustness and the amount of data which can be embedded without visible artifacts, often conflict. Most of early watermarking schemes have focused on embedding the watermark information applying a global power constraint such as the Peak-Signal-to-Noise-Ratio (PSNR) to satisfy fidelity constraints. But, the PSNR value is reflecting human s visual system because local image properties such as edges or textures are not considered. The watermarking systems have been proposed that allowed the embedded signal to be locally varied in response to the local properties of the corresponding host signal [38, 73, 77]. Authors in their paper [90] neglected the PSNR value and use the fact that all common lossy image compression schemes are PSNR optimized. They embedded watermark information by geometrically shifting objects and object borders in a given host image. If an observer has no original image for comparison, the embedding process is imperceptible. As a consequence, this approach turns out to be extremely robust to common image compression. Common lossy image compression is optimized for maintaining the geometric image structure. Hence, as they demonstrate, the embedded 36

63 information is not affected by a successive embedding approach in the compression domain. Authors in their paper [39] presented an improved invariant wavelet and designed a DCT based blind watermarking algorithm against Rotation-and Scaling-and Translation (RST) attacks by exploiting the affined invariance of the invariant wavelet. Surviving geometric attacks in image watermarking is considered to be of great importance. In the face of geometrical attacks, all shortcomings of almost all digital watermarking algorithms have been exposed. Therefore, this paper presents an improved invariant wavelet that is better than the bilinear interpolation and whose performance is close to of bi-cubic when scaling factor is very close to 1, and designs a novel blind image watermarking algorithm based on DCT in the (RST) Xiong s Invariant Wavelet, i.e. RSTXIW domain. The experiments show that this novel watermarking algorithm is robust against filter, noise and arbitrary RST geometrical attacks, however, sensitive to local crop attacks. In their paper [107], authors presented an image watermarking scheme based on 3-D DCT. A gray-level image is decomposed into a 3-D sub-image sequence by sub sample of zigzag scanning order that is transformed using block-based 3-D DCT. Simultaneously, they proved that the distribution of 3-D DCT AC coefficients follows the generalized Gaussian density function using the distribution relative entropy theory. To satisfy the balance between the robustness and the imperceptivity, a 3-D HVS model is improved to adjust the embedding strength. In watermark detecting, the optimum detector is used to implement the blind detection. It is shown in experiments that the scheme is strongly robust against various attacks. In paper [101] proposed digital watermarking scheme uses the properties of DCT and DWT to achieve almost zero visible distortion in the watermarked images. These schemes use a unique method for spreading, embedding and extracting the watermark. Embedding using a linear relation between the transform coefficients of the watermark and a security matrix has been proposed with satisfactory results. 37

64 In [59], algorithm is based on multistage Vector Quantization (VQ) that embeds both robust watermark for copyright protection or ownership verification and fragile watermark for content authentication or integrity attestation. The method in [122] combined DCT and VQ to simultaneously embed robust and fragile watermarks. In their paper [31], two simple dither modulation schemes for a pair of DCT coefficients are proposed. The first step is to handle the original image using the sub sampling technique as described in [14]. Then, transform it into DCT domain to obtain four sub images. By dividing them into two groups, we introduce distinguishing dither modulation processes in the two coefficient pairs with two robust watermarks. Experimental results show that the proposed method is blind and robust and through adopting dither modulation in sub images gained by sub sampling, two independent robust watermarks can be embedded in the original image. In the field of color images watermarking, many methods are accomplished by marking the image luminance, or by processing each color channel separately. Therefore in paper [55], authors proposed a new DCT domain watermarking expressly devised for RGB color images based on the diversity technique in communication system. The watermark is hidden within the data in the same sequence by modifying a subset of the block. DCT coefficients of each color channel. Detection is based on combination method which takes into account the information conveyed by three color channels. Even if a particular channel is severely faded, they are still able to recover a reliable estimated of transmitted watermark through other propagation channel. Experimental results, as well as theoretical analysis, are presented to demonstrate the validity of the new approach with respect to algorithm operating on image luminance only. 2.4 PROBLEM STATEMENT FORMULATION Since, financial implications of some of the application areas like fingerprinting and copyright protection are very high and till now no successful algorithm seems to be 38

65 available to prevent illegal copying of the multimedia contents, the primary goal of this thesis work is to develop watermarking schemes for images (which are stored in spatial domain as well as transformed domain) which can sustain the known attacks and various image manipulation operations. Out of image, audio and video, the image watermarking was chosen as a goal because any successful image watermarking algorithm may be extended to video watermarking also. Therefore, to keep the future extension in mind, the cover medium chosen is an image. Based on the literature survey presented in Sections 2.1, 2.2 and 2.3, the following issues were also identified: ISSUE 1: Till now there in no Generic nature in the watermarking algorithms available. More precisely, if certain approach is applicable for a gray level image, the same approach does not work for the other formats of an image. ISSUE 2: Even if gray color image watermarking algorithms are extended for RGB color images, the maximum work has been done for BLUE color channel only because human eyes are less sensitive to detect the changes in BLUE color channel. No attack impact analysis, i.e, which color channel may be affected by a particular attack, has been carried out. In view of the above, our problem statements are as follows: Problem statement 1: Choose Image Watermarking as a major problem. Problem statement 2: Identify, for multi-color channel images (True color windows BMP, uncompressed JPEG), the suitability of a color channel with respect to attack (if any). Problem statement 3: Explore the ways such that attack impacts may be minimized before the watermark embedding process. 39

66 ISSUE 3: In most of the research papers, once the watermarking scheme is finalized, it is applied to all test images. Since each image is different and has certain characteristics and after embedding the watermark data by a particular watermarking scheme, its performance against a particular attack may not be similar with other image. No study is conducted to make the embedding scheme based on some image characteristics. Thus, the next problem statement is: Problem statement 4: Explore the relationship between the performance of watermarking scheme and the cover image characteristics itself. ISSUE 4: Mostly watermarking schemes are developed in a way that first a scheme is developed based on the extension of earlier presented one and then see its performance against the common image manipulation and known attack. There are huge financial implications for watermarking schemes (say fingerprinting), but no scheme has been developed, which is, by design, resistant to at least one attack which can not be conducted by an attacker, leading to next problem statement: Problem statement 5: Embed an inherent nature in the developed watermarking schemes to guarantee that at least one serious attack having most financial implications cannot attack on watermarked images JUSTIFICATIONS OF THE PROBLEM STATEMENT CHOSEN While deciding the way to start the development of our watermarking schemes, first we resolved the ISSUE 4, because this must be dealt first among all the 4 issues listed above. It is known that the application area having the highest financial implications is Fingerprinting. If attacker has access to more than one copy of watermarked image, he/she can predict/ remove the watermark data by colluding them. This is known as Collusion attack. 40

67 Researchers working on fingerprinting primarily focus on the collusion attack. Network technology research center, Nanyang Technological University, Singapore website states that they pay at least equal attention to watermark attacks/counter-attacks as watermark designs [63]. To facilitate pirate tracing in video distribution applications, different watermarks carrying distinguishing client information are embedded at source. If few clients requesting for the same source data get their differently marked versions together, they may collude to remove or weaken the watermark leading to what is commonly called collusion attack. Collusion attacks are powerful attacks because they are capable of achieving their objective without causing much degradation in visual quality of the attacked data (sometimes, visual quality may even improve after attack.). In their paper Multi-bits Fingerprinting for Image [46], authors focused on collusion attack for fingerprinting application. It has been stated, that in fingerprinting different copies for each customer can be produced and this point is very helpful for attackers. Attackers compare several fingerprinting copies and find the location of the embedded information and destroy it by altering the values in those places where a difference was detected. One more work, specially conducted against collusion attack can be found as Collusionresistant watermarking and fingerprinting (US Patent Issued on June 13, 2006) [15]. Interested readers can find more literature based on collusion attack on watermarking system in [ ]. Therefore, Collusion attack is the most severe problem for the watermarking application area having the most financial impact. So while designing a watermark scheme, we decided that our proposed schemes must be designed in such a way that schemes are inherently collusion attack resistant. Therefore by this thesis, we are presenting a new term ICAR (Inherently Collusion Attack Resistant) as a requirement for a watermarking system. 41

68 The other 3 issues were taken into account while developing the watermarking schemes. After this, we had to decide the working domain and approaches of our developments based on the findings in the literature survey. Since transformed domain watermarking has been proved better than spatial domain watermarking, we decided to start with the transformed domain watermarking for gray level images and then subsequently move further for Colored and JPEG image watermarking keeping the first issue in mind. Apart from ICAR nature and resistant to common image manipulations and known attacks, we primarily focus the JPEG compression attack. This Lossy attack can reduce the size of an image up to 1% without altering much visual quality of an image. Therefore, we picked up the classical Middle Band Coefficient Exchange (MBCE) scheme (refer Section ) as a base for developing our schemes because this scheme takes the JPEG quantization table into consideration to hide the watermark data and thus ensures the robustness against JPEG compression attack. To move further, we again had to decide the categories of the watermarking application areas based on Figure 1.2, we are targeting to develop in this thesis work. Thus, Figure 2.6 is the same as Figure 1.2 but with highlighted types. 42

69 Figure 2.6: The Targeted types of to be developed watermarking schemes The first 2 red highlighters are already justified. The last one (destination based) is again understood as we are focusing ICAR nature in our watermarking schemes to be developed, which are highly correlated with fingerprinting which comes under the destination based watermarking. Among the visible and invisible, we picked up the non-visible watermarking because in most of the cases, the presence of the watermark or copyright data is to be hidden. The most crucial decision before us was to decide the choice among fragile versus robust watermarking. Since in the business, Temper detection have more serious financial 43

70 implications than the copy or copyright control, we decided to go for fragile watermarking. To conclude, it was decided to work on IMAGE WATERMARKING in TRANSFORMED DOMAIN (more precisely DCT based) to develop an ICAR watermarking scheme to hide an INVISIBLE watermark data which is FRAGILE in nature. In addition, the schemes to be developed should be generic in nature i.e. which could be extended to other images which are stored in spatial domain and transformed domain. 44

71 CHPATER-3 PRELIMINARIES This chapter covers the material which is required to be well understood before coming to next chapters containing the work contributions. This thesis has given special attention to JPEG compression of an image because this compression is most commonly used and it reduces the size of an image very much without noticeable degradation in image quality. Every image watermarking scheme must survive against JPEG compression attack. Therefore, first we are giving a brief introduction of JPEG and JPEG2000 image encoding. 3.1 IMAGE ENCODING STANDARDS JPEG ENCODING The JPEG Image Compression is a standard image compression mechanism. Developed by Joint Photographic Experts Group, JPEG compression is "lossy," because the compression scheme sacrifices some image quality for a reduction in the image data size. The JPEG compression scheme [56, 112] is shown in Figure 3.1. Figure 3.1: JPEG Compression Scheme First, the source image should be converted from RGB into a different color space called YCbCr. It has three components Y, Cb and Cr; the Y component represents the brightness of a pixel, the Cb and Cr components represent the chrominance (split into 45

72 blue and red components). The YCbCr color space conversion allows greater compression for the same image quality (or greater image quality for the same compression). The human eye can see more detail in the Y component (brightness) than in Cb (blue) and Cr (red). Using this knowledge, encoders can be designed to compress images more efficiently. The above transformation enables the next step, which is to reduce the Cb and Cr components (called "downsampling" or "chroma subsampling"). After subsampling, each channel is split into 8 8 blocks (of pixels). Next, each component (Y, Cb, Cr) of each 8 8 block is converted to a frequency-domain representation, using two-dimensional DCT. DCT is a widely used transform coding technique which is able to perform decorrelation of the input signal in a data independent manner. In case of image, we use 2D DCT. For more details related to 2D DCT, one may refer pp of Fundamentals of Multimedia [56]. Let us take an example of an 8 8, 8-bit sub image, as shown in Figure 3.2 below: Figure 3.2: An example sub image The next step is to transform the subimage from a positive range to the one which is centered on zero. For an 8-bit image, each pixel has 256 possible values (0 to 255). To center on zero, it is necessary to subtract each pixel by half the number of possible values, i.e Subtracting 128 from each pixel value yields pixel values in the range [ 128,127] resulting in the matrix shown in Figure

73 Figure 3.3: Example sub image after subtracting 128 from each pixel The next step is to take the 2-D DCT which is given by Equation 3.1 below: (3.1) Where, u is the horizontal spatial frequency, for the integers 0<= u < 8, v is the vertical spatial frequency, for the integers 0<= v < 8, is a normalizing function, g x,y is the pixel value at coordinates (x, y), and G u,v is the DCT coefficient at coordinates (u, v). If we perform this transformation on our matrix above given in Figure 3.3 and then round to the nearest integer, we get a DCT coefficient matrix, which is shown in Figure 3.4. Figure 3.4: DCT of sub image shown in Figure

74 It may be observed from Figure 3.4, that the top-left corner value is the largest one. This is the DC coefficient. The remaining 63 coefficients are called the AC coefficients. The DCT temporarily increases the bit-depth of the image since the DCT coefficients of an 8- bit/component image take up to 11 or 12 bits (depending on fidelity of the DCT calculation) to store. This may force the codec to temporarily use 16-bit data to hold these coefficients doubling the formal size of the image representation at this point. The advantage of the DCT is its tendency to aggregate most of the signal in one corner of the result, as may be seen above. The quantization step to follow accentuates this effect while simultaneously reducing the size of the DCT coefficients to 8 bits or less, resulting in a signal with a large trailing region containing zeros that the entropy stage can simply throw away. The temporary increase in size, at this stage, is not a performance concern for most JPEG implementations because typically only a very small part of the image is stored in full DCT form at any given time during the encoding or decoding process. After taking the DCT, next step is the quantization. The human eye is good at seeing small differences in brightness over a relatively large area, but at the same time it is not so good at distinguishing the exact strength of a high frequency brightness variation. This fact allows one to get away with greatly reducing the amount of information in the high frequency components. This is done by simply dividing each component in the frequency domain by a constant for that component, and then rounding to the nearest integer. This is the main lossy operation in the whole process. As a result of this, it is typically the case that many of the higher frequency components are rounded to zero, and many of the rest become small positive or negative numbers which take many fewer bits to store. A common quantization matrix is shown in Figure

75 Figure 3.5: JPEG Quantization matrix The quantized DCT coefficients are computed with the help of the Equation 3.2 given below: (3.2) where A is the unquantized DCT coefficients; Q is the quantization matrix above; and B is the quantized DCT coefficients. Using this quantization matrix with the DCT coefficient matrix in Figure 3.4, DCT values after quantization are given in Figure 3.6. For example, Figure 3.6: DCT values after quantization B 11 = round (A 1, 1 / Q 1, 1 ) = round (-415/16) = round ( ) = It may be noted that most of the AC coefficients are now ZERO. After quantization, as shown in Figure 3.1, entropy encoding is done as follows: 49

76 Step-1: Zigzag Scan - The resulting matrix after quantization will contain many zeros. The lower the quality setting, the more zeros will exist in the matrix. By re-ordering the matrix from the top-left corner into a 64-element vector in a zig-zag pattern, the matrix is essentially sorted from low-frequency components to high-frequency components. As the high-frequency components are the most likely to round to zero, one will typically end up with a run of zeros at the end of the 64-entry vector. This is important for the next step. Step-2: DPCM on DC component - On a block-by-block basis, the difference in the average value (across the entire block, the DC component) is encoded as a change from the previous block's value. This is known as Differential Pulse Code Modulation. Step-3: Run Length Encoding (RLE) on AC components - On the individual entries in the 64-element vector (the AC components), a Run Length Encoding stores each value along with the number of zeros preceding it. As the 1x64 vector contains lot of zeros, it is more efficient to save the non-zero values and then count the number of zeros between these non-zero values. The RLE stores a skip and a value, where skip is the number of zeros before this component, and the value is the next non-zero component. Step-4: Entropy Coding / Huffman Coding - A dictionary is created which represents commonly used strings of values with a shorter code. More common strings / patterns use shorter codes (encoded in only a few bits), while less frequently used strings use longer codes. So long as the dictionary (Huffman Table) is stored in the file, it is an easy matter to lookup the encoded bit string to recover the original values. Once image is stored compressed, it needs to be decompressed for viewing. This scheme of decompression is given in Figure

77 Figure 3.7: JPEG Decompression Scheme While decompressing, we will multiply the stored values (shown in Figure 3.6) with quantization matrix. Taking the entry-for-entry product with the quantization matrix results the matrix shown in Figure 3.8. Figure 3.8: DCT values regenerated in decompression Taking the inverse DCT of above values results in an image with values (still shifted down by 128) gives a matrix shown below in Figure 3.9 (a). Figure 3.9 (a): Sub image pixel values (still shifted down by 128) Adding 128 to each entry in the above matrix, we get 51

78 Figure 3.9 (b): Decompressed sub image pixel values This is the uncompressed subimage and can be compared to the original subimage (refer Figure 3.2) by taking the difference (original, uncompressed) results in error values: Figure 3.10: Error matrix for example sub image with an average absolute error of about 5 values per pixels as shown below: Higher compression ratio first affects the high-frequency textures in the upper-left corner of the image, and contrasting lines become fuzzier. The very high compression ratio severely affects the quality of the image, although the overall colors and image form is still recognizable. However, the precision of colors suffers less (for a human eye) than the precision of contours (based on luminance). This justifies the fact that images should be first transformed in a color model separating the luminance from the chromatic information before subsampling the chromatic planes (which may also use lower quality quantization) in order to preserve the precision of the luminance plane with more information bits. For example, an uncompressed 24-bit RGB bitmap image (73,242 pixels) would require 219,726 bytes (excluding all other information headers). The full quality image 52

79 (Q = 100) is encoded at 9 bits per color pixel, the medium quality image (Q = 25) uses 1 bit per color pixel. For most applications, the quality factor should not go below 0.75 bit per pixel (Q = 12.5), as demonstrated by the low quality image. The image at the lowest quality uses only 0.13 bit per pixel and displays very poor color. It could only be usable after subsampling to a much lower display size JPEG2000 ENCODING JPEG 2000 is a wavelet-based image compression standard [113]. Wavelet transform: In DCT, we use a special cosine based transform. If we carry out analysis based on both sine and cosine, then a concise notation assembles the results into a complex valued function of real valued frequency. Such decomposition results in very fine resolution in frequency domain. However, since a sinusoid is theoretically infinite in extent in time, such a decomposing gives no temporal resolution. Wavelet transform represents the signal with good resolution both in time and frequency, by using a set of besis functions called wavelets. There are two types of wavelet transforms: Complex Wavelet Transform (CWT) and DWT. Since image is a discrete signal, we are moving to discrete wavelet transform. Haar is the simplest form of the wavelet transform which forms the average and difference of the sequence of the values. If we repeatedly take average and difference and keep results for every step, we create a multi-resolution analysis of the sequence, as shown in Figure 2.5. For image, this would be equivalent to creating smaller and smaller summary images, one quarter the size for each step, and keeping track of differences from the average as well. As shown in Figure 2.5, at each step image is split in four subbands namely LL (low-low), HL (high-low), LH (low-high) and HH (high-high). The LL subband can be further decomposed to yield yet another level of decomposition. More details of DWT and Multiresolution analysis can be found in [56]. In the core of JPEG2000 is the Embedded Block Coding with Optimized Truncation (EBCOT) algorithm. The basic idea of EBCOT is the partition of each subband LL, LH, HL, HH produced by above wavelet transform into small blocks called code blocks. 53

80 JPEG 2000 requires more decompression time than JPEG and allows more sophisticated progressive downloads, yet it averages similar compression rates. Unlike JPEG, JPEG2000 becomes increasingly blurred with higher compression ratios rather than generating "blocking and ringing" artifacts. At high bit rates, where artifacts become nearly imperceptible, JPEG 2000 has a small machine-measured fidelity advantage over JPEG. At lower bit rates (for example, less than 0.25 bits/pixel for gray-scale images), JPEG2000 has advantage over certain modes of JPEG in a way that artifacts are less visible and there is almost no blocking. The compression gains over JPEG are attributed to the use of DWT and a more sophisticated entropy encoding scheme. JPEG2000 decomposes the image into a multiple resolution representation. The aim of JPEG2000 is not only improved compression performance over JPEG but also to add features such as scalability and editability. Very low and very high compression rates are supported in JPEG2000. To handle a very large range of effective bit rates is one of the strengths of JPEG2000. For example, to reduce the number of bits for a picture below a certain amount, the advisable thing to do with the first JPEG standard is to reduce the resolution of the input image before encoding it. That is unnecessary when using JPEG 2000, because JPEG2000 already does this automatically through its multiresolution decomposition structure. In JPEG2000, images have to be transformed from the RGB color space to another color space using any of the following two transforms [113]: 1) Irreversible Color Transform (ICT) which uses the well known YC B C R color space. It is called "irreversible" because of the quantization errors it introduces. 2) Reversible Color Transform (RCT) which uses a modified YUV color space that does not introduce quantization errors, so it is fully reversible. After color transformation, the image is split into so-called tiles, the rectangular regions of the image that are transformed and encoded separately. Tiles can be of any size, and we may consider the whole image as one single tile but all the tiles will have the same 54

81 size. Dividing the image into tiles is advantageous because the decoder will need less memory to decode the image and it can opt to decode only selected tiles to achieve a partial decoding of the image. The disadvantage of this approach is that the quality of the picture decreases. Using large number of tiles can create a blocking effect similar to the older JPEG standard. These tiles are then wavelet transformed to an arbitrary depth, in contrast to JPEG, which uses an 8 8 block-size DCT. JPEG2000 uses following two different wavelet transforms: 1) Irreversible: The CDF 9/7 wavelet transform. It is said to be "irreversible" because it introduces quantization noise that depends on the precision of the decoder. 2) Reversible: A rounded version of the biorthogonal CDF 5/3 wavelet transform. It uses only integer coefficients so that the output does not require rounding (quantization) and therefore, it does not introduce any quantization noise. It is used in lossless coding. After the wavelet transform, the coefficients are scalar-quantized to reduce the amount of bits to represent them at the expense of the loss of quality. The output is a set of integer numbers which have to be encoded bit-by-bit. The parameter that can be changed to set the final quality is the quantization step: the greater the step, the greater is the compression and the loss of quality. With a quantization step that equals 1, no quantization is performed (it is used in lossless compression). Above process results in a collection of sub-bands which represent several approximation scales. A sub-band is a set of coefficients which represent aspects of the image associated with a certain frequency range as well as a spatial area of the image. The quantized subbands are split further into precincts, the rectangular regions in the wavelet domain. They are typically selected in a way that the coefficients within them across the sub-bands form approximately spatial blocks in the (reconstructed) image domain, though this is not a requirement. Precincts are split further into code blocks. Code blocks are located in a single sub-band and have equal sizes, except those located at the edges of the image. The 55

82 encoder has to encode the bits of all quantized coefficients of a code block, starting with the most significant bits and progressing to less significant bits by a process called Embedded Block Coding with Optimal Truncation (EBCOT). In this encoding process, each bit plane of the code block gets encoded in three so-called coding passes, first encoding bits (and signs) of insignificant coefficients with significant neighbors (i.e. with 1-bit in higher bit planes), then refinement bits of significant coefficients and finally coefficients without significant neighbors. The three passes are called Significance Propagation, Magnitude Refinement and Cleanup respectively. Here we are limiting the discussion of JPEG2000. More details of JPEG2000 can be found in [56, ]. 3.2 IMAGE QUALITY MEASURES Through out this thesis, we have used Peak Signal to Noise Ratio (PSNR) and Cross- Correlation (CC) to measure the quality of the images PEAK SIGNAL TO NOISE RATIO The phrase Peak Signal to Noise Ratio, often abbreviated PSNR, is an engineering term for the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation. Because many signals have a very wide dynamic range, PSNR is usually expressed in terms of the logarithmic decibel scale. The PSNR is most commonly used as a measure of quality of reconstruction in image compression etc. It is most easily defined via the Mean Squared Error (MSE). For two m n monochrome images I (x, y) and K (x, y), where one of the images is considered a noisy approximation of the other, MSE is defined as: (3.3) 56

83 The PSNR in terms of MSE is defined as: (3.4) Here, MAX I is the maximum pixel value of the image. When the pixels are represented using 8 bits per sample, value of MAX I is 255. More generally, when samples are represented using linear PCM with B bits per sample, maximum possible value of MAX I is 2 B -1. For color images with three RGB values per pixel, the definition of PSNR is the same except the MSE is the sum over all squared value differences divided by image size and by three. Typical values for the PSNR in image compression are between 30 and 40 db CORRELATION COEFFICIENT (CC) The correlation coefficient, a concept from statistics, is a measure of how well trends in the predicted values follow trends in past actual values. It is a measure of how well the predicted values, from a forecast model, "fit" with the real-life data. The correlation coefficient is a number between 0 and 1. If there is no relationship between the predicted values and the actual values, the correlation coefficient is 0 or very low. As the strength of the relationship between the predicted values and actual values increases, the value of correlation coefficient also increases. A perfect fit gives a coefficient of 1.0. Thus the higher value of correlation coefficient is better. It indicates the strength and direction of a linear relationship between two random variables. We can use CC calculation to know the distortion level in our extracted watermark from an attacked watermarked image. If A is the original watermark of size m x n, and B is the extracted watermark, then, in Matlab, we can compute CC using the built-in function CORR2 ( ) which computes the correlation coefficient r as given in Equation

84 (3.5) 3.3 TEST DATA For testing the performance of our proposed watermarking schemes for gray and colored digital images, we are using standard test images available on various test images databases available on WWW. Figure 3.11 shows the gray level test images of Lena, Mandrill, Pepper and Barbara. These are gray level images of 256 colors. Figure 3.11: Test images of Lena, Mandrill, Pepper and Barbara (Courtesy: SPIHT based Image Coding Incorporating Perceptual Model and Scalability, Figure 3.12 shows the 24 bit true color Windows BMP test images of Lena, Mandrill, Pepper and goldhill. 58

85 Figure 3.12: Test images of Lena, Mandrill, Pepper and Goldhill (Courtesy: Image Processing/Video Codecs/Programming Figure 3.13 shows the monochrome watermark logo, used in all proposed watermarking schemes discussed in this thesis. Figure 3.13: Watermark logo used in the proposed schemes 59

86

87 CHAPTER-4 WATERMARKING OF GRAY IMAGES 4.1 INTRODUCTION Like most DCT based watermarking schemes, Middle-Band Coefficient Exchange scheme has proven its robustness against those attacks, which anyhow, do not attack on the perceptual quality of image (Refer Section ). For example, JPEG compression reduces the size of image considerably without having much distortion in visual quality. Therefore, most of the DCT based schemes are robust against JPEG compression attack. But in most of the research literature available, even if quality of extracted watermark logo is good enough to prove the ownership, PSNR value of extracted watermark logo is less. In this chapter, we have explained how PSNR value of extracted logo from watermarked image could be increased if watermarked image has been attacked by JPEG compression attack. Then we developed a watermarking scheme to increase the robustness against Histogram equalization attack, which attacks on perceptual quality of image. After developing the watermarking schemes which are robust against JPEG compression and histogram equalization attack, we developed a watermarking scheme which is collusion attack resistant by introducing redundancy in Middle Band Coefficient Exchange scheme. This scheme is not only collusion attack resistant but more robust against JPEG compression attack as compared to other similar state-of-the-art watermarking schemes. 61

88 4.2 INCREASING THE ROBUSTNESS OF IMAGE WATERMARKING SCHEMES AGAINST JPEG COMPRESSION Two, classical DCT and DWT based watermarking schemes have already been discussed in Section and We watermarked the images of Lena, Mandrill and Pepper, which are shown in Figure 3.11, by applying both the schemes. While watermarking the chosen images, we used a monochrome logo as a copyright data (or watermark), which is shown in Figure Then, watermarked images, obtained by applying the above said watermarking schemes, were compressed by JPEG low compression (Quality factor, Q = 20). From the JPEG compressed images, the watermark data was recovered. As it has already been mentioned that DCT and DWT based watermarking schemes are robust against JPEG attack, we found that extracted watermark logo is quite detectible to prove the ownership as shown in Figure 4.1. (a) (b) Figure 4.1 (a): Extracted watermark logos from test images of Lena, Mandrill and Pepper by applying DCT based scheme (b): Extracted watermark logos from test images of Lena, Mandrill and Pepper by applying DWT based schemes Though, the extracted watermark logos are quite detectible, we can see the presence of noise in extracted watermark logos and therefore the PSNR values of extracted watermark logos are less. Therefore, there is a possibility to further improve the quality of the extracted watermark logos with an increased PSNR value of extracted watermark logos. 62

89 To achieve this, we propose to change the image data or image pixel values such that it has less impact of JPEG compression attack after getting watermarked without loosing the perceptual quality to a great extent. We thought to change or modify the image such that the affect after the attack on the watermarked image could be minimized. We tried to accomplish this by creating the same effect in an image, before watermarking it, which this image shall have, after it has been attacked. More precisely, if we know that our watermarked image may have to suffer JPEG compression attack, whatever changes will be made by JPEG attack in the watermark image, we tried to incorporate those changes in the pixel values in advance so that changes caused by JPEG compression attack may be minimized. This led to the preprocessing of the images, i.e., doing some modifications in the image which are equivalent to the attack before we start watermarking on it, either by using DCT or DWT based watermarking schemes. To implement the idea, we decided to analyze the JPEG compression attack on an image which has been watermarked by DCT and DWT based watermarking schemes. We proposed three transformation steps before the watermarking of an image, which are as follows: 1) Take the gray level image which has to be watermarked; 2) Compress it using JPEG scheme; and 3) Convert back the compressed image to gray level image to get the Transformed Image. We applied the above 3 transformation steps on our chosen test images. First, we generated 3 transformed images of Lena s test image by keeping the JPEG quality factor Q = 20, Q = 40 and Q = 60. Then, in the same way, we generated 3 transformed images of remaining 2 test images of Mandrill and Pepper also. Then, we watermarked transformed images as well as original images, using both schemes stated above. So, total 12 images were watermarked separately by DCT as well as DWT based watermarking schemes. For each of the 3 test images, 4 copies of it were watermarked where 1 copy was the original image and other 3 copies were the transformed 63

90 images, generated by our proposed preprocessing steps. All watermarked images were then compressed using JPEG low compression (Q = 20). After retrieving the watermark logos, it was found that the quality of extracted watermark logos from transformed images was better than the quality of extracted watermark logos from original images. Table 4.1 summarizes the PSNR values (in decibel) of extracted watermark logos. It may be observed that for the test image Lena, PSNR values of extracted logos were better from all 3 transformed images as compared to PSNR value of extracted logo from original image for both DCT as well as DWT based watermarking schemes. But for the test images Mandrill and Pepper, only 1 transformed image generated by keeping Q = 40, gave the batter PSNR value of extracted logo as compared the PSNR value of extracted logo from their original image for both DCT as well as DWT based watermarking schemes. Thus, we conclude that the preprocessing for a certain Q enhances the quality of extracted logos to some extent and, therefore, to increase the robustness of watermarking schemes against some well known attacks, we must analyze the attack s characteristics and its impact on the image and then adjust or preprocess the image in such a manner that the impact of the attack could be minimized. 4.3 INCREASING THE ROBUSTNESS OF IMAGE WATERMARKING SCHEME AGAINST HISTOGRAM EQUALIZATION ATTACK In the previous section, we had discussed about the preprocessing of an image to improve the robustness of DCT and DWT based watermarking schemes against JPEG compression. We know that transformed domain based watermarking schemes like DCT and DWT based schemes which were under our consideration in previous section, are robust against the attacks which do not change the perceptual quality of an image like JPEG compression attack. We have seen that by our proposed preprocessing, a watermarked image became more resistant to JPEG compression attack. We decided to 64

91 see the effectiveness of our proposed idea of preprocessing against those attacks which alter the image perceptually. So, we focused on the histogram equalization attack. If we equalize the histogram of an image, it is affected badly. We would now check whether our proposed idea of preprocessing works in the case of histogram equalization? Table 4.1: PSNR (in decibel) of extracted watermark logo from JPEG compressed (Q = 20) watermarked image Results given by watermarking of original image Results given by watermarking of transformed image. Image Scheme Used PSNR of Extracted Logo if Original Image is not Transformed PSNR of Extracted Logo if Original Image is Transformed using Q = 20 PSNR of Extracted Logo if Original Image is Transformed at Q = 40 PSNR of Extracted Logo if Original Image is Transformed at Q = 60 Lena DCT DWT Mandrill DCT DWT Pepper DCT DWT We watermarked the images of Lena, Pepper, Mandrill and Barbara, which are shown in Figure 3.11, by applying both DCT and DWT based watermarking schemes. While watermarking the chosen images, we used a monochrome logo as a copyright data (or watermark) which is shown in Figure Then, for all watermarked images obtained by applying the above said watermarking schemes, we equalized their histogram and then recovered the watermark data from the histogram equalized images. We found that extracted watermark logos were quite detectible to prove the ownership, as shown in Figure 4.2, but all were very noisy. We now preprocess the image through the following proposed 3 transformation steps before watermarking the images: 65

92 1) Take the gray level image to be watermarked; 2) Adjust the image such that its histogram is equalized to get the Transformed Image ; and 3) Apply watermarking DCT and DWT based schemes to the image obtained in step 2. We applied the above 3 transformation steps on our chosen test images. We generated transformed images of Lena, Pepper, Mandrill and Barbara test image. For each of the 4 test images, 2 copies of it (1 copy of the original image and other copy of the transformed images generated by our proposed preprocessing steps) were watermarked by DCT and DWT based watermarking schemes. The histograms of all watermarked images were then equalized. After retrieving the watermark logos, it was found that the quality of extracted watermark logos from transformed images was better than the quality of extracted watermark logos from original images. (a) (b) Figure 4.2 (a): Extracted watermark logos from test images of Lena, Mandrill, Pepper and Barbara by applying DCT based scheme (b): Extracted watermark logos from test images of Lena, Mandrill, Pepper and Barbara by applying DWT based schemes Table 4.2 summarizes the PSNR values (in decibel) of extracted watermark logos. It may be observed that, the watermark logos, extracted from watermarked transformed 66

93 images have PSNR values slightly better then the PSNR values of extracted logos retrieved by watermarked original image. Even if, PSNR values were increased slightly, considerable improvement in perceptual quality was observed. Figure 4.3 shows the extracted logos from histogram equalized attacked watermarked images. Logos at left sides are recovered form attacked watermarked original image, whereas logos at right side in the figure are recovered logos form watermarked transformed images. We can easily find that quality of extracted watermark logos from transformed images is better for all the 4 test images. Table 4.2: PSNR of extracted log from attacked test images PSNR (DB) Watermarking PSNR of Extracted PSNR of Extracted Logos scheme used Logos from Original from Transformed Image Images Image DCT Lena DWT DCT Pepper DWT DCT Mandrill DWT DCT Barbara DWT

94 Figure 4.3: Extracted logos from original image (left) and transformed image (right) of Lena, Mandrill, Pepper and Barbara s (Top to Bottom) histogram equalized images (By applying DCT based scheme) Therefore, we conclude that preprocessing the images to minimize the impact of histogram equalization attack, made the test images more robust against said attack if DCT and DWT watermarking schemes were used. This favors our statement made in the previous section that we must analyze the attack s characteristics and its impact on the image and then adjust or preprocess the image in such a manner that the impact of the attack could be minimized. 4.4 DEVISING A COLLUSION ATTACK RESISTANT WATERMARKING SCHEME FOR IMAGES USING DCT After developing a technique to make DCT and DWT based watermarking schemes (discussed in Section and ) more robust against JPEG compression and histogram equalization attacks, we considered a malicious attack, the collusion attack which was discussed in Section Seeing the financial implications of this attack, we propose a new term or benchmark for watermarking schemes, the ICAR i.e. 68

95 Inherently Collusion Attack Resistant. We recommend that any watermarking algorithm, by definition, must be collusion attack resistant in nature. A watermarking scheme must be first ICAR and then it should focus on other common image manipulations and malicious attacks. Henceforth, all watermarking schemes that we are present are ICAR in nature. The classical Middile Band Coefficient Exchange (MBCE) scheme, A DCT based scheme discussed in Section , is known to be robust against common image manipulations and JPEG compression attack. But this scheme, however, cannot sustain collusion attack. If, any attacker takes more than one copy of a watermarked image, then by analyzing the patterns of block DCT coefficients, attacker can easily predict the watermark location and watermark data. Our aim is to develop an ICAR watermarking scheme which can sustain other common image manipulations and known attacks also over the existing MBCE scheme which is not capable of sustaining collusion attack. For developing the new ICAR scheme, the following 2 issues were kept in mind: 1) If only one pair is used to hide the watermark data, it might happen that by an attack or by any image manipulation, values of this pair are modified. So, instead to exchanging only one pair of coefficients from FM region, we should exchange more than one pair of the coefficient i.e. introduce some redundancy; and 2) To achieve ICAR nature in watermarking scheme, we must ensure that every copy of watermarked image has a different pattern of hiding watermark data so that attacker can not conclude the location and content of watermark data even after analyzing many copies of watermarked image. Issue no.1 is resolved as follow: There are 22 coefficients in the FM region in an 8 x 8 DCT block. Out of these 22 coefficients, we can form 17 pairs having nearly the same values in their corresponding 69

96 JPEG quantization table. Therefore, to introduce redundancy in MBCE scheme, we had a choice to exchange the n pairs where the value of n ranges from 1 to 11(as there are total 22 coefficients). We can not disturb or modify all 22 coefficients as it will affect the image perceptibility. We conducted some experiment on this issue and found that if we modify the values of 8 coefficients (i.e. 4 pairs are exchanged), no much degradation in the image perceptibility is recorded. Accordingly, we decided to set the value of n equal to 4. Issue no. 2 is simply resolved by choosing the combination of 4 pairs randomly in each watermarked image. MBCE scheme exchanges 1 pair of coefficient from FM region to hide 1 or 0. For example, if coefficients at (3,2) and (2,3) are decided to hide the watermark data, this scheme sets DCT (3,2) > DCT (2,3) to interpret 1 and set DCT (3,2) < DCT (2,3) to interpret 0 by exchanging the coefficient values. While decoding the watermark data, MBCE scheme takes 8 x 8 DCT of watermarked image and by looking the relative strength of the coefficients at these locations, it decodes the 1 or 0 to reconstruct the watermark data. The proposed ICAR scheme exchanges 4 pairs which indicates that either 0 or 1 is hidden in the block. One such combination of 4 pairs may be taken as: {(5,1) and (4,2), (6,3) and (5,4), (5,2) and (4,3), (3,2) and (2,3)}. A scheme is robust if it is able to recover watermark data even if most of the middle band conefficients are attacked. To achieve this, we need to develop some dependencies on low frequecny coefficients also. In Figure 2.3 and Figure 2.4, values present at location (0,1) and (1,0) in 8x8 block DCT are low frequency coefficients of an image and attacker can not change the values at these locations because it will affect the image badly. 70

97 Figure 4.4: Swapping of 4 pairs to hide 0 or 1 in conjunction with low frequency values We developed a scheme of exchanging 4 middle-band coefficient pairs in strong correlation with low frequency coefficients such that even if attacker successfully attacks on 3 pairs, only 1 pair of coefficient will decode the watermark data correctly. Swapping criteria of the proposed scheme is illustrated in Figure 4.4. More details of encoding and decoding process are given in Section and The proposed watermarking scheme can be defined as a 7-tuple (X, W, P, K, G, E, D), where 1) X denotes the set of instances X i, of a particular gray level image, (If N copies of an image are to be watermarked, then 0 i N); 2) W denotes the monochrome watermark logo; 3) P denotes the set of policies P i, 0 i N; 4) K denotes the watermark strength parameter; 5) G denotes the policy generator algorithm G: X i P i, where each X i will have a unique P i, i.e. a different policy to hide the watermark data; 6) E denotes the watermark embedding algorithm, E: X i x W x P i X i ; 71

98 7) D denotes the watermark detection algorithm, D: X i x P i W, where W represents extracted watermark. Out of these 7 tuples, last 3 tuples are algorithms as discussed below: G, THE POLICY GENERATOR ALGORITHM To watermark each copy X i of an image X differently, we need a different watermarking policy. Here Policy means that for every copy of the image, there will be unique combination of 4 pairs of middle band coefficients. To generate a policy, we simply take 8 x 8 DCT of the input image X i and randomly select 4 pairs out of 17 pairs of middle band region. So, number of policies that can be generated are 17 C 4 = 2380 which means that 2380 copies of a single image can be watermarked such that no two watermarked images have same policy. This step ensures that attacker can not conclude the location of watermark data by colluding many watermarked copies of an image. This also depicts that our proposed scheme is an ICAR scheme E, THE WATERMARK EMBEDDING ALGORITHM In this algorithm, each 8 x 8 DCT block of an image is used to hide a single bit of watermark logo. This algorithm is given as below: 1. Repeat steps 2 to 13 for i = 1..n; // where n is the number of copies of a single image to be watermarked // 2. INPUT (X i ); 3. Take 8 x 8 block DCT of X i ; 4. INPUT (W); 5. Convert W into a string S = (S j S j = {0, 1}, for j = 1..length of the watermark); 6. Let L = STRING_LENGTH (S); // where L is the length of watermark data. If L=1000, then first 1000 DCT block of Xi are used // 72

99 7. P i = CALL (G); // Each Pi shall be stored in an author s database for the detection purpose in future. Let the Pi, for chosen Xi, be {(5,2) and (4,3), (6,3) and (5,4), (5,1) and (4,2), (3,2) and (2,3)} // 8. Repeat steps 9 to12 for r = 1..L; 9. Read S r ; 10. If S r = 0 If (DCT (0, 1) > DCT (1, 0)) Swap the DCT coefficients from chosen Pi such that coefficients at (5,2), (6,3), (5,1) and (3,2) become larger than (4,3), (5,4), (4,2) and (2,3) respectively; If (DCT (0, 1) <= DCT (1, 0)) Swap the DCT coefficients from chosen Pi such that coefficients at (5,2), (6,3), (5,1) and (3,2) become smaller than (4,3), (5,4), (4,2) and (2,3) respectively; Else If S r =1 If (DCT (0, 1) <= DCT (1, 0)) Swap the DCT coefficients from chosen Pi such that coefficients at (5,2), (6,3), (5,1) and (3,2) become larger than (4,3), (5,4),(4,2) and (2,3) respectively; If (DCT (0, 1) > DCT (1, 0)) Swap the DCT coefficients from chosen Pi such that coefficients at (5,2), (6,3), (5,1) and (3,2) become smaller than (4,3), (5,4), (4,2) and (2,3) respectively; End; 11. For all swapped coefficients pairs repeat the step 12; 12. If (DCT (u 1, v 1 ) DCT (u 2, v 2 ) > K) If (DCT (u 1, v 1 ) > DCT (u 2, v 2 )) DCT (u 1, v 1 ) = DCT (u 1, v 1 ) + K/2; 73

100 Else End; DCT (u 2, v 2 ) = DCT (u 2, v 2 ) - K/2; DCT (u 1, v 1 ) = DCT (u 1, v 1 ) - K/2; DCT (u 2, v 2 ) = DCT (u 2, v 2 ) + K/2; // Like Classical MBCE scheme (Section ), robustness of the watermark can be improved by using a watermark strength constant K such that for all 4 chosen pairs, DCT (u 1, v 1 ) DCT (u 2, v 2 ) > K. If coefficients do not meet these criteria, they should be modified by using some random noise to satisfy the relation. Increasing K thus reduces the chance of detection errors at the expense of additional image degradation. This ensures that larger coefficients remains larger even after image manipulations because coefficients relative values will decide the decoding of the watermark data // 13. Take IDCT to reconstruct X i ; 14. End D, THE WATERMARK DETECTION ALGORITHM We decode 1 and 0 based on the swapping criteria shown in Figure 4.4. The detection algorithm steps are as follows: 1. INPUT (X i ); // Xi is the attacked copy of a watermarked image // 2. Take 8 x 8 block DCT of X i ; 3. For each P i in author s database, repeat the steps 4; // If initially 10 copies were watermarked, then out of 10 policies, for 1 policy, watermark will be recovered correctly. To explain further steps, we are assuming that now algorithm is in a loop where Pi is {(5,2) and (4,3), (6,3) and (5,4), (5,1) and (4,2), (3,2) and (2,3)}, which was used to watermark this particular Xi // 4. Repeat the steps 5 for j = 1.L; 74

101 // L is the length of watermark data. A single bit will be recovered form one 8x8 DCT block // 5. Take j th DCT block to form j th bit of watermark as follows: If (DCT (1, 2) > DCT (2, 1)) If (DCT (5, 2) > DCT (4, 3)) T1 = 1; else T1 = 0; If (DCT (5, 1) > DCT (4, 2)) T2 = 1; else T2 = 0; If (DCT (6, 3) > DCT (5, 4)) T3 = 1; else T3 = 0; If (DCT (3, 2) > DCT (2, 3)) T4 = 1; else T4 = 0; If (T1 + T2 + T3 + T4 > 1) Decode 0 ; Else decode 1 ; Else If (DCT (1, 2) <= DCT (2, 1)) If (DCT (5, 2) > DCT (4, 3)) P1 = 1; else P1 = 0; If (DCT (5, 1) > DCT (4, 2)) P2 =1; else P2 = 0; If (DCT (6, 3) > DCT (5, 4)) P3 = 1; else P3 = 0; If (DCT (3, 2) > DCT (2, 3)) P4 = 1; else P4 = 0; If (P1 + P2 + P3 + P4 > 1) Decode 1 ; Else decode 0 ; End; 6. Store W, the recovered watermark; 7. End. 75

102 Even if three pairs are attacked to confuse the decoder, only one pair in conjunction with the relationship between DCT (1, 2) and DCT (2, 1), enables us the detection of 1 or 0. That is why the line (T1 + T2 + T3 + T4 > 1) is written. If there is no change in watermarked image, all values will remain unaffected and we can set the condition (T1 + T2 + T3 + T4 > 3) PERFORMANCE OF THE PROPOSED SCHEME To incorporate the ICAR nature, we have introduced redundancy and randomness in classical MBCE scheme. Because of this attacker has no mechanism to conduct pattern analysis to find out the location of the watermark data. Therefore we can say that the proposed scheme s design ensures that pattern analysis by colluding many watermarked copies is not possible and thus the scheme is ICAR. Now, in order to check that injecting the ICAR nature in the scheme did not degrade the performance against common image manipulations and known attacks, we tested our scheme on 3 well known test images of Lena, Mandrill and Pepper of size 512 x 512 and 256 colors in Windows BMP format as shown in Figure We generated the watermarked copies at various watermark strength constant K. Values of K were chosen from 10 to 50, and then for all watermarked copies, watermark logos were recovered. Obviously, for higher values of K, the quality of extracted watermark logos were fine but the quality of watermarked image itself, was affected much. On the other hand, for the lower values of K, the watermarked image generated were of finer quality but the quality of extracted watermark logos from such images was poor. This is an obvious Imperceptibility versus Robustness trade-off. It was observed that, the value K = 20 was the best value under the circumstances. For this value of K, the recovery was good without losing much image quality. So, further tests were conducted by using K = PERFORMANCE AGAINST JPEG COMPRESSION: All watermarked test images were compressed using JPEG scheme at various JPEG quality factors. Even with quality factor, Q = 20 (9.1 % of original size, JPEG Low compression), extracted logos were quite detectible. Table 4.3 summarizes the PSNR values of extracted watermark 76

103 logos from JPEG compressed watermarked images. Figure 4.5 shows the extracted watermark logos from JPEG compressed watermarked test images. It may be observed from both the specified table and the figure that our proposed scheme is capable of sustaining JPEG compression attack and even at Q = 20, the recovery of the watermark logo is quite efficient PERFORMANCE AGAINST COMMON IMAGE MANIPULATIONS: All watermarked test images were then tested against Horizontal flip, Scaling, Brightness / Contrast (both - 20 to + 20) adjustment and Noising. Our scheme sustained all above image manipulations. Figure 4.6 shows the extracted watermark logos recovered by the test image of Lena, which had undergone all the above stated attacks. Same results were found for other 2 test images also. Table 4.3: PSNR of extracted watermarks after JPEG compression PSNR (DB) Quality Lena Watermarked Mandrill Pepper factor Image Watermarked Image Watermarked Image Figure 4.5: Extracted watermark logos after JPEG compression at Q = 20 from watermarked Lena, Mandrill and Pepper images COMPARATIVE STUDY WITH OTHER MECHANISMS: We compared the performance of the proposed scheme for the JPEG compression with other similar 77

104 state-of-the-art methodologies which are well known for their robustness against JPEG compressions. Schemes chosen were as follows: Scheme-A: Correlation based Schemes with 1 PN sequence (Section ) Scheme-B: Correlation based Schemes with 2 PN sequence (Section ) Scheme-C: DCT Domain based Scheme (Section ) Scheme-D: DWT Based Scheme. (Section ) Watermarked images, obtained by proposed scheme as well as by other four schemes (Scheme-A to Scheme-D) were then compressed at various JPEG quality factors. We named our proposed scheme as Scheme-E. As all the above said watermarking schemes were robust against the JPEG compression attack, we evaluated them at different scale. All schemes were evaluated for how rapidly the scheme would start losing its robustness as the JPEG quality factors goes down. It was observed that up to Q = 40, performance of all watermarking schemes were approximately equal but for lower values of JPEG quality factor (Q < 40), our scheme showed more resistant as compared to scheme-a and scheme-b. The percentage decrease in quality of extracted watermark with respect to JPEG quality factors were compared as shown in Figure 4.7. It may be observed that performance of proposed scheme is better then Scheme A and Scheme B for low JPEG compression. Proposed scheme loses its performance as compared to DCT and DWT based schemes because we are increasing robustness against collusion attack (by making it ICAR) at the expanse of quality (by introducing redundancy). Figure 4.6: Extracted watermark logos from Lena s image after Horizontal flipped, scaled, brightness /contrast adjusted and Noising (Left to Right, Top to bottom) 78

105 Schema-A Schema-B Schema-C Schema-D Schema-E 0 Q80 Q60 Q40 Q30 Q20 Figure 4.7: Percentage decrease in quality of extracted watermark with respect to JPEG quality factor So, even after introducing redundancy in classical MBCS scheme to fight against collusion attack, quality of recovered watermark does not decrease very much as compared to Scheme-C and Scheme-D and better than Scheme-A and Scheme-B. We, therefore, conclude that our proposed ICAR watermarking scheme is quite robust against JPEG compression and common image manipulations for watermarking of gray BMP images. 4.5 CONCLUSION To summarize this chapter, we can say that if DCT and DWT based watermarking schemes discussed in Section and are to be used for the watermarking of a gray BMP image, then the image becomes more resistant to JPEG compression attack if we transform the original image to JPEG image at certain JPEG quality factor and then convert it back to gray level image. Similarly, if we preprocess the image in such a way that its histogram is equalized, then also an image become more resistant to histogram equalization attack for the same watermarking schemes. So, a modification in the image such that the affect after the attack on the watermarked image could be minimized, increases the robustness of schemes for DCT and DWT based watermarking schemes. Then, we developed a DCT based ICAR watermarking scheme which was very robust against JPEG compression attack and other common image manipulations. 79

106 80

107 CHAPTER-5 WATERMARKING OF COLOR IMAGES 5.1 INTRODUCTION After satisfactorily developing the watermarking schemes for gray level images, we focused on developing the watermarking schemes for the color images. The proposed ICAR watermarking scheme given in the previous chapter was chosen as a base as it has already proved its resistance to JPEG compression attack and other common image manipulations and performed at par with other state-of-the-art watermarking schemes. In this chapter, we conducted a study to find out the suitability of color channel (Red/Green/Blue) to hide the watermark data while using the DCT based watermarking scheme. We present an ICAR watermarking schemes for true colored BMP images. 5.2 PERFORMANCE ANALYSIS OF COLOR CHANNEL FOR DCT BASED IMAGE WATERMARKING SCHEME Initially, the suitability of color channel to hide a monochromatic watermark in a 24-bit color Window s BMP image while using classical MBCE watermarking scheme, was examined as MBCE scheme is the base scheme used to develop the proposed ICAR watermarking schemes. Four well known 24 bit colored test images of Lena, Pepper, Mandrill and Monarch (Size 512 x 512 pixels), shown in Figure 3.12 were taken. The watermark logo used is shown in Figure Then, the watermark logo was embedded in these 4 test images using the MBCE watermarking scheme. To analyze the performance of Red, Green and Blue channels, the watermark was embedded separately in R, G and B channels one by one. So, total 4 images were watermarked and each of these images ware watermarked thrice. 81

108 Table 5.1: PSNR of Extracted watermark from JPEG compressed watermark test images LENA.BMP PEPPER.BMP JPEG Compression Q = 20 Q = 40 Q = 60 Q = 80 Q = 20 Q = 40 Q = 60 Q = 80 RED GREEN BLUE MANDRILL.BMP MONARCH.BMP JPEG Compression Q = 20 Q = 40 Q = 60 Q = 80 Q = 20 Q = 40 Q = 60 Q = 80 RED GREEN BLUE RED (PSNR-5.86) GREEN(PSNR ) BLUE(PSNR -4.45) Figure 5.1: Recovered watermarks for Lena.bmp after jpeg attack at Q = 40 After watermarking the test images in all three color channels, we compressed all 12 watermarked images using JPEG compression at 4 JPEG quality factors (Q = 80, 60, 40, and 20) and then recovered the watermark logos from JPEG compressed images. We calculated the PSNR values of all these 12 x 4 = 48 extracted watermark logos. Table 5.1 summarizes their PSNR values. The recovered watermark logos from all 3 Lena s test images, if they were JPEG compressed at Q = 40, are shown in Figure 5.1. It was observed that for all test images, quality of extracted watermark logo was better, if watermark is embedded in GREEN channel for all JPEG quality factors. This can be justified as follows: JPEG uses the YCbCr color model. While converting from BMP to JPEG, following color transformation occurs: Y = x R x G x B Cb = x R x G x B (5.1) Cr = x R x G x B 82

109 Where Y' is the luminance component and Cb and Cr are the blue and red chrominance components. Y'CbCr is not an absolute color space. It is a way of encoding RGB information and the actual color displayed depends on the actual RGB colorants used to display the signal. It is clear from Equations 5.1 that G is multiplied by relatively larger coefficient and thus green channel should carry the watermark data for the better recovery if images are JPEG compressed after the watermarking using the MBCE scheme. Now to further validate the concept of preprocessing introduced in previous chapter, color channels of all test images were histogram equalized one at a time, i.e., Lena image had now 3 copies where in one copy only red channel is equalized, in another copy only green channel is equalized and in the third copy only blue channel is equalized leading to 12 test images to be watermarked. The watermark logo was embedded in the histogram equalized color channel for all 12 test images. We performed the following attacks on the watermarked images: 1) JPEG Attack (low JPEG compression with Q = 20); 2) Noise Attack (adding 10% Gaussian noise in the watermarked images); and 3) Histogram Equalization (equalizing the histogram of the watermarked images). The watermark logos were recovered from the attacked images and their PSNR values were calculated. Table 5.2 summarizes the PSNR values of watermark logos recovered. It may be observed from Table 5.2 that for all cases if a color channel of the image was HISTOGRAM EQUALIZED before embedding the watermark, recovery of watermark is better i.e. PSNR values are higher. Therefore, our proposed idea of preprocessing worked well for colored BMP images also. It may be further observed that the difference in the PSNR values of recovered logos from original image and equalized image are high in the case of histogram equalization attack because our preprocessing step is itself the histogram equalization. These results further prove that a modification in the image such that the effect after the attack on the watermarked image could be minimized, increases the robustness against that attack for colored images watermarking algorithm. 83

110 It is, therefore, concluded that to decide the color channel to carry the watermark data, we will have to analyze the characteristics of attack itself. If there is high probability that watermark image may undergo JPEG compression, we should select the GREEN channel because while converting to JPEG format, green channel s data has higher impact as compared to other color channel s data. Color Channel Table 5.2: PSNR of extracted watermark from attacked watermarked test images LENA.BMP PEPPER.BMP Attack Jpeg Q20 Histogram Equalizati on Noise (12.5%) Jpeg Q20 Histogram Equalizati on Noise (12.5%) Original RED Equalized Original GREEN Equalized Original BLUE Equalized MANDRILL.BMP MONARCH.BMP Color Channel Attack Jpeg Q20 Histogram Equalizati on Noise (12.5%) Jpeg Q20 Histogram Equalizati on Noise (12.5%) Original RED Equalized Original GREEN Equalized Original BLUE Equalized

111 It is also clear from Table 5.2 that for attacks other than JPEG Compression, performance of color channels for all images had no fixed pattern which means that robustness may depend upon the attack characteristics as well as image characteristics also. Therefore, the goal for the further development was not only to develop an ICAR watermarking scheme but also to find out some relationship between the performances of our proposed schemes with the image characteristics itself. 5.3 DEVISING AN ICAR WATERMARKING SCHEME FOR COLORED BMP IMAGES In the previous chapter, we have proposed an ICAR scheme for watermarking gray level image. Results indicated that this scheme was not only an ICAR scheme but also very robust to JPEG compression attack and other common image manipulations. Therefore, we decided to extend the same approach for colored BMP images also. In the earlier proposed ICAR scheme, we have introduced redundancy in swapping and made the swapping criterion dependent on low frequency coefficient. To further improve the robustness, we propose a new swapping criterion with the assurance that no two watermarked copies of an image have same policy of watermarking. An attacker may attack on large number of middle band coefficients but if image has to remain perceptually unchanged, the average value (Av) of all middle band coefficients (total 22 in numbers) will not modify to a great extent. So, unlike the previous scheme where we swapped 4 pairs, we swapped 4 middle band coefficients (not pair) with the Av value. Details of this swapping mechanism are described in Section The proposed watermarking scheme can be defined as a 7-tuple (X, W, P, T, G, E, D) where: 1) X denotes the set of instances X i of a particular gray level image, (If N copies of an image are to be watermarked, then 0 i N); 2) W denotes the monochrome watermark logo; 3) P denotes the set of policies P i, 0 i N; 85

112 4) T is the watermark strength parameter ; 5) G denotes the policy generator algorithm G: X i P i, where each X i will have a unique P i, i.e. a different policy to hide the watermark data. 6) E denotes the watermark embedding algorithm, E: X i x W x Pi X i ; 7) D denotes the watermark detection algorithm, D: X i x P i W, where W represents extracted watermark. The parameter T is analogous to K of classical MBCE scheme. In classical MBCE scheme, relative strength of 2 coefficient s value of FM region decides the decoding of 1 or 0. If the relative strength of 2 values has to decide the decoding of 0 or 1, then larger value should remain larger even after image manipulations. So, we adjust these values in such a way that the difference between the 2 values becomes larger than a certain threshold value. We name this threshold value as Watermark Strength Parameter because this value decides the robustness of watermark data. Certainly, it has an impact on the image perceptibly. So, we have to decide this threshold value in such a way that our image does not loose its quality much. Out of these 7 tuples, last 3 tuples are algorithms which are discussed below: G, THE POLICY GENERATOR ALGORITHM Similar to our earlier proposed ICAR watermarking scheme for the gray image watermarking, we had to watermark each copy X i of an image X differently. Therefore, we need a different watermarking policy for each copy of the image to be watermarked. Here Policy means that for every copy of the image, there will be unique combination of 4 middle band coefficients. To generate a policy, we simply take 8 x 8 DCT of the input image Xi and randomly select 4 coefficients out of 22 middle band coefficient of FM region from any of the red, green or blue color channel. So, numbers of policies that can be generated are 22 C 4 = 7315 which means that 7315 copies of a single image can be watermarked such that no two watermarked images have same policy. This step ensures that attacker can not conclude the location of watermark data by colluding many 86

113 watermarked copies of an image. This also depicts that our proposed scheme is an ICAR scheme COLOR CHANNEL SELECTION Up to the development of this approach, we used only BLUE color channel to hide the watermark data. Bossen et al. [9] have stated that the watermarks should be embedded mainly in the BLUE color channel of an image. The human eye is least sensitive to change in BLUE channel. However, the suitability of color channel to hide the watermark data is dependent on the image itself and therefore, we have discussed some interesting results related to this issue in the Chapter 6. In this section, we are using BLUE channel to hide the watermark data E, THE WATERMARK EMBEDDING ALGORITHM In this algorithm, each 8x8 DCT block of an image is used to hide a single bit of watermark logo. Our embedding algorithm is based on averaging the coefficients of FM region. We can fight against collusion attack by swapping more than one pair but if attacker is ready to loose some quality, he/she can disturb all the coefficients in FM region. Therefore, even if we introduce redundancy with randomness, our watermark data may still be attacked. So, we propose that an attacker cannot alter the image such that the average of coefficients of FM region changes much. Accordingly, we are hiding 1 or 0 by using relative value of a coefficients and the average Av of coefficients of FM region. This algorithm is given as below: 1. Repeat steps 2 to 11 for i = 1..n; // where n is the number of copies of a single image to be watermarked // 2. INPUT (X i ); 3. Take 8x8 block DCT of X i ; 4. INPUT (W); 5. Convert W into a string S = (S j S j = {0,1}, for j = 1..length of the watermark); 87

114 6. Let L = STRING_LENGTH (S); // L is the length of watermark data. If L = 1000, then first 1000 DCT block of Xi are used // 7. P i = CALL (G); // Each generated Pi shall be stored in an author s database for the detection purpose in future. Let the Pi for chosen Xi be, Pi = {(5,1), (4,2), (6,3) and (5,4)} in BLUE channel // 8. Calculate the average Av of remaining 18 middle band coefficients. 9. Repeat steps 10 to11 for r = 1..L; 10. Read Sr; // Now like classical MBCE scheme, relative strength of average Av and chosen 4 coefficients in step 7 will interpret 0 or 1 of watermark data. To hide 0, for all 4 chosen coefficients in step 7, we assigned the value of coefficients which is T less than the average Av. To hide 1, for all 4 chosen coefficients in step 7, we assigned the value of coefficients which is T greater than the average Av // If (S r = 0) DCT (5, 1) = Av - T; DCT (4, 2) = Av - T; DCT (5, 4) = Av - T; DCT (6, 3) = Av - T; Else DCT (5, 1) = Av + T; DCT (4, 2) = Av + T; DCT (5, 4) = Av + T; DCT (6, 3) = Av + T; End; 11. Take IDCT to reconstruct X i ; 12. End D, THE WATERMARK DETECTION ALGORITHM Watermark extraction is the reverse procedure of watermark embedding. To extract the watermark from the watermarked image, we calculated the average Av in the same way as in embedding algorithm. Owner should have a record of all policies used to watermark 88

115 the image. Based on policies, owner of the image can recover watermark using following rules: 1) If at least 1 out of 4 chosen coefficients are less than average, Interpret 0 ; and 2) If at least 1 out of 4 chosen coefficients are greater than average, interpret 1. The detection algorithm steps are as follows: 1. INPUT (X i ); // Xi is the attacked copy of a watermarked image // 2. Take 8x8 block DCT of X i and calculate Av; 3. For all P i in author s database, repeat the steps 4; // If initially 10 copies were watermarked, then out of 10 policies, for 1 policy, watermark will be recovered correctly. To explain further steps, we are assuming that now algorithm is in a loop where Pi is {(5,1) (4,2) (5,4) and ( 6,3)}, which was used to watermarked this particular Xi // 4. Repeat the steps 5 for j = 1.L; // L is the length of watermark data. A single bit will be recovered form one 8x8 DCT block// 5. Take j th DCT block to form j th bit of watermark as follows: If (DCT (5, 1) <= Av) T1 = 1; Else T1 = 0; If (DCT (4, 2) <= Av) T2 = 1; Else T2 = 0; If (DCT (5, 4) <= Av) T3 = 1; Else T3 = 0; If (DCT (6, 3) <= Av) T4 = 1; Else T4 = 0; If ( T1 + T2 + T3 + T4 >= 1) 89

116 Decode 0 If (DCT (5, 1) > Av) P1 = 1; Else P1 = 0; If (DCT (4, 2) > Av) P2 = 1; Else P2 = 0; If (DCT (5, 4) > Av) P3 = 1; Else P3 = 0; If (DCT (6, 3) > Av) P4 = 1; Else P4 = 0; If ( P1 + P2 + P3 + P4 >= 1) Decode 1 End; 6. Store W, the recovered watermark; 7. End. It may be observed from both the algorithms that even if attacker alters the values of the coefficient of FM region, if Av is not changed much, then we can recover the watermark data correctly and attacker cannot aim to attack the image in such a manner which modifies Av PERFORMANCE OF THE PROPOSED SCHEME Our proposed scheme does not need any testing to check whether or not it is robust against the collusion attack, as it is designed in such a way that the attacker can not analyze the pattern by colluding many watermarked copies. We needed to check the performance of the proposed scheme against the JPEG compression and other common image manipulations and known attacks. For this, we tested our scheme on 3 test images 90

117 Lena, Mandrill and Pepper of size 512 x 512 in Windows 24 bit BMP format, shown in Figure Firstly, we chose an appropriate value of T which affects least the image quality as well as optimizes the recovery of watermark data. Our experiments suggested that if we were hiding the watermark using T = 150, there was approximately no loss in the perceptual quality of the images and recovered watermark logos were of very fine quality. Figure 5.2 shows the watermarked test images after hiding watermark logo by keeping T = 150. It may be seen that, images are not disturbed at all. Figure 5.3 shows the extracted watermark logos from these watermarked copies of Lena, Mandrill and Pepper without performing any attack or manipulations on the watermarked images. This fixed up the value of T = 150 for further tests. Figure 5.2: Watermarked test images keeping T = 150 Figure 5.3: Extracted watermark from watermarked Lena, Mandrill and Pepper images respectively at T = PERFORMANCE AGAINST JPEG COMPRESSION: We applied JPEG compression on watermarked images (generated by keeping T = 150) at different JPEG quality parameters Q and then recovered the watermark logos. Table 5.3 summarizes the 91

118 PSNR of extracted watermark logos. It may be observed from Table 5.3 that even at Q = 20, quality of extracted watermark is very fine and logos are quite detectible PERFORMANCE AGAINST COMMON IMAGE MANIPULATIONS: We performed the following attacks on the watermarked test images: Attack-1: Equalize the Histogram; Attack-2: Apply uniform scaling (Zoom); Attack-3: Adjust the brightness to +40 and contrast to +25; Attack-4: Adjust the hue and saturation to +10 each; Attack-5: Add 10 % Gaussian noise; and Attack-6: Blur the image using Gaussian blur with 1 pixel radius. Table 5.3: PSNR of extracted watermark logos after JPEG compression PSNR (DB) Quality factor Lena Watermarked with T = 150 Mandrill Watermarked with T = 150 Pepper Watermarked with T = 150 Q = Q = Q = Q = Then, we recovered the watermark logos from attacked images and calculated the PSNR value of watermark logos. Table 5.4 summarizes the PSNR values of extracted logos recovered from all test images. Our proposed scheme sustained all the attacks and the quality of the extracted watermark logos is quite good. Figure 5.4 shows the recovered logos from attacked images. 92

119 Table 5.4: PSNR of extracted watermark logo from watermarked test images after attacks PSNR (DB) Histogram Equalization Zoom Brightness- Contrast Adjustment Hue- Saturation Gaussian Noise Gaussian Blur Lena Mandrill Pepper COMPARATIVE STUDY RESULTS WITH OTHER SCHEMES: We compared our scheme against JPEG compression with other similar and state-of-the-art methodologies which are well known for their robustness against JPEG compressions. The chosen schemes are as follows: Scheme-A: Correlation based Schemes with 2 PN sequence (Section ) Scheme-B: Classical MBCE Scheme (Section ) Scheme-C: Scheme proposed in Section 4.4 is also based on Middle Band Coefficient Exchange (MBCE) scheme and ICAR in nature. So, we decided to compare the performance of our scheme with this scheme also. This scheme swaps 4 pairs of coefficients in FM region in correlation with low band coefficients. We are naming this scheme as Scheme-C. Then, we re-implemented the chosen schemes for the colored images and hid the watermark data in BLUE channel. Scheme-D: We are naming our proposed scheme as Scheme-D. It is observed that all the above schemes are robust against JPEG compression attack but if we compress the watermark images by very low quality factors (less then Q = 20), our proposed scheme outperforms the other schemes. We compressed the watermarked test 93

120 images by keeping JPEG quality factor Q = 15, 10, and 5. No scheme, other than the proposed one, was able to extract the detectible watermark logos. Figure 5.4: Recovered logos from attacked images Table 5.5 summarizes the PSNR values of extracted logos from highly compressed watermark test images using various schemes. Figure 5.5 shows the recovered watermark logos from highly compressed watermarked images using our proposed scheme. It may be observed that recovered logos are quite detectible and proposed scheme is more efficient than the other chosen schemes. 94

121 Table 5.5: PSNR values of extracted logos from highly compressed watermarked test images using various schemes PSNR (DB) JPEG Quality Lena Mandrill Pepper Factors Schemes Scheme-A Scheme-B Scheme-C Scheme-D Q = Q = Q = Q = Q = Q = Q = Q = Q = Q = Q = Q = Test Images / JPEG Q Factor Lena Mandrill Pepper Q = 15 Q = 10 Q = 05 Figure 5.5: Extracted logos using proposed scheme from highly compressed watermarked test images 95

122 Results indicate that proposed scheme recovers the watermark even from an attacked image which is compressed up to Q = 5 quality factor of JPEG (i.e. after 95-99% size reduction). This proves that the proposed scheme is not only an ICAR scheme but also very robust to JPEG compression. In addition to this, the proposed scheme is resisting common image manipulations like cropping, scaling, flipping, histogram equalization, brightnesscontrast adjustment, Hue-saturation alteration, Gaussian noise and Gaussian blur. 5.4 CONCLUSION In this chapter, we have discussed the watermarking of the colored images. Since a colored image has R, G and B color channel, firstly we presented a study to find the suitability of a color channel to carry the watermark data with respect to the robustness against an attack. It was found that if an image has to undergo JPEG compression attack, then the watermark data should be hidden in GREEN color channel to ensure the best recovery of the watermark logo. Then, we presented an ICAR watermarking scheme based on the average of the FM coefficients. Results indicted that the proposed scheme is very robust against JPEG compression and common image manipulations and better then other similar state-of-the-art schemes. 96

123 CHAPTER-6 WATERMARKING OF JPEG IMAGES 6.1 INTRODUCTION In the Chapter 4, we have discussed that we can improve the robustness of DCT and DWT based watermarking schemes against some well known attacks by preprocessing the images. Since, Fingerprinting is the most crucial demand of today, we developed an ICAR scheme for the watermarking of gray level images also. We further expanded our scope for the colored images watermarking in Chapter 5 and developed an ICAR scheme for watermarking of 24-bit colored BMP images. Since, most of the images present on World Wide Web are in JPEG format, which is a highly compressed image format and store the images in the transformed domain, i.e. store the frequencies not the pixels values, we decided to develop an ICAR watermarking scheme for JPEG images. We also explored a relationship between the robustness and some of the image characteristics. 6.2 DEVELOPMG AN ICAR WATERMARKING ALGORITHM FOR JPEG IMAGES Most of the images present on WWW are in the Joint Photographic Experts Group (JPEG) format where as relatively less work is found for watermarking the JPEG images. Therefore, we decided to extend our earlier proposed ICAR schemes for the watermarking of JPEG images also. In our earlier proposed ICAR schemes, we inserted the ICAR nature in by introducing redundancy in the coefficients swapping of FM region. We also made the swapping criteria dependent on some very robust data elements (in the scheme presented in Section 4.4, it was the relative value of low frequency coefficient and in the scheme presented in 5.3, it was the average value of all middle band coefficients) so that decoding algorithm may perform a good recovery of the watermark data. But as it may be observed that we deployed the coefficients of FM region which were generated by taking the 8 x 8 DCT of pixels values. So, to continue the same 97

124 approach for the JPEG images, we needed to use coefficients belonging to FM region. More pricelessly, JPEG image format does not store the pixel s actual value but it stores the image in frequency domain. So, we need to convert the JPEG image into spatial domain and then take 8x8 block DCT on its color channels to get the FM region. To inject the ICAR nature, we need to introduce redundancy in coefficient swapping. Since JPEG is a very high compressed format, we know that as soon as we convert this spatial domain image into JPEG format, lots of its coefficients will be changed. This would create problem in recovering the watermark data by only considering the relative strengths of coefficients of FM region. We must, therefore, provide extra robustness by involving some coefficients whose value does not change much during the conversion of spatial domain to frequency domain and vise versa. To resolve this issue, we decided to take the advantage of JPEG compression-decompression scheme itself. In an 8x8 DCT block, large value of the top-left corner is called the DC coefficient. The remaining 63 coefficients are called the AC coefficients. This DC coefficient is the major dominating value while decompressing. This DC value alone can regenerate the best approximated image by taking the IDCT. If this value is altered, then image is largely affected. So we decided to take the contribution of this DC coefficient apart from coefficients from FM region to interpret the watermark data to make our scheme robust. We have seen that in our earlier scheme, we developed a swapping criteria based on the average of all 22 coefficients of FM region by claiming that it was difficult for any attacker or for any image manipulation to alter this value significantly if the image has to remain perceptually similar. Therefore, for our newly proposed watermarking scheme for JPEG images, we interpreted the watermark data in FM region based on the average of 22 coefficients from FM region and the DC coefficient. More details of the watermark embedding algorithms are described in Section To ensure ICAR property, liker our earlier proposed schemes, we watermarked each copy of a single JPEG image with a different policy. The proposed watermarking scheme can be defined as a 7-tuple (X, W, P, T, G, E, D) where: 98

125 1. X denotes the set of instances X i, of a particular JPEG image, (If N copies of an image are to be watermarked, then 0 i N); 2. W denotes the monochrome watermark logo; 3. P denotes the set of policies P i, 0 i N; 4. T is the watermark strength parameter ; 5. G denotes the policy generator algorithm G: X i P i, where Each X i will have a unique P i, i.e. a different policy to hide the watermark data; 6. E denotes the watermark embedding algorithm, E: X i x W x P i X i ; 7. D denotes the watermark detection algorithm, D: X i x P i W, where W represents the extracted watermark. The parameter T is analogous to K of classical MBCE scheme. In classical MBCE scheme, relative strength of two coefficients value of FM region decides the decoding of 1 or 0. If the relative strength of two values has to decide the decoding of 0 or 1, then larger value should remain larger even after image manipulations. So, we adjust these values in such a way that the difference between the two values becomes larger than a certain threshold value. We name this threshold value as Watermark Strength Parameter because this value decides the robustness of watermark data. Certainly, it has an impact on the image perceptibly. So, we need to decide this threshold value in such a way that our image does not loose its quality much. The value of T may differ for each image. Out of these 7 tuples, last 3 tuples are algorithms, which are discussed below: G, THE POLICY GENERATOR ALGORITHM Similar to our earlier proposed ICAR watermarking scheme for the gray image watermarking and colored image watermarking, we need to watermark each copy X i of an JPEG image X differently. Therefore, we need a different watermarking policy for 99

126 each copy of the image to be watermarked. Here Policy means that, for every copy of the image, there will be unique combination of 4 middle band coefficients. First we had to convert the source JPEG image into its equivalent true colored 24-bit BMP image. Then, to generate a policy, we simply take 8 x 8 DCT of a chosen color channel of the input image X i and randomly select 4 coefficients out of 22 middle band coefficient of FM region from any of the red, green or blue color channel. So, numbers of policies that can be generated are 22 C 4 = 7315 which means that 7315 copies of a single image can be watermarked such that no two watermarked images have same policy. This step ensures that attacker can not conclude the location of watermark data by colluding many watermarked copies of an image. This also depicts that our proposed scheme is an ICAR scheme. Policy generator algorithm also returns the color channel to be used to carry the watermark COLOR CHANNEL SELECTION: Bossen et al. [9] have stated that the watermarks should be embedded mainly in the BLUE color channel of an image because human eye is least sensitive to change in BLUE channel. However, the suitability of color channel to hide the watermark data depends on the image itself. The color channel which should be used can be found on the basis of the amount of the color present in the image or on the basis of histogram of each color channel (i.e. color with spreader histogram should be given priority). We also know that for few images, BLUE channel may not give the optimum results. We, therefore propose that the color channel with the lowest Standard Deviation (SD) should be selected. More details of this finding and result related to this issue are given in the Section E, THE WATERMARK EMBEDDING ALGORITHM In this algorithm, each 8x8 DCT block of an image is used to hide a single bit of watermark logo. Our embedding algorithm is based on averaging the coefficients of F M region and the DC coefficient. As we know that attacker cannot alter this average (Av) of coefficients of FM region and the DC coefficient badly as it will heavily impact the quality of image, we are hiding 1 or 0 by using the relative values of four coefficients with this Av. 100

127 This algorithm is given as follows: 1. Repeat steps 2 to13 for i = 1..n; // where n is the number of copies of a single image to be watermarked // 2. INPUT (X i ); 3. Convert the X i into its equivalent spatial domain 24-bit colored image; 4. Take 8 x 8 block DCT of X i ; 5. INPUT (W); 6. Convert W into a string S = (S j S j = {0,1}, for j = 1..length of the watermark); 7. Let L = STRING_LENGTH (S); // L is the length of watermark data. If L = 1000, then first 1000 DCT block of Xi are used // 8. P i = CALL (G); // Each generated Pi shall be stored in an author s database for the detection purpose in future. Let the Pi for chosen Xi be, Pi = {(5,1), (4,2), (6,3) and (5,4)} in the chosen color channel // 9. Calculate the average Av of remaining 18 middle band coefficients and DC coefficient. Av = (DCT (0, 0) + Sum (22 Middle band coefficients) - Sum (4 chosen coefficients chosen by P i )) / Repeat steps 11 to13 for r = 1..L; 11. Read S r ; // Now like classical MBCE scheme, relative strength of average Av and chosen 4 coefficients in step 7 will interpret 0 or 1 of watermark data. To hide 0 for all 4 chosen coefficients in step 7, we assigned the value of coefficients which is T less than the average Av. To hide 1, for all 4 chosen coefficients in step 7, we assigned the value of coefficients which is T greater than the average Av // If (S r = 0) DCT (5, 1) = Av - T; DCT (4, 2) = Av - T; DCT (5, 4) = Av - T; 101

128 DCT (6, 3) = Av - T; Else DCT (5, 1) = Av + T; DCT (4, 2) = Av + T; DCT (5, 4) = Av + T; DCT (6, 3) = Av + T; End; 12. Take IDCT to reconstruct X i ; 13. Convert X i back to its JPEG format; 14. End D, THE WATERMARK DETECTION ALGORITHM Watermark extraction is the reverse procedure of watermark embedding. To extract the watermark from the watermarked JPEG image, first we convert it into its equivalent 24 bit colored images and then calculate the average Av in a same way, as in embedding algorithm. Owner has a record of all policies used to watermark the images. Based on policies ; owner of the image can recover watermark using following rule: 1) If at least 1 out of 4 chosen coefficients are less then Av, Interpret 0 ; and 2) If at least 1 out of 4 chosen coefficients are greater then Av, interpret 1. The detection algorithm steps are as follows: 1. INPUT (X i ); // Xi is the attacked copy of a watermarked image// 2. Convert X i into its equivalent 24 bit colored image; 3. Take 8x8 block DCT of X i and calculate Av; 4. For all P i stored in author s database, repeat the steps 5; // If initially 10 copies were watermarked, then out of 10 policies, for 1 policy, watermark will be recovered correctly. To explain further steps, we are assuming that now algorithm is in a loop where Pi is {(5, 1) (4, 2) (5, 4) and (6, 3)}, which was used to watermarked this particular Xi // 102

129 5. Repeat the steps 5 for j = 1.L; // L is the length of watermark data. A single bit will be recovered form one 8x8 DCT block.// Take j th DCT block to form j th bit of watermark as follows: If (DCT (5, 1) < = Av) T1 = 1; Else T1 = 0; If (DCT (4, 2) < = Av) T2 = 1; Else T2 = 0; If (DCT (5, 4) < = Av) T3 = 1; Else T3 = 0; If (DCT (6, 3) < = Av) T4 = 1; Else T4 = 0; If ( T1 + T2 + T3 + T4 > = 1 ) Decode 0 If (DCT (5, 1) > Av) P1 = 1; Else P1 = 0; If (DCT (4, 2) > Av) P2 = 1; Else P2 = 0; If (DCT (5, 4) > Av) P3 = 1; Else P3 = 0; If (DCT (6, 3) > Av) P4 = 1; Else P4 = 0; If ( P1 + P2 + P3 + P4 > = 1) 103

130 Decode 1 ; End; 6. Store W, the recovered watermark; 7. End. It may be observed from both the algorithms that even if attacker alters the values of the coefficient of FM region, if Av is not changed much, then we can recover the watermark data correctly and attacker cannot aim to attack the image in such a manner which modifies the Av PERFORMANCE OF THE PROPOSED SCHEME Our proposed scheme does not need any testing to check whether or not it is robust against the collusion attack as it is designed in such a way that the attacker can not analyze the pattern by colluding many watermarked copies. We needed to check the performance of the proposed scheme against the JPEG compression and other common image manipulations and known attacks. We have tested our scheme on four JPEG test images of Lena, Mandrill, Pepper and Goldhill shown in Figure 3.12 and watermark logo is shown in Figure We measured the image quality in terms of Peak Signal to Noise Ratio (PSNR) and Correlation Coefficient (CC). Firstly, we choose an appropriate value of T which affects least the image quality as well as optimizes the recovery of the watermark data. Based on our earlier experiences discussed in Section 5.3.5, we embedded the watermark logo in test images by keeping T = 150 (in blue color channel) and then recovered watermark logos. Our experiments suggested that in Lena, Mandrill and Pepper test images, there was, almost no loss in the perceptual quality of the images (as shown in Figure 6.1) and recovered watermark logos were of very fine quality. Figure 6.2 shows the watermark logos obtained from Lena, Mandrill, Pepper and Goldhill. It was observed that for Goldhill test image, recovery was not good. Therefore, we continued to experiment the same process for the Goldhill test image at various values of T and we found that at T = 100, Goldhill test image was giving the best recovered logo without much loosing its perceptibility. Figure 6.3 shows the 104

131 goldhill test image after the watermark logo was embedded and the recovered logo. Therefore, considering the imperceptibility versus Robustness trade-off, we fixed up the value of T = 150 for the further tests for Lena, Mandrill, and Pepper JPEG test images, and T = 100 for the Goldhill test image. Figure 6.1: Watermarked test images generated by keeping T = 150 Figure 6.2: Extracted watermark logos from watermarked Lena, Mandrill, Pepper and Goldhill test images respectively at T = 150 Figure 6.3: Goldhill test image after hiding the watermark logo and the recovered logo at T = COLOR CHANNEL SELECTION AND PERFORMANCE AGAINST JPEG COMPRESSION: Standard deviation (SD) depicts the spread of the frequency values in a range. If the histogram of a chosen color channel of a particular image has less spread, the image has less number of frequencies of the chosen color channel. Since, it is the color channel i.e. the particular color frequencies that actually carry the watermark data, we conclude that SD must play an important role. To explore the relationship 105

132 between the selection of a color channel to carry the watermark data and the efficiency of recovery, we decided to experiment on SD of all three color channels. Table 6.1 shows the standard deviation of all three color channels for test images. Table 6.1: SD values of color channels for test images Lena Mandrill Pepper Goldhill R channel G channel B channel First, we hid the watermark data in the BLUE channel of all four test images. Then, we compressed watermarked images using JPEG technique at various quality factors and then recovered the watermark logos. We calculated the PSNR and CC values of extracted logo. Table 6.2 summarizes the results. It was found that extracted watermark from Mandrill and Goldhill test images were having poor values of PSNR and CC. Therefore, for these two images, we repeated the above process by using GREEN Channel. The qualities of the extracted watermark logos from these two images were improved. Therefore, we have related the performance of our scheme with color channel selection. As, it may be observed from the Table 6.1 that for Lena s and Pepper s test images, BLUE channel have lesser SD, whereas for Mandrill s and Goldhill s images, GREEN channel has lesser SD. So it was concluded that lesser the SD better is the recovery of the watermark data. This fixed up the BLUE channel for Lena s and Pepper s watermarking and GREEN channel for rest two images. It is clear from Table 6.2 and Table 6.3 that after using GREEN channel for Mandrill s and Goldhill s images, performance was increased. It may be further observed from Table 6.3 that our proposed scheme is quite robust against JPEG compression PERFORMANCE AGAINST IMAGE MANIPULATIONS: We performed the following attacks on the watermarked test images: 106

133 Attack-1: Equalize the Histogram; Attack-2: Add 10 % Uniform noise; Attack-3: Adjust the brightness to + 40 and contrast to + 25; Attack-4: Adjust the hue and saturation to + 10 each; Attack-5: Flip Horizontal; and Attack-6: Apply uniform scaling (Zoom). Our proposed scheme sustained all the attacks and qualities of extracted watermark logos were very fine. Table 6.4 summarizes the CC of extracted logos from all test images. Figure 6.4 shows the recovered logos from attacked images. Table 6.2: PSNR and CC of extracted logo by using BLUE channel for all images JPEG Quality Factor Lena Mandrill Pepper Goldhill PSNR Q = 60 CC PSNR Q = 40 CC PSNR Q = 20 CC

134 Table 6.3: PSNR and CC of extracted logo by using BLUE and GREEN channels for images JPEG Quality Factor Lena (BLUE), T = 150 Mandrill (GREEN) T = 150 Pepper (BLUE) T = 150 Goldhill (GREEN) T = 100 PSNR Q = 60 CC PSNR Q = 40 CC PSNR Q = 20 CC COMPARATIVE STUDY WITH SIMILAR, STATE-OF-THE-ART SCHEMES: We compared the performance of the proposed scheme against JPEG compression with other similar schemes which are DCT based and well-known for their robustness against JPEG compression. Test Images /Attacks Table 6.4: CC of the extracted logos Lena Mandrill Pepper (BLUE), (GREEN), (BLUE), T = 150 T = 150 T = 150 Goldhill (GREEN), T = 100 Histogram Equalization Uniform Noise (10%) Brightness (+ 40) & Contrast (+ 25) Hue and saturation adjust (10 each) Horizontal Flip Uniform scaling

135 Figure 6.4: Extracted logos from attacked watermarked images The schemes chosen were: Scheme-A: Correlation based Scheme (Section ) Scheme-B: The Classical Middle Band coefficient exchange scheme (Section ) Scheme-C: Collusion attack resistant watermarking scheme (Section 4.4): Scheme proposed in Section 4.4 is also based on MBCE scheme and ICAR in nature. This scheme swaps 4 pairs of coefficients in F M region in correlation with low band coefficients. We are naming this scheme as Scheme-C. Scheme-D: We named our proposed scheme as Scheme-D. We re-implemented the first three chosen schemes ideas for JPEG colored images. In their work A Novel DCT-based Approach for Secure Color Image Watermarking [7] author have compared their proposed scheme against JPEG compression with Tsai [102], cox [19], Fridrich [28] and Koch [48] approaches but they have given the results only up to JPEG Quality factor Q = 20. Therefore, we compared our proposed scheme for very 109

136 less JPEG quality factors such as Q = 5 and Q = 10. Most of the schemes started loosing their efficiency at these quality factors. We conclude that all the above schemes were very robust against JPEG compression attack but if we compressed the watermark images at very low quality factors (less than Q = 20), our proposed scheme outperformed the other schemes. No scheme, other than the proposed one, was able to extract the detectible watermark logo at Q = 10 and 5. Figure 6.5 shows the graph of CC values of recovered logos obtained from JPEG compressed (at Q = 10) images which were watermarked using various schemes. Figure 6.6 shows the graph of CC values obtained from JPEG compressed (at Q = 5) images. Therefore, the proposed scheme is not only an ICAR scheme but also enhances the performance. Results indicate that the proposed scheme recovers the watermark even from highly attacked images which are compressed up to Q = 5 quality factor of JPEG (i.e. after 95-99% size reduction). In addition to this, the proposed scheme is resisting common image manipulations like cropping, scaling, flipping, histogram equalization, brightness- contrast adjustment, hue-saturation alteration and Gaussian noise Lena (Blue) Mandrill ( Green) peper (Blue) Goldhill (Green) Scheme-A Scheme-B Scheme-C Scheme-D Figure 6.5: Comparison of correlation coefficients at Q =

137 Lena (Blue) Mandrill ( Green) peper (Blue) Goldhill (Green) Scheme-A Scheme-B Scheme-C Scheme-D Figure 6.6: Comparison of correlation coefficients at Q = A DWT BASED WATERMARKING SCHEME FOR JPEG IMAGES During the development of the above schemes, popularity of JPEG2000 image compression/encoding increased. JPEG2000 is a wavelet-based image compression standard. This standard has also been created by the Joint Photographic Experts Group committee in the year 2000 with the intention of superseeding their original DCT based JPEG standard (created in the year 1991). The standardized filename extension for JPEG2000 image is.jp2. JPEG 2000 has a much more significant advantage over certain modes of JPEG in that the artifacts are less visible and there is almost no blocking. The compression gains over JPEG are attributed to the use of DWT and a more sophisticated entropy encoding scheme. Since.jp2 format is new upcoming image format and very less watermarking efforts have been presented against this format conversion in the literature, we need to focus this attack because BMP and JPEG images may have to undergo.jp2 image format conversion/compression. To ensure that our watermarked images do not lose their robustness against JPEG2000 format conversion attack, we further develop a watermarking scheme which can sustain JPEG2000 format conversion attack also. 111

138 6.3.1 EXPLORATION OF DWT DOMAIN Till now, all of our proposed watermarking schemes are DCT based, and therefore, very robust against JPEG compression attack because JPEG encodes the images using DCT. Both, DCT and DWT encode (or compress) the image very differently. Since JPEG2000 encodes the image using DWT, a DCT based scheme may not be fruitful if we are targeting.jp2 conversion attack resistant nature in our watermarking scheme. Our earlier results of preprocessing (Sections 4.2) also supported this fact. So, we decided to explore DWT domain for the watermarking of JPEG images ISSUES IN USING DWT: Because of their inherent multi-resolution nature, wavelet-coding schemes are especially suitable for applications where scalability is important. The use of DWT is gaining popularity in signal processing, image compression and watermarking. DWT gives extremely good results in the case of lossless compression. But DWT has a serious issue when it comes to comparison with DCT for the watermarking purposes. We cannot assume lossless manipulation in images; both in watermark embedding and while the image is being attacked. In watermarking, one has to ensure that the watermark data is recoverable even from highly destroyed/manipulated/compressed/lossy cover image. Now, while using DCT domain, in most of the cases, we take 8 x 8 DCT and thus have hundreds of DCT blocks. In each DCT block, there are FL, FM and FH regions, as shown in Figure 2.3. We cannot use FH because any manipulation operation will attack first on FH. FL has the major dominating coefficient to recreate the image. If we use FL to hide the watermark data, cover image perceptibility will be affected seriously. Therefore, we use FM region, or since, there are so many FL regions, we can work out to devise a watermarking scheme that takes FL region also into consideration without changing FL coefficient values. On the other hand, DWT takes the complete image into consideration as shown in Figure 6.7 and breaks it into four parts, namely LL, HL, LH, and HH region. This policy may have several advantages but for watermarking, it has a very serious issue. Like DCT blocks, we should not use HH region. LL region coefficients can also not be altered much because these will heavily affect the image perceptibility (LL coefficients will alone 112

139 generate a very good approximated image and we cannot alter these coefficients much). HL and LH coefficients may be altered seriously by any image manipulation operation. Unlike DCT based transformation (where there are so many FM regions to hide the watermark data), there is only one LL region in DWT. Therefore, we have very less space to hide the watermark data. Either we disturb heavily DWT coefficients and thus affect the image perceptibility while hiding watermark data or to preserve to image quality, hide watermark data in those regions which are less susceptible to get modified by image manipulation operations and thus affecting the robustness of the watermarking scheme. We thus conclude that if we use DWT for watermarking purpose, Imperceptibility vs. Robustness balance is the new challenge for us. More precisely, the classical CDMA- DWT based scheme as given in Section , a highly referred scheme which is very robust against JPEG compression, affects the image quality up to a great extent. On the other hand, if sub-band based technique [36] does not affect the image perceptibility after hiding the watermark data, we may recover the watermark data from JPEG compressed image only up to compression ratio (Q = 70 approx). So, both the above wellknown schemes do not have a good balance in Imperceptibility vs. Robustness tradeoff. Figure 6.7: 2-D Haar DWT 113

140 Therefore, in this section, our target is to develop a watermarking scheme which is: 1) ICAR in nature (because it ensures the maximum coverage of financial implications.) 2) JPEG2000 attack resilient (because it is upcoming DWT based image format). 3) Being a DWT based scheme, achieve a good balance in Imperceptibility vs. Robustness trade-off, as most of the DWT based watermarking scheme do not satisfy much of this quality. We decided to explore Haar DWT for watermarking purposes because CDMA-DWT [42][52] and Sub-band based scheme [36] used Haar DWT and in both these schemes, use of Haar DWT has shown its robustness against JPEG compression as well as image imperceptibility separately BACKGROUND OF THE PROPOSED SCHEME We used a monochrome logo as a watermark data which we first converted into a string of 0 s and 1 s. Now, we needed to hide 0 and 1, in our JPEG image, which we converted into its equivalent RGB image. As we have said above that a single DWT block of the image does not give us enough space to hide the data, we planned to take 8 x 8 DWT on a specified color channel of JPEG so that we have a large number of DWT blocks and thus have enough opportunities to hide the watermark data. We used color channel with lesser Standard Deviation (SD) (as discussed in Section 6.2). We inherited the idea of classical MBCE scheme i.e. instead of actually embedding any data, we interpret 0 or 1 by using the relative strength of two values. We claim that the average value of all coefficients of a single LL region is less susceptible to modification because LL coefficients are the major dominating coefficients and one cannot change all coefficients much. Even if some of these have been altered after one pass of coding and decoding (Taking DWT and then IDWT), the altered coefficients will again try to get their original value (if we are not changing the perceptual quality of the image). Therefore, Average of all LL coefficients may provide us a good robustness. 114

141 Even if it is slightly modified, it is very less probable that relative values of Averages of two consecutive LL blocks get modified. So we decided to hide 0 or 1 by using the relative value of average of LL coefficients of two consecutive 8x8 DWT blocks DUAL WATERMARKING Both DCT and DWT encode the image very differently. Since DWT based watermarking scheme provides coverage against DWT based attack, our watermarking scheme may not give good result against DCT transformation based attacks as in the case of JPEG compression. Since we have a very robust DCT based scheme in hand (proposed in Section 6.2), we decided to watermark the images using both schemes, one after another, to ensure the maximum coverage against attacks. So, first we watermark an image (I) using a DWT based approach to generate a watermarked copy (I ) and the on I, we again apply a DCT based scheme, presented in Section 6.2, to generate a final watermarked copy I THE DWT BASED WATERMARKING In our proposed dual watermarking, the DWT based watermarking scheme for JPEG images is defined as a 7-tuple (X, W, P, T, G, E, D) where: 1. X denotes the set of instances X i, of a particular JPEG image, (If N copies of an image are to be watermarked, then 0 i N); 2. W denotes the monochrome watermark logo; 3. P denotes the set of policies P i, 0 i N; 4. T is the watermark strength parameter. 5. G denotes the policy generator algorithm G: X i P i, where each X i will have a unique P i, i.e. a different policy to hide the watermark data. This ensures the ICAR nature; 6. E denotes the watermark embedding algorithm, E: X i x W x P i X i ; 7. D denotes the watermark detection algorithm, D: X i x P i W, Where W represents the extracted watermark. 115

142 P, THE POLICY: P is a set of policies P i, where each P i belongs to a unique X i, the instance of an image. A Pi is generated by G and is of the form (Starting block (r,s), offset, color channel). For example, for Lena test image which we used in our experiments, P i is (starting block (0,0), offset (1), Blue) G, THE POLICY GENERATOR ALGORITHM: Similar to our earlier proposed ICAR watermarking scheme for the gray image watermarking and colored image watermarking, we need to watermark each copy X i of a JPEG image X differently to ensure the ICAR nature. Policy generator algorithm is called by E, after taking 8 x 8 DWT of the image. Since average of two consecutive LL blocks have to interpret 0 or 1, G ensures that no two copies of the same original image use the same pattern. So, G achieves this by providing E, the calling routine, a starting block (which is chosen randomly) and an offset. E can start grouping of 2 consecutive blocks using this data. For example, consider an image of size 80 x 80. There will be 100, 8 x 8 DWT blocks in each color channel as shown in Figure 6.8 and 6.9. If G returns the starting block (0,0) and offset 1 in a specific color channel, then the blocks to be chosen to hide the watermark data are shown in Figure 6.8. If G returns the starting block (5,5) and offset 2, then blocks to be chosen to hide the watermark data are shown in Figure 6.9. We assume the circular queue of the DWT blocks. If our source image is of 512 x 512 size, then there are 4096, 8 x 8 DWT blocks. Using G, we can generate thousands of policies which ensure that no two watermarked copies will share same way to hide the watermark data. The overhead of this G is that the author / owner has to record all policies in his / her database to use in the decoding phase. This depicts that our proposed scheme is an ICAR scheme. 116

143 These first 2 blocks will hide first bit These 2 consecutives blocks will hide second bit Figure 6.8: An example of 2 consecutive DWT blocks (5, 5) block These 2 blocks, having offset 2, will hide first bit These 2 blocks will hide second bit Figure 6.9: An example of 2 consecutive DWT blocks 117

144 Policy generator algorithm also returns the color channel to be used to carry the watermark data. As discussed in the Section , we used the color channel with lesser Standard Deviation (SD) to hide the watermark data E, THE WATERMARK EMBEDDING ALGORITHM: To explain the embedding algorithm, we assume that G returns the DWT (0, 0) block as starting block with offset as 1. A simple watermark embedding approach is shown in Figure Embedding algorithm steps are as follows: 1. Repeat steps 2 to12 for i = 1..n; // where n is the number of copies of a single image X to be watermarked// 2. INPUT (X i ); // Xi is the instance of X. 3. Convert the X i into its equivalent spatial domain 24-bit colored image; 4. Take 8 x 8 block DWT of X i ; 5. INPUT (W); 6. Convert W into a string S = (S j S j = {0,1}, for j = 1..length of the watermark); 7. Let L = STRING_LENGTH (S); // L is the length of watermark data. If L = 1000, then first 2000 DWT block of Xi are used to hide the watermark data // 8. P i = CALL (G); // Each Pi shall be stored in an author s database for the detection purpose in future. Let the Pi, for chosen Xi, be Pi = {DWT (0, 0), Offset (1), BLUE} which is shown in Figure 6.8 // 9. Repeat steps 10 to 12 for r = 1..L; 10. Read S r ; // Based on Pi, the average AV1 and AV2 of 2 chosen DWT blocks is calculated as follows: // AV1 = (Sum of all LL coefficients of DWT (0, 0))/16; AV2 = (Sum of all LL coefficients of DWT (0, 1))/16; If (S r = 0) If (AV1 - AV2 > 0) v = (AV1 - AV2) / 16; 118

145 End; Subtract v from all coefficients of DWT (0, 0); Add v in all coefficients of DWT (0, 1); // Now relative value of AV1 and AV2 reflects the watermark bit. To further increase the robustness, we adjust the LL coefficients values such that difference of AV1 and AV2 become at least T, the watermark strength parameter // Subtract T / 2 from all LL coefficients of DWT (0, 0); Add T / 2 in all LL coefficients of DWT (0, 1); Else If (S r = 1) If (AV1 - AV2 < = 0) v = (AV2 - AV1) / 16; Subtract v from all coefficients of DWT (0, 1); Add v in all coefficients of DWT (0, 0); End; // Now relative value of AV1 and AV2 reflects the watermark bit. To further increase the robustness, we adjust the LL coefficients values such that difference of AV1and AV2 become at least T, the watermark strength parameter // Subtract T / 2 from all L coefficients of DWT (0, 1); Add T / 2 in all LL coefficients of DWT (0, 0); End; 11. Take IDWT to reconstruct X i ; 12. Convert X i back to its JPEG format; 13. End. 119

146 Generated using G LL LL 2 Consecutive 8x8 blocks will hide a single bit 0 or 1 Divide image in 8x8 block Take average of these 16 coefficients (AV1) Take average of these 16 coefficients (AV2) Adjust AV1 and AV2 (By changing LL coefficients little bit) such that their relative values reflect the watermark bits 0 or 1. If AV1 > AV2 => 1 else 0. Now consider next 2 consecutive blocks to hide next bit and so on So, 1000 watermark bits will be hidden using 2000 consecutive 8 x 8 DWT blocks. Then IDWT will be taken. After then, image will be dual watermarked using DCT based watermarking scheme presented in Section 6.2. Figure 6.10: Watermark embedding approach D, THE WATERMARK DETECTION ALGORITHM: Watermark extraction is the reverse procedure of watermark embedding. To extract the watermark from the watermarked JPEG image, first we converted it into its equivalent 24 bit colored images, took 8 x 8 DWT and then calculated the average AV of consecutive blocks based on policies stored in author s database. The detection algorithm steps are as follows: 120

147 1. INPUT (X i ); // Xi is the attacked copy of a watermarked image // 2. Convert X i into its equivalent 24 bit colored image; 3. Take 8 x 8 block DWT of X i for the specific color channel; // Based on Pi, author knows which color channel was used to hide the watermark data for a specific image // 4. Repeat step 5 to step 7 for each P i ; 5. Based on each Pi, group the DWT blocks in pairs; 6. For i = 1 to 2 * (L-1) repeat step 7; // L is the length of W // 7. For each pair (AV1, AV2) of DWT blocks; If (AV 1 > AV2) Decode 1 ; Else Decode 0 ; 8. Reconstruct W, the extracted watermark; 9. End THE DCT BASED WATERMARKING After hiding the watermark logo using DWT based watermarking presented above, we dual watermarked the images using DCT based watermarking presented in Section RESULTS We applied the proposed dual watermarking scheme on three standard JPEG test images of Lena, Mandrill and Pepper. In this section, we used a different watermark logo, which is shown in Figure

148 Figure 6.11: The watermark logo THE VALUE OF T : Our proposed DWT based scheme takes a watermark strength parameter as an input. This T itself tries to balance the Imperceptibility versus Robustness trade-off. To decide the optimal value of this parameter, we hid the watermark data in test images at various values of T and then calculated the PSNR values of the watermarked images. Some of those values (which will lead us to a final value) are shown in Table 6.5. After this, the watermark data was recovered and the quality of the watermark data was measured using Correlation Coefficient. Table 6.5: PSNR of watermarked image and CC of extracted logo for various values of T Lena Mandrill Pepper T PSNR of PSNR of PSNR of CC of CC of CC of recovered (LL color color color channel recovered logo recovered logo logo Band) channel channel Table 6.6: Revised Table 6.5 T = 500 T = 600 T = 700 PSNR Lena CC PSNR Mandrill CC PSNR Pepper CC

149 PSNR CC PSNR CC PSNR CC Lena Mandrill Pepper T=500 T=600 T=700 Figure 6.12: Graph of the values shown in Table 6.6 Table 6.5 represents the above results. It is obvious that for the higher values of T, the PSNR values of the watermarked images decrease but at the same time, the CC of the extracted logos increase. To decide the value of T, we first brought the values of PSNR data in the range of CC data, by multiplying by 2.5 and then reproduced the Table 6.6. Figure 6.12 shows the graph of the values shown in Table 6.6. It may be observed that series for T = 600 is always lying between the series of T = 500 and T = 700. It means that value T = 600 is the best value, under the Imperceptibility versus robustness trade off. Similarly for other values of T, if PSNR value is good, CC value is poor and viceversa. We conducted further tests by using T = 600 for all test images. It may be noted that our target was to embed the JPEG2000 attack resistant nature using DWT based embedding without loosing the robustness against those attacks which our DCT based scheme could sustain. Therefore, first we hid the watermark logo using DWT based scheme, and then checked its robustness against JPEG2000 attack. As presented in Table 6.6, the quality of the watermarked image did not decrease considerably. We converted the watermarked 123

150 JPEG images (without applying DCT based scheme) to JPEG2000 format. Then, we recovered the watermark logos from these watermarked images (which are converted to JPEG2000 format). Table 6.7 represents the CC coefficients of extracted logos. Figure 6.13 shows the extracted logos from JPEG2000 converted watermarked Lena, mandrill and Pepper s test images. Table 6.7: CC of extracted logos from JPEG2000 attacked images Test Image CC Lena Mandrill Pepper Figure 6.13: Extracted logos from Lena, Mandrill and Pepper s test images It may be observed from Table 6.7 and Figure 6.13 that our proposed DWT based watermarking scheme is capable of sustaining JPEG2000 format conversion attack. In order to implement the dual watermarking scheme, we further applied the DCT based scheme on the watermarked images which were generated by applying DWT based scheme. Now we had to check the effect on the image perceptibility as well as robustness against JPEG2000 format conversion attack. Table 6.8 shows the decrement in the PSNR values after the application of DCT based scheme. Though decrement is natural, it is not perceptually visible in the PSNR values. It is a compromise with the image quality to make the watermarked images very robust against more DCT based attacks, which we will present later in this section. 124

151 Table 6.8: Decrement in the PSNR values after the application of DCT based scheme PSNR if only DWT based scheme is applied PSNR if both DWT & DCT based scheme are applied Lena Mandrill Pepper After applying dual watermarking scheme, we again conducted the JPEG2000 format conversion attack on the watermarked images. Now we had a choice. We could recover the watermark logos either by applying DWT based recovery or by applying DCT based recovery. Table 6.9 shows the CC values of the extracted watermark logos recovered by both recovery methods which clearly indicate that DCT based recovery gave better results. Figure 6.14 shows the extracted logos using DWT based method and Figure 6.15 shows the extracted logos using DWT based method. Table 6.9: CC values of the extracted watermark logos recovered by both recovery methods CC if DWT based recovery is applied CC if DCT based recovery is applied Lena Mandrill Pepper Figure 6.14: Extracted logos using DWT based method 125

152 Figure 6.15: Extracted logos using DCT based method PERFORMANCE AGAINST JPEG COMPRESSION: We have seen in the previous section that our proposed DCT based scheme is very robust against JPEG compression attack. We need to check the robustness of the dual watermarking scheme presented in this section against JPEG compression. We compressed all test images after applying dual watermark at very low JPEG quality factor Q = 20, 10 and 5 and recovered the watermark logos using DCT based recovery. Our proposed dual watermarking sustained this attack very strongly even at Q = 5. Table 6.10 shows the CC of the extracted logos. Figure 6.16 shows the extracted logos. Table 6.10: CC of extracted logo from highly compressed jpeg image using DCT based recovery Q = 20 Q = 10 Q = 5 Lena Mandrill Pepper Q = 20 Q = 10 Q = 5 Lena Mandrill Pepper Figure 6.16: Extracted logos from highly compressed JPEG images 126

153 PERFORMANCE AGAINST COMMON ATTACKS AND IMAGE MANIPULATIONS: Since we know that transform domain based schemes are very robust against those attacks which can reduce the size but not the perceptual quality, we conducted some attacks on our dual watermarked images, which change the perceptual quality of an image too. The attacks are as follows: Attack 1: Adding uniform noise (10%), Attack 2: Adding Gaussian noise (10 %), Attack 3: Equalizing histogram, Attack 4: Applying uniform scaling, Attack 5: Adjusting brightness (+ 40) and contrast (+ 25), Attack 6: Horizontal flipping, and Attack 7: Adjustment of hue and saturation (+ 10 each). Table 6.11 shows the CC of the extracted watermark logos and Figure 6.17 shows the extracted watermark logos. It may be observed that the proposed dual watermarking scheme sustained all the above mentioned attacks COMPARATIVE STUDY WITH DCT BASED SCHEMES: We compared the performance of the proposed scheme against JPEG compression with other similar schemes which are DCT based and well known for their robustness against JPEG compression. We re-implemented these schemes for JPEG images. Schemes chosen were: Scheme-A: Correlation based scheme (Section ) Scheme-B: The classical Middle Band Coefficient Exchange scheme (Section ) Scheme-C: Collusion attack resistant watermarking scheme (Section 4.4): Scheme proposed in Section 4.4 is also based on MBCE scheme and ICAR in nature. This scheme swaps 4 pairs of coefficients in FM region in correlation with low band coefficients. We are naming this scheme as Scheme-C. Scheme-D: The scheme presented in Section 6.2. Scheme E: The proposed dual scheme in this section. 127

154 Table 6.11: CC of the extracted watermark logos Attacks Lena Mandrill Pepper Adding uniform noise (10%) Adding Gaussian noise (10 %) Equalizing histogram Applying uniform scaling Adjusting brightness (+ 40) and contrast (+ 25) Horizontal flipping Adjustment of hue and saturation (+ 10 each) Figure 6.17: Extracted watermark logos after applying common attacks 128