DIGITAL IMAGE WATERMARKING BY USING INTERMEDIATE SIGNIFICANT BITS IN GRAY SCALE IMAGES ALI SHARIFARA A dissertation submitted in partial fulfillment of the requirements for the award of the degree of Master of Science (Computer Science) Faculty of Computer Science and Information Systems Universiti Teknologi Malaysia AUGUST 2012 To my lovely father, AliAkbar Sharifara to support and encourage me in the whole of my life specially for being my source of inspiration… Thanks for being there & no words could describe your support & encouragement I love you… ACKNOWLEDGEMENT My appreciation first of all goes to my supervisor, Prof. Ghazali Bin Sulong, I would like to take this opportunity to thank him who gives me a lot of encouragement and guidance during the dissertation. His guidance has been a valuable asset for me during this dissertation. I also would like to thank all my friends who are given appreciated opinion and idea of this dissertation. They are also willing to share their knowledge with me. All the discussion and recommended opinion of my course mare and friends were appreciated by me. Finally, I would like to send my appreciation to my lovely family for their support and encouragement throughout my whole life. ABSTRACT The rapid growths of the computer technologies have been increased over the last half century in terms of amount and complexity of data. Also, access to the data has become much easier due to rapid growth of the networks such as Internet. Furthermore, most of the people use image to represent information and it is transferred throughout the internet. Digital watermarking techniques are used to protect the copyrights of multimedia data by embedding secret information inside them, for example, embedding in images, audios, or videos. Digital Image watermarking also has been used to detect original images against forged images by embedding an evidence of the owner of image. Digital Image watermarking can be categorized into two domains namely; spatial, and frequency domains. The former has been chosen for the current research and some remaining problems have been studied such as imperceptibility, security, and robustness of watermarked image. A repeated method is proposed to solve the problem of robustness, and a Zig-zag algorithm is also recommended to change the order of embedding in watermarked image. Furthermore, the proposed method aims to embed watermark in different bit planes of host image which can improve both the visual quality and robustness. To evaluate the proposed technique, some attacks have been applied for the approach and the experimental results have shown that the proposed technique successfully withstood against most of the attacks, and at the same time preserved the watermarked image quality. ABSTRAK Kadar tumbesaran teknologi komputer telah melonjak naik sejak dari separuh abad yang lalu dari segi aspek perubahan jumlah dan kerumitan data. Juga, capaian data menjadi lebih mudah dengan kemajuan pesat jaringan seperti internet. Selanjutnya, kebanyakkan orang lebih cenderung menggunakan imej sebagai medium utama untuk menyampaikan maklumat melalui internet. Teknik Tera Air digunakan untuk menlindungi hak milik data multimedia dengan membenamkan maklumat rahsia ke dalamnya, sebagai contoh, pembenaman dalam gambar, audio dan video. Tera Air digital juga digunakan dalam mengesan keaslian imej daripada yang palsu dengan membenamkan bukti pemilik asal imej tersebut. Tera Air digital boleh di kategorikan dalam 2 domain iaitu spatial dan frekuansi. Domain spatial dipilih dalam kajian ini dan masalah yang masih wujud ini dikaji seperti ketampakan, keselamatan dan kekukuhan imej tera air . Kaedah pengulangan dicadangkan untuk menyelesaikan masalah ketampakan dan algoritma zig-zag dicadangkan untuk mengubah susunan pembenaman imej tera air. Seterusnya kaedah yang dicadangkan ini mensasarkan untuk membenamkan tera air dalam satuan ‘bit’ yang berlainan pada imej luas supaya dapat meningkatkan kualiti visual dan kekukuhan pada tera air tersebut. Untuk menilai kaedah yang dicadangkan ini, beberapa serangan telah dilakukan dan keputusan kajian telah menunjukan bahawa kaedah tersebut berjaya mempertahankan hampir keseluruhan serangan dan pada masa mengekalkan kualiti imej tera air. TABLE OF CONTENTS CHAPTER 1 2 TITLE PAGE DECLARATION ii DEDICATION v ACKNOWLEDGEMENT vi ABSTRACT (ENGLISH) vii ABSTRACT (MALAY) viii TABLE OF CONTENTS ix LIST OF TABLES xiii LIST OF FIGURES xiv LIST OF ABBREVIATIONS xvi INTORDUCTION 1 1.1 Introduction 1 1.2 Background of the Problem 3 1.3 Statement of the Problem 4 1.4 Aim of dissertation 5 1.5 Objectives of the Study 5 1.6 Scope of the Study 6 1.7 Significance of the study 7 1.8 Dissertation organization 7 LITERATURE REVIEW 8 2.1 8 Introduction 2.2 Digital Watermarking history 11 2.3 Digital watermarking overview 13 2.4 Importance of digital watermarking 13 2.5 Applications of watermarking 14 2.5.1 Ownership assertion 14 2.5.2 Fingerprinting 14 2.6 2.7 Embedding Watermark Techniques 15 2.6.1 Visible watermarking 15 2.6.2 16 Invisible watermarking Extraction of watermark 17 2.7.1 Non-blind-watermarking 17 2.7.2 18 Blind-watermarking 2.8 Peak Signal-to-Noise Ratio (PSNR) Test 18 2.9 Measuring performance of watermarking 18 2. 9.1 Robustness 19 2. 9.2 Imperceptibility 19 2.9.3 Security 19 2.10 Watermarking Attacks 20 2.10.1 Classification of attacks 20 2.10.1.1 Removal attacks 21 2.10.1.2 Geometric attacks 21 2.10.1.3 Protocol attacks 22 2.10.1.4 Cryptographic attacks 23 2.11 2.12 Spatial Domain Methods 23 2.11.1 Least Significant Bit 24 2.11.2 Most Significant Bit 26 Frequency Domain Methods 26 2.13 3 28 RESEARCH METHODOLOGY 31 3.1 Introduction 31 3.2 Research environment 32 3.3 Introduction of the new method 33 3.3.1 33 The best quality of image 3.3.2 The best robustness 34 3.3.3 Variable Repetition 36 3.4 Integrate watermarking 36 3.5 Security in watermarking 37 3.5.1 Random Pixel Manipulation methods 38 3.5.2 Zig-zag embedding matrix 39 3.6 4 Related work Enhancement of capacity in watermarked Image by PVD 40 3.7 Evaluation of Quality and Robustness 41 3.8 Embedding process 42 3.9 Embedding Phase 43 3.10 Extracting Phase 44 3.11 Applying attacks 45 3.12 Dissertation Framework 46 3.13 Summary 47 IMPLEMENTATION 48 4.1 Introduction 48 4.2 The chosen attacks 49 4.3 The implementation 50 4.4 Evaluation of watermarking methods 51 4.4.1 Mean Squared Error (MSE) 51 4.4.2 51 Pick Signal to Noise Ratio (PSNR) 4.5 Watermarking with the best quality 52 4.6 Watermarking with improve security 52 4.7 Implementation and result 54 4.8 Improving the capacity 67 4.9 Summary 67 5 CONCLUSION 69 5.1 Introduction 69 5.2 Contribution and Achievement 69 5.3 Future Work 71 5.3 Summary 72 REFERENCES 73 LIST OF TABLES TABLE NO TITLE PAGE 2.1 Intensity Matrices 29 4.1 The impact of attacks on Images per pixel 50 4.2 The watermarked images by using proposed method for each bit plane and their PSNR value (Lena) 4.3 The watermarked images by using proposed method for each bit plane and their PSNR value (Lake) 4.4 56 The watermarked images by using LSB until MSB for each bit plane and their PSNR value (Lena) 4.5 55 57 The watermarked images by using LSB until MSB for each bit plane and their PSNR value (Lake) 58 4.6 NCC value of proposed method 59 4.7 NCC Value for LSB until MSB under different attacks 60 4.8 NCC value for LSB until MSB under different attacks 60 4.9 Extracted watermark after attack in proposed method (Lena) 61 4.10 Extracted watermark after applying attacks in proposed method 62 4.11 Extracted watermark after applying the attacks in LSB method 63 LIST OF FIGURES FIGURE NO TITLE PAGE 1.1 The approach of study 6 2.1 Sample of analog watermarking 9 2.2 Embedding watermark process 10 2.3 Detecting watermark process 10 2.4 Extracting watermark process 10 2.5 Watermarking Schemes 11 2.6 Visible image watermarking 15 2.7 Invisible image watermarking 17 2.8 Watermarking Attacks 20 2.9 8 bit planes of a gray scale image 22 2.10 LSB and MSB bit-Planes 24 2.11 pixel value of the cover image and watermark 25 2.12 Digital watermarking based on DCT 27 2.13 pixel intensity matrixes 29 3.1 Grayscale watermark and host images 32 3.2 Eight layers of host and watermark images 33 3.3 Divided cover image into blocks 34 3.4 Repeating bits in each block 35 3.5 different sizes of blocks 36 3.6 Random Pixel Manipulation process 38 3.7 Zig-zag embedding matrix 39 3.8 Capacity improvement techniques 41 3.9 Embedding watermark phase’s framework 43 3.10 Extracting watermark phase’s framework 44 3.11 Applying attack phase’s framework 45 3.12 Dissertation Framework 46 4.1 Original Host image and different attacks 49 4.2 Improving security by using Zig-zag algorithm 53 4.3 Applying zig-zag algorithms to embed watermark 54 4.4 Applied salt and pepper attack for both proposed and LSB method 64 4.5 Applied Gaussian attack for both proposed and LSB method 65 4.6 Applied Speckle attack for both proposed and LSB method 65 4.7 Applied Poisson attack for both proposed and LSB method 66 4.8 Applied blurring attack for both proposed and LSB method 66 LIST OF ABBREVIATIONS LSB - Least significant Bit MSB - Most significant Bit ISB - Intermediate Significant Bit CWT - Continues Wavelet Transform DFT - Discrete Fourier Transform DCT - Discrete Cosine Transform HVS - Human Visual System PSNR - Peak Signal to Noise Ratio MSE - Mean Squared Error NCC - Normalized Cross Correlation CHAPETR 1 INTRODUCTION 1.1 Introduction Over the last half century the pace of change in the digital technologies has been widely increased. Digital images as one of the digital technologies also have been replaced with the analog images. Moreover, Internet also became as one of the most important tools to transfer digital images from one part to other parts of the world. For this reason, the security of digital documents became a challenging concern and digital image watermarking as a solution is use to decrease the number of digital forged documents (Langelaar, 2000; Kumar N.M. , 2011 ). Information hiding methods have been using since the presence of paper and after that used to many application areas such as digital image, audio, video which put an evidence of the owner of documents such as a logo, serial numbers and etc (Moulin, 2003). This might be helpful to decrease the number of unauthorized copy of them. In addition, information hiding must be imperceptible and the signals of embedded data must be low enough when projected into the human eyes. Digital watermarking as a solution has been presented to be very practical for identifying the owner of documents (Yongjian Hu, 2004). It has been presented for some purposes like: copyright protection, data authentication, fingerprint, medical applications, and broadcast monitoring. Furthermore, documents can be divided into two main groups, analog and digital. Image watermarking is the process of embedding an image into a host image. For Instance, watermarks are embedded in bank cheque for preventing forgery. Consequently, unauthorized modification of data is the concern of researchers about copyright of documents and numerous image watermarking methods have been proposed with different complexity levels so far, and all of them try to set up the balance between quality and robustness of images. Moreover, for extracting watermark there are two general approaches. The first one is blind watermarking and the second one is non-blind watermarking (Hyeong-In Choi, 2010). In blind watermarking there is no need to have the original document, while in the non-blind extraction we need to have the original document to detect the watermark (Song-Hwa Kwon, 2011). In addition, watermark can be visible or invisible in both digital and analog documents. As an example, a visible watermark can be a signature in an image that is used to point out the ownership of image; meanwhile an invisible watermark is not apparent easily. The embedded watermark can extract by using an extraction algorithms for identifying the copyright owner. Existing digital watermarking techniques can be categorized into one of the two domains including spatial and frequency, according to the embedding domain of the host image. There are lots of researches that have been proposed base on frequency and spatial domains. For example, digital image watermarking by applying LSB (Least Significant Bit) and MSB (Most Significant Bit) are two common techniques that try to achieve robustness and quality at the same time in spatial domain(Ibrahim Nasir, 2007 ). 1.2 Background of the Problem There are lots of images on the internet without having watermark, and everybody can download and modify them illegally. Furthermore, the owner of the image can be missed without watermarking (Nour El-Houda Golea, 2010). Consequently, watermarking is used to protect this behavior by adding an image as a watermark into the host image. Regardless of images we also can use it for other important documents such as video, audio, and text. Generally, there are some concerns about this area of research that still need to be solve, such as imperceptibility, robustness, security (Jiang Nan, 2006). The first concern about the digital image watermarking is imperceptibility. It means a watermark can be inserted in the cover image without making any kind of degradation. It means, after embedding the watermark, the watermarked image and host image must be identical. In other words, the watermarked image must be the same as original image and nobody can realize the differences between them by the naked eye. In fact, the embedded watermark is really imperceptible if human eyes cannot distinguish between watermarked image and the original image. The second one is robustness; it means the watermarking scheme must be able to protect a watermark against geometric distortion attacks like rotation, JPEG compression, and also signal processing attack like sharpening, blurring, adding noise and so on. Robustness related to ability of recovering the watermark after performing various processing attacks on watermarked image. The robustness must be sufficient if any kind of attacks occurs. The watermarking scheme should be able to protect watermark against possible signal processing operations, in addition for evaluating the quality of the watermark after applying attacks. Security is another important issue in watermarking, which means protecting the watermark from unauthorized users and a digital watermark must be completely invisible. Due to use algorithms for embedding the watermark into the cover image and extract it as well, presence of a secret key is needed for keeping those algorithms safe; otherwise unauthorized users can detect, remove, or modify the watermark form the host image easily. Moreover, the human eyes have different sensitivity for different image regions and a watermark should embed in the highest priority part of the host image. In the processing of watermarking, it is important to detect the best parts of the image and embed the watermark within these areas of images. Robustness and imperceptibility are two main parameters in the watermarking that tries to adjust beside the capacity of information which can be embedded into the host image (Ching-Tang Hsieh 2001). 1.3 Statement of the Problem Illegal copying, modifying and copyright protection have become very significant issues with the quick use of internet and no one can deny the importance of Internet in our lives. Moreover, everyone can access to the vast of information on the Internet that includes image, video, audio and text formats and also use them for personal or commercial goals. Hence, malicious users can abuse of these digital documents. So, digital watermarking can protect them from forgery. Therefore, the digital watermarking methods have been identified as a possible solution to the copyright protection, and have become an area of increased research activity over the last decade (Gang Liu 2010). Digital watermarking should be satisfied the below questions: 1- How we can embed an image as a watermark into the host image without causing visible degradation? 2- How we can achieve quality of watermarked image without losing robustness? 3- How much information we can put into the host image? 4- How we can extract watermark from the watermarked image? 1.4 Aim of dissertation The main aim of this research is to propose a technique based on ISB (Intermediate Significant Bit) to achieve high robustness and imperceptibility of gray scale watermarked image. Applying ISB in watermarking is robustness enough to protect image from attacks and also preserve quality of watermarked image in spatial domain. 1.5 Objectives of the Study To answer the problem statement, following objectives of this project are: 1- To study and apply watermarking in gray scale image by using ISB (Intermediate Significant Bits). 2- To improve the robustness of gray scale image watermarking. 3- To apply image attacks and evaluate the quality and robustness of proposed watermarking method by using PSNR (Peak Signal to Noise Ratio) and NCC (Normalized Cross Correlation). 4- To compare proposed method to LSB and MSB methods after and before applying attacks. 1.6 Scope of the Study In this paper we deal with gray scale images as host image that we want to add and extract watermark, and the approach is using non-blind watermarking in spatial domain by applying ISB. The applied attacks are Gaussian filter, salt and pepper, speckle, and Blurring attacks. Host image and watermark images are a grey scale image with 256 × 256 pixels and a gray scale logo with 50 × 50 pixels, respectively and both of the images are in TIF format. Figure 1.1 The approach of study 1.7 Significance of the study Over the past few years digital watermarking has became more popular due to its significance in content authentication and legal ownership for digital multimedia data. Furthermore, digital watermarking and data hiding has become an important tool for protecting digital images from theft, illegal copying and unlawful reproduction. 1.8 Dissertation organization This research consists of five chapters. The first chapter presents introduction to the project which includes the problem background, problem statements, aim of project, the main objectives and scope of the project. In the chapter two covers information about literature review on watermarking, which focused on current algorithms that have been using for protecting image against attacks. The project methodology is discussed in Chapter three where comparative study and pre-lab testing have been used as the research strategy. In Chapter 4, the implementation of the methodology where the findings of comparative study and pre-lab testing take place and eventually, in Chapter five the result and findings of the lab testing is explained and the overall project will concluded as well. CHAPTER 2 LITERATURE REVIEW 2.1 Introduction The growth of the Internet over the last several years has been highlighted the requirement of techniques for protecting ownership of digital documents in the digital world and also identical copies of digital documents such as images, video, and audio can distribute on the internet easily (Arya, 2010). Hence, without having the powerful techniques for protecting digital documents, it is impossible to say who the owner is and who the plagiarist is. In consequence of all above reasons, digital watermarking is one of the proper solutions for protecting copyright of digital documents against forgers. By using watermarking in digital documents, we can protect of misusing behavior in important documents like Will, Cheque and etc. The figure 2.1 depicts the watermark into the money as an analog document which has been used to protect money against forgers. Figure 2.1 Sample of analog watermarking Embedding a hidden stream of bits in an image is called digital image watermarking. The main aim of digital image watermarking is embedding image as a watermark into an image as a host image and can be extracted later which provides evidence of its authenticity. Hence, using watermark allows someone to identify the original owner. Digital image watermarking categorized into two main groups. They can be visible or invisible (Arya, 2010). However, more focus is on invisible watermarking. Due to, they don’t have any effects on the quality of images and watermarks in this group are imperceptible as well. In visible watermarking the watermark can be seen with the naked eyes, whereas in invisible methods, the watermark cannot be seen easily and we need to use an extraction algorithm to take out the watermark from the watermarked image. In fact, using each group of watermarking depends on their applications. For example, television channels are using a visible watermark as a logo that specifies the owner of channels. There are three main processes involved in digital image watermarking which includes embedding, detection, and extraction of a watermark (S. Shefali, 2007). We also can secure watermarked image by adding a secure or public key in watermarking process. It causes that only authorized receptions can extract watermark form the digital content .In the figures 2.2-2.4, the main processes for producing a watermarked image have been demonstrated. Figure 2.2 Embedding watermark process Figure 2.3 Detecting watermark process Figure 2.4 Extracting watermark process Digital image watermarking algorithms can divide into spatial and frequency domains (K.Ganesan, 2010). There are many methods for achieving the goal have been proposed in both spatial and frequency domains. In the rest, we will describe other techniques which have been using to overcome the watermarking problems. However, the main aim of this research is to achieving robustness and quality of watermarked image in spatial domain. 2.2 Digital Watermarking history Data hiding is a global term that encloses a wide range of problems beyond that of embedding information in documents. The term of hiding here means trying to keep the watermark imperceptible and keeping the information secure as well. Hence, the data hiding has been divided into watermarking and steganography. Figure 2.5 Watermarking Schemes As we can see in the figure 2.5, data hiding can divide into watermarking and steganography (Cox, 2008). Furthermore, the word of Steganography invented by Trithemius and he was the first author of publications about cryptography. In addition, this word derived from Greek words Steganos and Graphia which means covered and writing respectively. In fact, Steganography is the art of hiding communication. In the late 20th century digital systems had become popular for the watermarking of several documents. The main focus was on audio, video and photographs. After that, when the attack occurred on September 11, 2001 Interest in steganography increased, when it became clear that means for camouflaging the communication itself are likely to be used for criminal actions. The first steganalytic techniques focused on the most usual type of hiding which called LSB (Least Significant Bit) embedding in bitmap images. After that, other types of images like GIF, JPEG, etc beside audio formats had become popular (Cox, 2008). Furthermore, the concern over copyright protection of documents was the main reason to increase in watermarking interest. The internet had become more popular by with the first web browser in 1993 and people wanted to download digital media like videos, music and pictures form the internet. Moreover, internet is a distribution system with high performance and cheap price in comparison by other communication methods (Cox, 2008). On the whole, due to need for copyright protection, digital watermarking technology getting more and more popular along with associates to digital data over the world that contains different multimedia documents such as image, video, and audio. 2.3 Digital watermarking overview The art of embedding copyright information into original contents is called Digital watermarking and the information embedded is also called “watermarks”. Digital watermarks usually do not put an obvious object in the content and do not influence its appearance. Moreover, these are imperceptible and only can detect by suitable authorities. In fact, digital watermarking is a signal which carrying data that is embedded into another signal, for example into an image or video signal. In other words, the basic idea behind the digital watermarking is addition of a signal into another signal which is imperceptible and also cannot be identified without having the knowledge of the parameters of the watermarking algorithm. 2.4 Importance of digital watermarking Due to digital broadcast of contents on the networks (especially Internet), legal issues of copyright protection have become more important in the last several years. Moreover, all of the information on the internet is provided as digital contents. Hence, a desirable technique for protecting copyright is digital watermarking. Adding a watermark which can authenticate the legal copyright owner and that cannot be removed and manipulated easily by forgers without having the secure key or other security restrictions. For instance, if cheque could be duplicated easily, trust in its authenticity would be greatly decreased. For this reason, watermarks are employed in currencies to reduce the risk of forgery as well. However, watermark is not only the technique that has been using for preventing forged and illegal use. As a result, digital watermarking protects the ownership of host signal by embedding data such as the copyright data , company logo or any other digital data that specifies the original owner of content. In addition, Watermarked object will be produced when we embed watermark signal into the cover object. 2.5 Applications of watermarking Digital Watermarks are potentially useful in many applications such as ownership assertion, fingerprinting, medical industry, etc. in the rest; we described some of them in detail. 2.5.1 Ownership assertion Watermarks can play a role as evidence for the owners of the digital contents. In other words, watermarks have been using in digital objects for protecting of the ownership of the contents. For asserting ownership of a digital content like image, a watermark signal can be generated by using an algorithm and also can be embedded it into the host image. By doing this only the receiver and recipient of the digital content can realize and extract the watermark by using the suitable algorithm. 2.5.2 Fingerprinting Fingerprinting is one type of watermarks which recognizes the receiver of a digital object as well as its owner. In other words, fingerprinting provides indication of the ownership, and indication of the identity of a licensed user, by embedding information in the digital content. This information can be imperceptible or perceptible. Furthermore, Fingerprinting attempts to provide copyright owners with the desired degree of protection. 2.6 Embedding Watermark Techniques There are two fundamental ideas for embedding watermark into the digital content. The watermark can be embedded in the digital content as visible or invisible. 2.6.1 Visible watermarking A visible watermark is a visible content which is called watermark that covers on the original content. In general, visible watermarking should be easily visible and must be difficult to remove (Yongjian Hu, 2004). A visible watermark is like a number or an image which is used to specify the ownership of the content and everybody can see the watermark in the digital content easily even with naked eyes. This technique allows the original content to be showed, but it still provides evidence by marking the content as its owner’s possession. As an example, in the figure 2.6 we demonstrated how visible image watermarking works. In the below figure, (a) is a cover image and (b) is a watermark. The watermarked image is also has been shown in figures 2.6 (c). Figure 2.6 Visible image watermarking Although we can use some techniques to make it difficult for removing watermark form the original content, but we have to accept that remove the watermark is not impossible (Yongjian Hu, 2004). As a result, visible watermarking is a good solution for digital contents which are distributed on the public networks like Internet. Moreover, users can realize the main owner of digital content with naked eye and it could decrease the security of watermarked image. 2.6.2 Invisible watermarking Most of the data hiding techniques try to manipulate the cover digital objects in order to can embed the secret data into them. Although, manipulate may be small and imperceptible. Invisible watermarking is also one of data hiding techniques which applied to digital contents like: image, video and audio. In addition, invisible image watermarking embeds watermark into the cover image in order to alter the pixel values without changing the quality of cover image and nobody can notice it easily. Moreover, watermark can be extracted only with proper decoding algorithm. The following figure shows invisible image watermark process. Figure 2.7 invisible image watermarking There are two fundamental approaches for extracting watermark. The watermark can be extracted from the digital content as non-blind or blind watermarking. Depending on the method used for watermark extraction digital watermarking can be divided into blind watermarking and non-blind watermarking (Kumar N.M. , 2011 ). 2.7.1 Non-blind-watermarking Distinguishing between non-blind and blind watermarking schemes depending on whether or not the original digital document is required at extraction process. The Non-blind watermarking is used for extracting the watermark from the cover content. In other words, cover content is required during the detection process for extracting watermark (Kumar N.M. , 2011 ). Essentially, non-blind watermarking method extracts the watermark by comparing the watermarked picture with the original picture. 2.7.2 Blind-watermarking The Blind-watermarking is also used for extracting the watermark from the cover content. Unlike Non-blind-watermarking, cover digital content is not required during the detection process (Makhloghi, 2011). In fact in Blind watermarking method does not need original image to extract the watermark from the watermarked image. 2.8 Peak Signal-to-Noise Ratio (PSNR) Test The Peak Signal-to-Noise Ratio (PSNR) value has been used to evaluate the quality of the watermarked images and common values for PSNR can be between 30 and 40 dB. If the value of PSNR for a watermarked image is more than 30, we can say that it is difficult to notice of the differences between cover image and watermarked image by Human Visual System (Gil-Je Lee, 2008). 2.9 Measuring performance of watermarking In order to obtain the purposes of digital image watermarking especially for copyright protection, watermark techniques should satisfy the following requirements and also performance of a watermark technique will be measured by them. 2. 9.1 Robustness When we can say that a watermark method is robust which the watermarked image suffers from different attacks and the embedded image (Watermark) still can be extracted from the watermarked image (Kumar N.M. , 2011 ). In other words, the attacks which we descried them in section 2.10, must do not have any effects in the extraction process. For example, when a watermarked image is rotated, the algorithm still must identify the watermark that we have embedded inside the cover image and extract it as well. 2. 9.2 Imperceptibility When we can say that a watermark method is imperceptible which is so hard for human visual system (HVS) to recognize the differences between the original image and the watermarked image (Kumar N.M. , 2011 ). This means that both original and watermarked image must be identical and nobody can notice and find the watermark easily. 2.9.3 Security Security in the watermarking process plays an essential role. In order to encoding the image by embedding secure or public key (Kumar N.M. , 2011 ) . In fact, only the owner of the digital media must be able to extract or modify the watermark from the digital media. Otherwise our method could be insecure and anybody can manipulate the digital media and put their watermark as well. 2.10 Watermarking Attacks Watermark attacks wished to ruin or remove the watermark signal form the digital watermarked content. Hence, it is desirable to understand how those attacks will be occurred to design a more robust watermarking method against them. 2.10.1 Classification of attacks Watermarking attacks can be divided into the four main categories. They are removal, geometric, protocol and cryptographic attacks (Voloshynovskiy, 2001). In the below figure illustrates them along with some sample of common their attacks. Figure 2.8 Watermarking Attacks (Voloshynovskiy, 2001) 2.10.1.1 Removal attacks This type of attacks aimed to remove the watermark from the watermarked image without using any decryption algorithm. In other words, this type does not aim to find out the algorithm which is used for embedding watermark into the host image. In fact, removal attacks affect in the magnitude of watermark signals and can removal the watermark or decrease the energy of host signals. Hence, extract the watermark from the watermarked content becomes difficult. This type of attack includes noising, histogram, filtering, blur, and sharpen attacks. 2.10.1.2 Geometric attacks Geometric attacks aimed to the embedded data in the watermarked image with distort the watermark detector synchronization. In other words, they trend to distort signals in the watermarked objects (Toshihiko Yamasaki, 2009). Some of the common attacks in this category are image rotation, translation, skewing, cropping, and scaling. Furthermore, there are three directions to solve such these attacks. The first method is in geometric constant domain. The second one is embed an extra template to evaluate geometric transformation. The last one is digital watermarking based on the feature points (Jinguang Sun, 2010). Moreover, two new attacks which try to destroy low ISB bits (2,3,4 bit plans) in watermarked image have been introduced including Set Removal attack, Reset Removal attack (Mir Shahriar Emami, 2011). Figure 2.9 8 bit planes of a gray scale image (Mir Shahriar Emami, 2011) Several techniques have been proposed in literature for solving this problem so far. For example, algorithms by employing SIFT. In fact, SIFT is a theory for extracting features. It can detect those points that are changeless to location, rotation and scale (Jinguang Sun, 2010). 2.10.1.3 Protocol attacks In this type of attack, attackers try to add their own watermark into the watermarked object instead of the main watermark (Martin Kutter, 2000). It causes that realizing the main owner of the object become difficult. For an instance, copy attack is one of the protocol attacks which do not aim to destroy the watermark, but it tries to find out the watermark and copy the watermark into some other objects. Furthermore, the discovered watermark will be accustomed to the other objects and will fulfill their imperceptibles. Some of the common attacks in this category are invertible and copy attack. 2.10.1.4 Cryptographic attacks The main of this attack is to find out the watermark and remove it from the watermarked object or to put misleading watermarks (Martin Kutter, 2000). In fact, this type of attack trends to crack the security of the watermarked object. Some of the common attacks in this category are brute-force and Oracle attack. Furthermore, due to their complexity, application of this type of attack is limited. 2.11 Spatial Domain Methods As we already mentioned, there are two types of watermarking techniques based on the domain, namely spatial and frequency domains. However, both of them have their own benefits. One of the benefits of the spatial domain method is that they can be applied to any images easily. Spatial watermarking is also popular due to provide tradeoff between robustness and quality in the watermarked objects. There are two important keys for the spatial watermarking which are the region of the host image that we can embed the watermark and as well as the information inside the watermark. Furthermore, there are also three main factors which can define the parameters of algorithms in spatial domain watermarking (Megalingam, 2010). Firstly, watermark is closely connected to the quality and size of the watermark image. Secondly, using a secret key in the process of watermarking can improve the security of watermark. In term of using secret key only authorized recipient can extract watermark. The last one is related to the masking property of the image. It is also connected to the quality of the image that indicates the transparency of the watermark into the host image. 2.11.1 Least Significant Bit Least Significant Bit (LSB) is one of the first techniques for watermarking and it uses the lowest bit plane of the bitmap images. It can modify bits of both cover and watermark image for embedding the given watermark. The basic idea behind this technique is substitution of the lowest bit plane of the host image with the watermark. In other words, the spatial domain technique embeds the direct watermark into the cover image by altering the pixel values (Gil-Je Lee, 2008). Moreover, LSB is extremely easy in term of implementation and also eye cannot recognize very minor digressions (Ker, 2004). However, this method is imperceptible and there is no significant difference between watermarked image and original image, but it is not robustness enough to protect the authority of watermarked images. This means that in the spatial domain embedding capacity can be large, but the watermark could be easily found by unauthorized forgers. Figure 2.10 LSB and MSB bit-Planes In addition, there are 8 bits for storing each pixel in a grayscale image system. As can be seen from the figure 2.10, we have one pixel value of the cover image and also we have the secret data in the figure, which their value are 192(011000000)2 and 1(1)2 respectively. It applies to 1bit LSB that altered the value of pixel and changes it to 193(011000001)2. Undoubtedly, LSB is an imperceptible technique; due to changing the least significant bit value of each pixel does not have any significant effects in a digital media (Puneet Kr Sharma, 2012). Figure 2.11 one pixel value of the cover image (a), one pixel value of secret data or watermark (b), change one pixel value of cover image by a bit of secret data (c). Obviously, by using LSB we will be able to store 1 bit in each pixel. For example, in an image with size of 128*128, LSB can only store 16384 bits into the cover image. As a result, this type of watermarking technique is not robust and forgers easily can access to the secret data and remove it from the digital content. It also could be more fragile, when we put the secret data in a sequence order. 2.11.2 Most Significant Bit The main primary concept of Most Significant Bit (MSB) is the same as LSB that we have described in the previous section. The only difference between them is that, LSB uses the least significant bit, meanwhile in the MSB is completely different and it uses most significant bits instead. However, the base idea of LSB and MSB is the same, but they have big difference in term of imperceptibility and robustness. As we discussed in the previous section, those algorithms which use LSB are not robust. We also mentioned that LSB has some advantages like imperceptibility, but algorithms which use MSB are not imperceptible. Because changing the most significant bit in the pixel value of an image has considerable effect on the image. For instance, if we have 193(011000001)2 as a pixel value of a cover image and 3 bits of our watermark will be 4 (100)2, the result for MSB will be 129 (100000001) 2. As can be seen, the result is very far from the cover pixel. 2.12 Frequency Domain Methods Watermarking algorithms in frequency domain may use different transform functions such as Discrete Sine Transform, Discrete Cosine Transform, Discrete Wavelet Transform, and Discrete Hartley Transform which abbreviation for them are DST, DCT, DWT, and DHT respectively (Megalingam, 2010). In addition, in frequency domain image watermarking, both host and watermark image must transmute into the frequency domain. Therefore, it is extremely hard to find the watermark in the frequency domain. (1) Where, Vi’: is the result of added DCT of the both host and watermark image. Wi: is DCT value of the watermark image. Xi: is DCT value of the host image. Also, the visibility of the watermark will be adjusted by Į and ȕ as two constants. For instance, in the above formula if we adjust ȕ as a low value, watermark will be invisible and if we change it to a high value, watermark will be more visible as well. Also, if we want to extract watermark, we need to know the values of Į and ȕ, otherwise the process of extracting watermark will not be completed. It causes that the security of the watermarked images will be increased. Moreover, it can be more secure when the value of each block for Į and ȕ are different and only the owner of image knows about their values. Figure 2.12 Digital watermarking based on DCT It the figure 2.12, we have demonstrated watermarking in frequency domain by using Discreet Wavelet Transform. In the figure there are a logo and a Lena image which are our watermark and host image, respectively. Watermarked image also is provided in the above figure. However, computation of DCT is more complex than as shown in the figure and it also involves lots of complex matrix manipulations. 2.13 Related work A new LSB algorithm based on the digital watermarking scheme has been presented (Gil-Je Lee, 2008). Moreover, the proposed algorithm works in the grayscale image and this technique also embed the same data into the cover image as traditional LSB technique. In fact, the proposed algorithm enhances the traditional LSB algorithm to make it more secure and robust. The authors believe that their algorithm is robust against some attacks like cropping and equalization attacks and attackers must get the seed for detecting the watermark. A simple algorithm for both embedding and extracting watermark in the spatial domain has been proposed (Megalingam, 2010). The formulas are also provided as: Ǔ=Y+ĮI (2) Where, Ǔ: is the watermarked image. Y (i, j): is the intensity of the host image in the position of i and j. Į I: is the watermark. Moreover, by using the below formula, the authorized owner can extract watermark by comparing the watermarked image and the original image. I = (Ǔ - Y) / Į (3) As can be seen in the figure 2.8, there are four diversified pixel intensity matrices which called lowest intensity matrix(LL), Intermediate Intensity Matrix(LH) ,Higher Intensity Matrix(HL) and Highest Intensity Matrix(HH). The pixel intensity matrix of the host image will be compared with four constant values which constants are 63,127,191 and 255 for 8-bit encoded pixel data. Figure 2.13 pixel intensity matrixes In fact, (Megalingam, 2010) believe that their algorithm is working different by other conventional algorithms that have been proposed so far in spatial domain watermarking. Also, their mentioned which their method is robustness enough and it guarantees an average PSNR for all images. In their algorithm, all values of each pixel will be compared with all four diversified intensity matrices and will be classified by regarding to the table 2.1. Table 2.1 Intensity Matrices (Mir Shahriar Emami, 2011) proposed two new Geometric attacks including Set Removal (SR) and Reset Removal (RR) attack which aim to replace and remove the embedded watermarks in a watermarked media and also evaluates the accuracy of the watermarking robustness metrics. It also uses all possible capacity of the watermark embedding bit-plane, so the watermark will be embedded several times in this bit-plane. In the extracting process, in order to determine the robustness, the highest obtained value as the degree of the similarity between the original watermark image and the attacked watermark image is considered. (Chan, 2004) proposed a simple watermarking algorithm based on the Least Significant Bit in grayscale image. They believe which their algorithm is more robust than the traditional LSB technique. In fact, they proposed a new LSB algorithm which uses the third and the fourth least significant bits of cover image content’s value for embedding watermark inside the cover digital content. They also added that their algorithm does not have any obvious distortion in the watermarked image. CHAPTER 3 RESEARCH METHODOLOGY 3.1 Introduction In this chapter, the methodologies and techniques which have been applied to achieve the objectives of the study have been presented. In addition, the watermark and host images that have been used as datasets for the study are introduced as well. The approach tries to achieve the best quality, high robustness, and improving the security of watermarked image in comparison by the LSB technique. In the rest, a mathematic formula as a key for embedding watermark has been presented to improve the security and a repeating method is discussed for improving the robustness as well. Moreover, in this chapter we are going to verify them and apply them to making the tradeoff between robustness and security as well. At the end of current chapter, two main phases have been discussed including embedding and extracting phases. In the first phase watermark will be embed into the cover image and then some attacks will be applied on them for evaluation and finally by using the extracting phase, the watermark will be extracted from the watermarked image. 3.2 Research environment The algorithms of this study have been implemented by MATLAB R2010b and visual C .Furthermore, both watermark and host images have been prepared by Adobe Photoshop cs3. The below figure (Figure 3.1) depicts watermark and host images in size of 50*50 and 256*256, respectively. Figure 3.1 Grayscale watermark (50*50) and host images (256*256) As can be seen from the Figure 3.1, two standard grayscale host images have been used, namely Lenna and Lake. The dimension of each one is 256* 256 pixels and these images are standard images that can be found and downloaded from the Internet. 3.3 Introduction of the new method A new approach to discover the best watermarked image quality as well as the best robustness is presented. In this section, a trade-off between quality and robustness must be achieved. To reach the goal of robustness, a repeating method has been introduced. Furthermore, by using a non-sequential mathematic formula for embedding, security has been reached. 3.3.1 The best quality of image There is no doubt that the quality of the watermarked image is the most significant parameter in all invisible methods. In all invisible watermarking methods, watermark must be embedded with having the least effect on the quality of the host data. In order to improve the quality of the watermarked image, beside the security an Intermediate Significant Bit (ISB) method has been used. This means that only one bit between bit 2 and bit 7 will be changed to guarantee the quality of watermarked image and the other 7 bits will not be changed. As can be seen from the figure 3.1, all eight layers of both watermark and host images have been presented. Figure 3.2 Eight layers of host and watermark images 3.3.2 The best robustness Robustness as another concern in watermarking also must be improved. In this section, this goal is achieved by repeating the embedded bits of watermark for certain times. This idea tries to make the algorithm more robust against attacks. Hence, the watermark is encoded into main signals T times. For instance, in this case we have considered T=3 and image is divided into blocks of with size of 3 pixels which all of the values of each block must be same as it is demonstrated in the figure 3.3. Figure 3.3 divided cover images into blocks To clarify, the below figure depicts how one pixel of watermark is embedded into the cover image. If the first pixel of watermark contains (11011101), the pattern of embedding by considering this assumption that the method of embedding is sequential, must be: Figure 3.4 Repeating bits in each block By this assumption that each bit of watermark image is repeated into the cover image three times, in embedding process, for extracting process also must be done the reverse of this process. This procedure protects watermark against attacks and can decrease the risk of losing bits that can be damaged by the attacks. This process, verify the majority of bits while extracting watermark to guess that what the value of watermark is. In other words, the algorithms read all three bits for one pixel and make decision between them. For example, if T=3, (000) represents 0, the (101) depicts 1, the (001) represents 0 and so on. To improve the robustness of the watermarked image, one bit in each pixel can be embedded in a block of pixels instead of embedding in only one pixel. This method has some advantages and disadvantages as well. The most important disadvantage of this method is decreasing the size of the watermark data which can be embedded in the host image. On the other hand, this approach can make a good robustness by increasing the size of each block. Moreover, each block can be divided into 3, 5, 7, and 9 pixels as can be seen from the figure 3.5. Figure 3.5 different sizes of blocks 3.3.3 Variable Repetition To further improve robustness of watermarking besides keeping a high capacity of watermarked image, the above mentioned method of repeating the embedding bits, based on average of pixels value can be chosen manually. The number of repeating might be high for the Most Significant Bits (MSB) of the embedded image and low for the Least Significant Bits (LSB). This method may ensure a high quality of embedding image, besides keeping the large amount capacity. 3.4 Integrate watermarking Beside the improvement of robustness in the watermarking system, other parameters also need to improve such as security and capacity of image which we can hide information inside it. Hence, a few steps need to be developed which guarantee that the proposed method is more robust, secure as well. Security beside the robustness is also important and may help to cut-off the attacks of malicious users. It is important to note that all of these stages can be executed separately. In the following section more details have been discussed. 3.5 Security in watermarking Security is a vital parameter in the watermarking system which must be considered. This means a strategy must be used to protect watermark against malicious users and also this strategy must be hidden. In other words, the watermark must not be able to detect by the people who do not right to modify or manipulate the watermarked image. In addition, the security in watermarking is represented same as security techniques which have been used in encryption methods (Seitz, 2005). The security of watermarking needs a separate research effort. Nevertheless, in this study a method for embedding bits of watermark has been presented to enhance the security of watermarked images. This method is using the encryption method to improve the security of watermark. As an example, Data Encryption Standard (DES) is using the shared key to encrypt the message. This technique might use to make more additional security to the watermarked image. The “Random Pixel Manipulation” is another example of encryption (Venkatraman, 2004). The idea behind this method is using the random positions of data instead of sequential positions to overcome the security problems. In the below section this method has described in more detail. 3.5.1 Random Pixel Manipulation methods The Random Pixel Manipulation technique uses a key that can specify the sequence of bits. This key provides an integer number to help generating a repeated sequence of unique numbers which its rage starts form 1 to N; where N is the number of current pixels. One of the advantages of this method is that the pixels are distributed to the whole image and decrease the distortion to the host image. Figure 3.6 Hashing bits into the host image (M. Zeki, 2011) As can be seen from the figure 3.6, it is clear that all the bits in watermark can be embedded thorough the host image to make it both robust and secure. This technique is quiet secure, but the problem is that we need to keep the places of bits in an extra place such as table or file. Regardless of this problem, this method is the best way to secure images even it can be executed with the repeated bits which have been described in section 3.3.2. Hence, if an area of watermarked image will be destroyed, the algorithm can use the other two bits from the other parts of image to extract the watermark in extracting process. 3.5.2 Zig-zag embedding matrix To improve the security of watermarked image, a non-sequential algorithm is needed. A Zig-zag array is a square collection of the first N^2 integers and the numbers increase sequentially as can be seen from the figure 3.7. Hence, the watermark pixels are not sequential and watermark cannot easily extract by unauthorized users (M.Padmaa, 2010). Figure 3.7 Zig-zag embedding matrix (M.Padmaa, 2010) The main aim of using Zig-Zag matrix is to improve the security of watermarking, besides using a repeating method to increase robustness. Hence, each pixel of watermark will be repeated three times within cover image by using this embedding algorithm to improve both security and robustness. 3.6 Enhancement of capacity in watermarked image by Pixel value Differencing (PVD) method The word of capacity in watermarking deals with the amount of data which can be embedded inside the host image. The main aim of capacity is to embed data as much as possible into the host image, and this goal should be achieved beside the other parameters like robustness and security which have been discussed in previous sections. In the embedding process, the host image must be divided into blocks, for example 3*3. This process also uses a key for embedding. Then, the watermark must be converted into the stream of binary bits and each bit must be embedded within the block of host image. Moreover, the number of bits which can be embedded in to each block is different and it depends on the each block and their values. Hence, each pixel might be different by another pixel. The sequence of watermarked data must be selected wisely and inherited by key which we can use this key for recovering the watermark later in extracting process. In the figure 3.8 the process of this technique has been illustrated as well. Figure 3.8 Capacity improvement techniques 3.7 Evaluation of Quality and Robustness To evaluate the robustness of the proposed algorithm we have used the Normalized Cross Correlation (NCC) .NCC is an important performance factor in any extracting component (Latha Parameswaran, 2006). (3.1) Where, W’(x,y) is the extracted watermark and W(x,y) is the watermark image. The PSNR greater or equal than 30 db (PSNR>=30 db) is the highest NCC and the algorithm will keep the position which is considered the best place for embedding watermark. Furthermore, to evaluate the quality of watermarked image we have used the Peak Signal to Noise Ratio (PSNR). Generally, a processed image is acceptable to the human eyes if its PSNR is greater than 30 dB (Chen, 2003). The greater PSNR has better image quality. The PSNR is defined as given by (2). Where m and n are the size of the host and watermarked images, ȕij is the pixel of the watermarked image and Įij is the pixel of the host image. 3.8 Embedding process In this phase, the watermark image must be embedded into the host image by using the techniques which have been described in previous sections. As can be seen from the previous sections, the methods which described try to improve the robustness and imperceptibility. Figure 3.9, indicates the process of embedding the watermark. The process will be started by converting watermark from pixel to bits. Hence, for each pixel there are 8 bits which each bit will embed by using a Zig-zag algorithm within different bit-planes in cover image. This process will be continued until all of the bits in watermark have been embedded. 3.9 Embedding Phase Figure 3.9 Embedding watermark phase’s framework 3.10 Extracting Phase In this phase, the watermark will be extracted from the watermarked image. This phase uses the reverse algorithms which have been used to embedding bits through the host image. The process has been shown on the below figure. Figure 3.10 extracting watermark phase’s framework 3.11 Applying attacks In this section, we will apply some attacks into the watermarked image and then quality and robustness of image will be evaluated. Figure 3.11 Applying attack phase’s framework 3.12 Dissertation's Framework Start Read the watermark image Extract stream bits of watermark image Find the best place for putting watermark into the host image Select the best layer to put the bits of watermark between 2 and 7 Repeating one bit of watermark tree times within the image Y Satisfied condition s? N Evaluate PSNR and NCC of watermarked image Read the watermarked image Watermarked Image Convert watermarked image to a string of bits Apply attacks such as Sharpening Filter, adding salt and pepper noise, Blurring Extract three bits of each pixel of watermark (T1, T2, and T3) Compare 3 bits together and make a decision between them to find of the exact value of the watermark’s bit. If two or three bits are 0 then value is 0, else the value is 1 Put the bits into the watermark buffer and make the watermark End Figure 3.12 Dissertation Framework 3.13 Summary In this chapter, some methods for embedding watermark into the host image have been proposed such as variable repatriation, Random Pixel Manipulation and etc. The proposed methods will be implemented in two main phases which are embedding and extracting watermark and they aim to reach the main objectives of this dissertation. The proposed algorithm in this methodology will be solved the problems in watermarking area like: imperceptibility, quality and robustness. The proposed algorithm uses Intermediate Significant Bits (ISB) as the main method for embedding information. At the end, we will evaluate our selected bits by using PSNR, MSE, and NCC to guarantee the quality and robustness of watermarked image, respectively. CHAPTER 4 IMPLEMENTATION AND EXPERIMENTAL RESULTS 4.1 Introduction In the previous chapter, we have theoretically verified the methods which have been used to implement the proposed technique. In the current chapter, we want to explain it more and the actual embedding into some standard host images have been implemented and the results are also provided. Moreover, a discussion to find the best bit plane of image in spatial domain has been presented as well. First of all, the host image is divided into the different sizes of blocks to enhance the robustness as the main parameter of watermarking. Then, the watermark is embedded into the different blocks to make it robust. As we explained in the previous chapter, each bit of watermark is repeated three times. Next, the place of each bit of watermark will be chosen by using the embedding algorithm through the host image. Next, the watermarked image will be created. Finally, some attacks will be applied into the watermarked image and the process of extracting tries to extract watermark as well as possible. 4.2 The chosen attacks First of all, some different attacks which might be happen for the images and make effects on them have been provided in the Figure 4.1. The attacks include: Gaussian filter, Speckle noise, Salt and Pepper, Passion, Set Removal (SR), Reset Removal (RR), and Blurring. Also, the comparison between them which includes the average percentage of effect per pixel has been displayed in the Table 4.1. Figure 4.1 Original Host image and different attacks The percentage of effect per pixel has been calculated by the blow mathematic formula. Where N is the number of pixels, P is the original pixel and P’ is the value of pixel after applying the attacks. As can be seen from the table 4.1, blurring filter has the most significant effect on the images, and the Gaussian filter has the least significant effect on the images. (4.1) Table 4.1 the impact of attacks on Images per pixel Attacks Average percentage of attacks’ effects Gaussian Filter 1.5735 Speckle Noise 5.9014 Salt and pepper 2.2041 Poisson 4.2337 Blurring 7.3540 4.3 The implementation In this section, the implementation about the methodology which has been discussed in the previous chapter has been done and the result also provided at the end of this chapter. The implementation tries to improve the robustness, the quality, and the security of current image watermarking methods. For achieving these goals, a repeating method has been proposed for the robustness and security as well. For reaching the goal of quality in the watermarked images an ISB method has been proposed. 4.4 Evaluation of watermarking methods There are several mathematic equations which have been using to qualify the watermarking algorithm, examining tests on the resulted watermarked image. They can evaluate the quality of watermarked image by comparing the watermarked image with the original cover image. In the rest, some of the equations for evaluating the watermark have been presented. 4.4.1 Mean Squared Error (MSE) One of the basic tests that were executed to test whether two images are similar is Mean Squared Error (MSE). This function could be easily written according to equation 4.2 and gets two images for measuring the average of the squares of the errors. The error is the amount by which the value specifies by the estimator differs from the quantity to be estimated. "# $ "%# & ! '( 4.4.2 Pick Signal to Noise Ratio (PSNR) Pick Signal to Noise Ratio (PSNR) is commonly used as a measure of quality of renovation for image compression and has a better evaluation than MSE, since it also evaluate the signal strength and error together. The equation 4.3 shows how this value is obtained. 2345 8 6*7 )*+, - ./0 1 (4.3) 4.5 watermarking with the best quality The quality in digital image watermarking plays a vital role and this parameter must be achieved during embedding process. In the embedding process there are eight options to put a bit of watermark in the host image. The most important thing is which bit planes (or layers) of host image is the most suitable and has the least effect in the quality of watermarked image. The table 4.2 has been demonstrated embedding image into all eight bit planes from Least Significant Bit (LSB) until Most Significant Bit (MSB). The table also provides the information about the quality of each watermarked image by using PSNR and MSE. 4.6 watermarking with improve security In this section, the approach which has been used to improve security has been discussed and implemented as well. The main idea is that information about the secret key leaks from the observations, for example, watermarked parts of content, available to the opponent. The security of watermarked image is achieved by using a non-sequential embedding algorithm. At the first, watermark will be converted into a stream of bits and then these bits are embedded by using Zig-zag algorithm in cover image. Furthermore, each bit of watermark will be embedded in each pixel of cover image (between bit-plane 2 and 7). The figure 4.2 depicts the process of embedding watermark. The simple LSB method is using sequential embedding method which watermark can be easily extracted by forgers and replace with another watermark. It can change the authority of the owner of watermarked image. This technique causes that security improve and also each bit of watermark will be embedded by using Zig-Zag algorithm in different bit-planes of cover image. So, if the attacker wants to recover watermark by using this technique, it will not work and still is secure. Figure 4.2 improving security by using Zig-zag algorithm By using two early mentioned techniques for improving robustness and security, we have come up by some results that have been provided in the next section. The result compares the proposed method by LSB method for two host images. The results also calculated PSNR and NCC for each bit planes in proposed and LSB method. Figure 4.3 applying Zig-zag algorithms to embed watermark 4.7 implementation and result As mentioned before, PSNR is used to evaluate quality of image and NCC is used to assess imperceptibility of watermarked image. The input for these functions are two images, including cover image and watermarked image then the function return a number to show the percentage accuracy of watermarked image. In the next section, the result for proposed method has been provided and also it is compared with LSB and MSB techniques. Table 4.2: watermarked images by using proposed method for each bit plane and their PSNR value (Lena). Bit Plane Watermarked Image s PSNR Bit Plan es 1,2,3 45.7101 3,4,5 33.2965 1,3,4 39.8263 3,5,6 31.7147 2,3,4 39.6664 4,5,6 31.3294 2,4,5 33.4402 5,6,7 30.1412 Watermarked Image PSNR Table 4.3: Watermarked images by using proposed method for each bit plane and their PSNR value (Lake). Bit Plane Watermarked Image s 1,2,3 PSNR Bit Plan es 45.6327 3,4,5 1,3,4 31.6079 39.9086 4,5,6 31.2706 34.7739 34.5692 3,5,6 2,4,5 PSNR 40.0745 2,3,4 Watermarked Image 5,6,7 29.3477 Table 4.4: Bit Plan e The watermarked images by using LSB until MSB for each bit plane and their PSNR value (Lena). PSNR Bit Plan e 1 59.0652 5 34.5335 2 52.9430 6 35.1333 3 46.9393 7 35.0785 4 40.8651 8 33.7072 Watermarked Image Watermarked Image PSNR Table 4.5: Bit Plane Watermarked images by using LSB until MSB for each bit plane and their PSNR value (Lake). PSNR Bit Plane 1 59.0291 5 36.141 7 2 52.9543 6 33.917 9 3 46.9160 7 33.178 2 4 41.1983 8 34.290 9 Watermarked Image Watermarked Image PSNR As can be seen from the above results, the proposed method by using repeating bits in three different Bit-planes (1, 2 and 3) has the least effect on the cover image with PSNR over 45; meanwhile the worst bit planes are 5, 6 and 7 with PSNR under 30. Nevertheless, the LSB method by using the first bit has the best PSNR with around 60, but LSB is so fragile and the watermark can be extracted easily by forgers. The proposed method uses different bit planes and uses a Zigzag embedding algorithm to implant watermark onto the cover image. In this part, the chosen attacks have been applied on the watermarked images and the NCC has been calculated for both proposed method and LSB in tables 4.6, 4.7, 4.8, respectively. Table 4.6: NCC value of proposed method BitPlanes Blurring Speckle Gaussian Salt and pepper 1,2,3 0.7536 0.7680 0.7221 0.7966 1,3,4 0.6535 0.6876 0.6249 0.7954 2,3,4 0.7537 0.7503 0.7444 0.7970 2,4,5 0.7228 0.71.88 0.7257 0.7942 3,4,5 0.6532 0.6105 0.6488 0.7944 3,5,6 0.6359 0.5855 0.6097 0.8073 4,5,6 0.6056 0.5595 0.5761 0.8068 5,6,7 0.5939 0.5470 0.5536 0.8499 Table 4.7: NCC Value for LSB until MSB under different attacks (Lena) BitPlane Blurring Speckle Gaussian Salt and pepper 1 0.7809 0.7804 0.7820 0.7853 2 0.7803 0.7800 0.7820 0.7869 3 0.7789 0.7781 0.7820 0.7857 4 0.7760 0.7748 0.7820 0.7815 5 0.7759 0.7740 0.7820 0.7810 6 0.7383 0.7346 0.7820 0.7467 7 0.7367 0.7298 0.7820 0.7421 8 0.8352 0.8167 0.7820 0.8386 Table 4.8: NCC value for LSB until MSB under different attacks (Lake) BitPlane Blurring Speckle Gaussian Salt and pepper 1 0.9520 0.9919 0.7820 0.9935 2 0.9514 0.9948 0.7820 0.9943 3 0.9503 0.9923 0.7820 0.9919 4 0.9453 0.9881 0.7820 0.9900 5 0.9348 0.9788 0.7820 0.9778 6 0.9641 0.9994 0.7820 1 7 0.9893 1 0.7820 1 8 0.7124 0.7130 0.7820 0.7289 The extracted watermark‘s result for proposed method has been done and illustrated in the tables 4.8 and 4.9. The attacks that have been chosen, described in previous chapter s and the amount of their effects has been written at the first of this chapter. Table 4.9: Bitplanes Extracted watermark after attack in proposed method (Lena) Salt and Pepper Gaussian Speckle Poisson Blurring 57.8942 35.1767 35.1767 35.1211 35.1767 33.1767 35.1767 31.9604 30.2204 58.0448 29.2292 28.8208 29.1933 28.9032 54.8024 29.1511 29.0914 29.2553 28.7651 57.1409 29.0285 29.7285 28.9973 28.2251 55.2978 28.9489 29.7255 29.9973 29.0686 55.2978 28.8878 31.5728 30.5728 28.9786 58.4180 29.4963 35.1767 31.9604 28.6906 1,2,3 PSNR 1,3,4 PSNR 58.0278 2,3,4 PSNR 2,4,5 PSNR 3,4,5 PSNR 3,5,6 PSNR 4,5,6 PSNR 5,6,7 PSNR Table 4.10: BitPlanes Extracted watermark after applying attacks in proposed method (Lake) Salt and Pepper Gaussian Speckle Poisson Blurring 58.0448 31.1767 34.1767 32.1211 30.1767 58.0448 30.1767 31.9604 31.2204 58.0444 29.2292 29.5179 29.1933 28.9032 54.8011 29.1511 30.5596 29.2553 28.7651 56.1412 29.0285 31.1895 28.9973 28.2251 55.2978 28.9417 32.2277 29.9973 29.0686 55.2978 28.8878 32.0976 30.8828 28.7786 58.4180 29.4963 36.5238 31.9604 28.6906 1,2,3 PSNR 1,3,4 PSNR 30.4285 2,3,4 PSNR 2,4,5 PSNR 3,4,5 PSNR 3,5,6 PSNR 4,5,6 PSNR 5,6,7 PSNR Table 4.11: BitPlanes Extracted watermark after applying the attacks in LSB method Salt and Pepper Gaussian 41.3402 27.2872 40.9581 27.4718 42.6408 Speckle Poisson Blurring 27.2553 27.3459 27.5265 27.3376 27.4557 27.2759 27.3816 27.3793 27.2623 41.3980 27.3839 27.5088 27.4279 27.7070 41.3206 27.4665 29.5288 27.8759 27.8265 41.6746 27.6354 32.2871 29.7421 28.2866 41.8143 29.4687 34.7684 32.8219 27.4516 40.7806 32.2926 39.5445 36.0818 27.4319 1 PSNR 27.4521 2 PSNR 3 PSNR 4 PSNR 5 PSNR 6 PSNR 7 PSNR 8 PSNR As can be inferred from the tables 4.9, 4.10, and 4.11, the proposed method is more robust in comparison by LSB and MSB methods. Nevertheless, most of the attacks tend to destroy low bits (1, 2, and 3), but the proposed method can extract watermark after the attacks by acceptable percentage of healthy pixels in watermark even in low bit-planes. Surprisingly, as can be seen from the result of proposed method, the average PSNR in all of the 8 layers is approximately same by over 30 db. Hence, above result proves that watermark can embed onto the all bit-planes of cover image regardless of the visibility of watermark. The below figures also provided information about the PSNR of watermark after extraction for two LSB and proposed method. The x-axis indicated 8 bit-planes and the PSNR is indicated on yaxis. Figure 4.4: Applied salt and pepper attack for both proposed and LSB method Figure 4.5 Figure 4.6 Applied Gaussian attack for both proposed and LSB method applied Speckle attack for both proposed and LSB method Figure 4.7 Applied Poisson attack for both proposed and LSB method Figure 4.8 Applied Blurring attack for both proposed and LSB method 4.8 Improving the capacity The capacity is also another important parameter of watermarking and it is needed to be increased. Obviously, the best method of watermarking in term of capacity is embedding more information into the host image. The capacity of the watermark method can be simply evaluated by increasing the length of the watermarking message. Furthermore, any watermarking method is not capable of holding more than a definite length of data or it might endanger its imperceptibility. 4.9 Summary In this chapter, the quality and robustness of watermarked image has been found to enhance by using a Zig-zag algorithm and repeating method in different bitplanes. Zig-zag algorithm tries to embed watermark in a non-sequential way into the cover image, and repeating method helps to embed watermark in different places and different bit-planes of cover image. Therefore, it causes that if a bit-plane has been destroyed by mentioned attacks, uses two other bit-planes and makes decision between them by using the majority of bit values. The experimental results also provided in this chapter that specify the proposed method has many advantages compared to other spatial methods such as LSB and MSB. The host image was portioned into block size of 3 pixels and proposed method was also applied to these blocks. The result also shows that the NCC and PSNR value were improved in the embedding number for all attacks and all mentioned attacks has approximately the same effect in all bit-planes at over 30db. The proposed method aims to embed and extract watermark pixels in grayscale, but it could be black and white image. Hence, each pixel in watermark can be embedded in each pixel of the host image. Therefore, regardless of the repeating method for improving the robustness, the space for the watermark could be the same as host image. For example, if we have a cover image with size of 256*256, we can embed a watermark the same is cover image size. CHAPTER 5 CONCLUSION 5.1 Introduction Over the last half century the pace of change in term of types of documents and methods of transferring data from one part to another part has been increased where billions of bits of data is created in every fraction of a second and transfer through the Internet rapidly. Hence, digital watermarking came as a method and a tool to overcome copyright concerns for digital documents. Watermarking is a method that is use to hide information or identify data within digital documents. Moreover, digital documents can be divided into video, audio and image. In this dissertation, we have focused primarily on the watermarking of digital images. Digital watermarking is becoming popular, mainly for embedding undetectable marks, such as owner or copyright information. 5.2 Contribution and Achievement In this study, gray scale images have been applied to embed an evidence of the owner of the digital image. Digital images can be divided into two domains including frequency and spatial domains and the proposed method implemented in spatial domain to embed watermark within cover image. The proposed method also used Intermediate Significant Bits (ISB) to improve the security and quality of watermarked image. At the same time, some attacks also applied to the watermarked image due to by attacking watermarked image it become very complex to recover watermark back from the watermarked image and even if it extracted one may no longer use it to prove the ownership and copyrights. Therefore, one of the goals was to find such regions in the cover image which are very resistant and stable against attacks. In this study, we have proposed a new watermarking method to improve previous works in spatial domain in terms of robustness and imperceptibility. Moreover, we have discussed about the previous studies in the literature. Also, some of the remained problems in spatial domain have been discussed such as security, imperceptibility, quality, capacity, and robustness. The proposed method has been aimed to improve robustness of watermarked image by using repeated variable for each pixel of watermark and it improves the probability of extracting intact bits after attacks and the security guaranteed by using a secret key and none-sequential embedding method (Zig-zag). The experimental result for the current dissertation has been recognized based on the literature review which has been discussed in chapter two about spatial domain methods such as LSB and ISB techniques. However, the experiments in chapter four have been proved that the problems that occur in the spatial domain have been solved and the result compared with other recent techniques. Therefore, the achievements of the dissertation are summarizes as follow: a. Improve the problem of quality and imperceptibility of watermarked image by using an ISB method in spatial domain. For this goal, intermediate bits have been chosen to be imperceptible enough and no one can detect the existence of watermark easily. b. Improve the robustness of watermarked image by using a repeating method for each pixel of watermark within cover image. Using an reparative technique in different bit planes helps to improve watermarking robustness. In other words, if a watermarked image has been attacked between sender and receiver or during its life time, it still can extract and can use other bits of repeated pixel within the watermarked image. c. Improve the security of watermarked by distributing pixel throughout the cover image (none-sequential method). Obviously, using a none-sequential embedding algorithm within cover image helps to improve the security of image against forgers. In other words, we have used a secret key for embedding and then it will be used in extracting process. 5.3 Future Work Although the proposed method is tried to achieve the best quality, robustness, and security for watermarking in grayscale images, but some other important parameters must be considered as well. In the following some of the remained problems have been listed as: 1. The most important remained parameter in watermarking is capacity. In this study, a repeated method for improving robustness has been used, meanwhile using this technique causes that the amount of watermark information decreased due to watermark must be repeated T times into the cover image. Nevertheless, some researchers have been tried to increase the amount of information into the cover image, but it still needs to improve. 2. The Integrity is also another important concern in digital image watermarking. In other words, when a watermark is embedded within the cover image, might be extracted and embedded with a fake watermark. Hence, the ownership of cover image will endanger about this type of attack. Therefore, the methods need to overcome these kinds of problems. 3. Although, grayscale image for this study were used, but the present study also can extend for colored image (RGB). 5.4 Summary The general idea of this dissertation is to improve the imperceptibility and robustness of digital image watermarking by proposing new intermediate significant bits method and reparative watermark bits within different bit planes of watermarked image, respectively. The security also in the dissertation has been improved, due to using none-sequential embedding process and using a secret key to embedding and extracting watermark. In order to evaluate the performance of proposed method, several experiments have been conducted. Finally, the proposed method has been compared with conventional method and also the experiment results did reveal that the proposed method is capable to solve the problems related to imperceptibility and robustness. REFERENCES ARYA, D. 2010. A Survey of Frequency and Wavelet DomainDigital Watermarking Techniques. 1. CHAN, C.-K. 2004. Digital Image Watermarking Using Localized Biorthogonal Wavelets Pattern Recognition, 469-474. CHING-TANG HSIEH , Y.-K. W. 2001. Digital Image Multi resolution Watermark Based on Human Visual System Using Error Correcting Code Tamkang Journal of Science and Engineering,, Vol. 4, 8. COX, I. J. 2008. Digital Watermarking and Steganography, Morgan Kaufmann Publishers. GANG LIU , Y. W., WENJUAN HE , JING LIU 2010. A watermarking algorithm based on direction of image specific edge. 3 1146-1150. GIL-JE LEE, E.-J. Y., KEE-YOUNG YOO 2008. A New LSB Based Digital Watermarking Scheme with Random Mapping Function. Ubiquitous Multimedia Computing. Hobart, ACT. HYEONG-IN CHOI, T.-W. K., SONG-HWA KWON,HWAN PYO MOON,SUNG HA PARK,HEON-JU SHIN,JUNG-KYO SOHN 2010. Digital watermarking of polygonal meshes with linear operators of scale functions. Comput. Aided Des., 42, 163-172. IBRAHIM NASIR, Y. W., JIANMIN JIANG 2007 A New Robust Watermarking Scheme for Color Image in Spatial Domain. Third International IEEE Conference on Signal-Image Technologies and Internet-Based System. JIANG NAN , W. J. N. X. Y. Y. 2006. The Quantificational Relation of Imperceptibility, Robustness and Hiding Rate in Digital Watermarking. Guilin. JINGUANG SUN, S. L. 2010. Geometrical Attack Robust Spatial Digital Watermarking Based on Improved SIFT. Proceedings of the 2010 International Conference on Innovative Computing and Communication and 2010 Asia-Pacific Conference on Information Technology and Ocean Engineering. IEEE Computer Society. K.GANESAN, T. K. G. 2010. Multiple Binary Images Watermarking in Spatial and Frequency Domains An International Journal(SIPIJ), 2, 148-159. KER, A. D. 2004. Improved Detection of LSB Steganography in Grayscale Images, Springer Berlin Heidelberg. KUMAR N.M. , M., T. , SAPTHAGIRIVASAN, V. 2011 Non blind image watermarking based on similarity in contourlet domain. Recent Trends in Information Technology (ICRTIT), 2011. LANGELAAR, G. C., SETYAWAN, I. ; LAGENDIJK, R.L. 2000. Watermarking digital image and video data. A state-of-the-art overview. Signal Processing Magazine, IEEE. LATHA PARAMESWARAN, A. K. A. 2006. A Robust Image Watermarking Scheme using Image Moment Normalization. World Academy of Science, Engineering and Technology. M. ZEKI , A. A. M. 2011. ISB Watermarking Embedding: A Block Based Model. 10, 841-848. M.PADMAA, D. Y. V. 2010. ZIG-ZAG PVD – A Nontraditional Approach. 5, 5-10. MAKHLOGHI, M. 2011. A new robust blind DWT-SVD based digital image watermarking. Electrical Engineering (ICEE). Tehran. MARTIN KUTTER, S. V., ALEXANDER HERRIGEL 2000. The Watermark Copy Attack. Security and Watermarking of Multimedia Content II. San Jose, CA, USA. MEGALINGAM, R. K., NAIR, M.M. , SRIKUMAR, R. , BALASUBRAMANIAN, V.K. , SARMA, V.S.V. 2010. Performance Comparison of Novel, Robust Spatial Domain Digital Image Watermarking with the Conventional Frequency Domain Watermarking Techniques. Signal Acquisition and Processing, 2010. ICSAP '10, 349- 353. MIR SHAHRIAR EMAMI, G. B. S. 2011. Set Removal Attack: A New Geometric Watermarking Attack 2011 International Conference on Future Information Technology. Singapore. MOULIN, P., O'SULLIVAN, J.A. 2003. Information-theoretic analysis of information hiding. Information Theory, IEEE Transactions on, 49, 563- 593. NOUR EL-HOUDA GOLEA, R. S., REDHA BENZID 2010. A bind RGB color image watermarking based on singular value decomposition. Proceedings of the ACS/IEEE International Conference on Computer Systems and Applications - AICCSA 2010. IEEE Computer Society. PUNEET KR SHARMA, A. R. 2012 INFORMATION SECURITY THROUGH IMAGE WATERMARKING USING LEAST SIGNIFICANT BIT ALGORITHM CS & IT-CSCP 2012 S. SHEFALI, S. M. D., AND S. G. TAMHANKAR 2007. Attack Detection through Image Adaptive Self Embedding Watermarking. World Academy of Science, Engineering and Technology 29 2007, 298-304. SEITZ, J. 2005. Digital Watermarking for Digital Media. SONG-HWA KWON, T.-W. K., HYEONG IN CHOI,HWAN PYO MOON,SUNG HA PARK,HEON-JU SHIN,JUNG KYO SOHN 2011. Blind digital watermarking of rational Bézier and B-spline curves and surfaces with robustness against affine transformations and Möbius reparameterization. Comput. Aided Des., 43, 629-638. TOSHIHIKO YAMASAKI, Y. N., AND KIYOHARU AIZAWA 2009. AN OBJECT-BASED NON-BLIND WATERMARKING THAT IS ROBUST TO NON-LINEAR GEOMETRICAL DISTORTION ATTACKS. Image Processing (ICIP), 2009 16th IEEE International Conference Cairo. VENKATRAMANS , A. A. P., M. 2004. Significance of steganography on data security. International Conference on Information Technology: Coding and Computing (ITCC'04). VOLOSHYNOVSKIY, S., PEREIRA, S. , PUN, T. , EGGERS, J.J. , SU, J.K. 2001. Attacks on digital watermarks: classification, estimation based attacks, and benchmarks. Communications Magazine, IEEE, 39, 118-126 YONGJIAN HU , K., S. , JIWU HUANG 2004. Using invisible watermarks to protect visibly watermarked images. Circuits and Systems, 2004. ISCAS '04. Proceedings of the 2004 International Symposium on.