DUAL ADAPTIVE WATERMARKING SCHEMES FOR DICOM IMAGES

A PROJECT REPORT submitted by MANJARI TYAGI(091237) PALLAVI JAIN(091310) TAPAS TRIVEDI(091324) UNDER THE SUPERVISION OF DR.SHISHIR KUMAR (HOD CSE)

May-2013 submitted in partial fulfillment for the award of the degree of

Bachelor of Technology IN
Department of Computer Science & Engineering

Department of Computer Science & Engineering JAYPEE UNIVERSITY OF ENGINEERING & TECHNOLOGY, AB ROAD, RAGHOGARH, DT. GUNA-473226 MP, INDIA

JAYPEE UNIVERSITY OF ENGINEERING & TECHNOLOGY

A.B. ROAD, P.B. No. 1, RAGHOGARH, DIST: GUNA (M.P.) INDIA. Phone: 07544 267051, 267310 - 14 Fax: 07544 267011 Website: www.juet.ac.in

CERTIFICATE

This is to certify that the work titled Dual Adaptive Watermarking for DICOM images submitted by Manjari Tyagi (091237), Pallavi Jain (091310), and Tapas Trivedi(091324) in partial fulfillment for the award of degree of Bachelor of Technology in Computer Science of Jaypee University of Engineering & Technology, Guna has been carried out under my supervision. This work has not been submitted partially or wholly to any other University or Institute for the award of this or any other degree or diploma.

Signature of Supervisor (Dr. Shishir Kumar ) Sr. Lecturer Date:

i

ACKNOWLEDGMENT
We are extremely grateful and remain indebted to our guide Dr. Shishir Kumar(HOD) for being a source of inspiration and for his constant support in the Design, Implementation and Evaluation of the project. We are thankful to him for his constant constructive criticism and invaluable suggestions, which benefited us a lot while developing the project on DUAL ADAPTIVE WATERMARKING FOR BIOMEDICAL IMAGES. He has been a constant source of inspiration and motivation for hard work. He has been very co-operative throughout this project work. Through this column, it would be our utmost pleasure to express our warm thanks to him for their encouragement, co-operation and consent without which we mightnt be able to accomplish this project. We also express our gratitude to Mr. Puneet Pandey for providing us the infrastructure to carry out the project and to all staff members who were directly and indirectly instrument in enabling us to stay committed for the project.

Manjari Tyagi [091237]

Pallavi Jain [091310]

Tapas Trivedi [091324]

ii

EXECUTIVE SUMMARY

In Dual Adaptive Watermarking for Biomedical Images we mainly deal with the watermarking procedures currently being carried out in the field of biomedical images, and also try to improve upon the same by proposing an improved scheme. The digital form of medical images have a lot of advantages over its analog form such as ease in storage and transmission. Medical images in digital form must be stored in a secured environment to preserve patient privacy. It is also important to detect modifications on the image. These objectives are obtained by watermarking in medical images. In this project we will mainly deal with the DWT(Discrete Wavelet Transform) for watermarking of biomedical images. It is a method for decomposing an image into 4 subbands of varying frequencies, that aids in localization of values in an image, energy compaction and also decorrelation of values, leading to greater security than normal spatial domain based transforms. We firstly employ a conventional DWT based watermarking scheme on a set of biomedical images, and then apply attacks on them to ascertain their quality of robustness and imperceptibility. We also employ a proposed scheme under which we highlight the regions of interest and non-interest in an image, and apply separate watermarks based on their desired requirements. This composite image, is tested by applying several image processing and geometrical distortion attacks. We will then compare the results of both, the original scheme and the proposed scheme by comparing the state of the images after applying the attacks stated.

iii

TABLE OF CONTENTS
CERTIFICATE.....i ACKNOWLEDGEMENT..................ii EXECUTIVE SUMMARY....iii 1. Introduction.1 1.1. Introduction.1 1.2. Digital Images....1 1.2.1. Types of Images.1 1.2.2. Formats of images2 1.3. Digital Image Watermarking..4 1.3.1. Introduction....4 1.3.2. Structure of a Digital Watermark..5 1.3.3. Principle of Watermarking.6 1.3.3.1. 1.3.3.2. 1.3.3.3. 1.3.3.4. 1.3.3.5. 1.3.3.6. Watermark Insertion.....6 Watermark Generation..6 Encoding Process...7 Watermark Extraction7 Decoding Process..7 Comparison Process...7

1.3.4. Requirements of Watermarking.....8 1.3.5. Applications of Watermarking....9 1.3.6. Attacks on Watermarks...10 1.3.7. Types of Watermarking...11 1.3.8. Watermark Embedding Techniques............12 1.3.8.1. 1.3.8.2. 1.3.8.3. Spatial Domain Techniques............12 Transform Domain Techniques...13 Contourlet Domain Techniques13

2. Bio-Medical Image Watermarking.14 2.1. Introduction.....15 2.2. Security and Medical Information.15 2.3. Watermarking Medical Applications.17 2.4. Case Studies on Watermarking for Bio-medical Images18 2.4.1. Authentication and Tracing.18
iv

2.4.2. EPR Diffusion..18 2.5. Requirements for Medical Image Watermarking19 2.5.1. Reversible Watermarking..19 2.5.2. Defining Regions of Interest and Regions of Insertion...19 2.5.3. Integrity Control...19 2.5.4. Authentication..19 2.5.5. Dual Watermarking Scheme20 2.5.6. Adaptive Watermarking...20 3. Related Studies...21 3.1. General Watermarking System..22 3.2. Watermarking in Spatial Domain..22 3.2.1. Insertion of Watermark22 3.2.2. Extraction of Watermark..24 3.3. Watermarking in Transform Domain25 3.3.1. Discrete Cosine Transform (DCT)...25 3.3.2. Discrete Wavelet Transform (DWT)...26 3.3.3. Singular Value Decomposition28 3.4 Watermarking in Contourlet Domain..31 3.5 Test Standards..33 3.5.1 Peak Signal to Noise Ratio (PSNR)..33 3.5.2 Structured Similarity Measure (SSIM).............33 3.5.3 Luminance Comparison (LC)....33 3.5.4 Contrast Comparison (CC)34 3.5.5 Structured Comparison (SC)..34 3.5.6 Bit Error Rate (BER).............34 3.5.7 Normalized Cross Correlation (NCC)...34 4. Design and Implementation...35 4.1 Implementation tools and pre-requisites....36 4.2 Algorithms Implemented..37 4.2.1. Basic DWT Algorithm.....37 4.2.1.1 Embedding..37 4.2.1.2 Extraction37
v

4.2.2. DWT Dual and Adaptive Algorithm........38 4.2.2.1 Embedding in Region of Interest....38 4.2.2.2 Embedding in Region of Non Interest....38 4.2.2.3 Extraction from Region of Interest................39 4.2.2.4 Extraction from Region of Non Interest39 4.3. Attacks to test Robustness and Imperceptibility..40 4.3.1 Table showing comparison of PSNR values of original to proposed scheme.40 4.3.2 Table showing comparison of NCC values of original to proposed scheme.41 5 Conclusion and Appendices..45 5.1 Conclusion...46 5.2 Appendices...47 5.2.1 Appendix A : Attacks done on images47 5.2.2 Appendix B : Extracting Region of Interest and Region of Non-Interest...48 REFERENCES..50 BIBLIOGRAPHY..52 PERSONAL DETAILS.........53

vi

LIST OF FIGURES
Fig1.1 Fig.1.2 Fig.1.3 Fig.1.4 Fig.1.5 Fig1.6 Fig1.7 Fig.1.8 Fig 1.9 Fig 1.10 Fig 2.1 Fig 2.2 Fig 3.1 Fig 3.2 Fig 3.3 Fig 3.4 Fig 3.5 Fig 3.6 Fig 3.7 Fig3.8 Fig 4.1 Fig 4.2 Fig 5.1 File formats Watrmark Embedding Watermark Extraction Principle of watermarking Watermark Embedder (Encoder) Simple Decoding Process Comparing Process Primary requirements of Watermarking Algorithms Watermark Attacks Types of Watermarking Example of Medical image tampering Security Tools Embedding and extraction of watermark DCT Implementation Flowchart DWT Implementation DWT hybrid watermark embedding(using SVD) DWT watermark extraction block schematic SVD based watermark Embedding Block Diagram SVD based watermark Extracting Block Diagram Contourlet Domain embedding and extraction block schematic Implementation of pure DWT algorithm (non-adaptive) Region of interest and non-interest, before, and after watermarking procedure Region of Interest and non-Interest

vii

CHAPTER 1 INTRODUCTION

viii

1.1. INTRODUCTION
In the recent years, medical images are produced from a wide variety of digital imaging equipments, such as computed tomography (CT), magnetic resonance imaging (MRI), computed radiography (CR) and so forth. With the increasing use of internet and appearance of new system such as picture archiving and communication systems (PACS), the usability of digital form of medical images has been increased . Images in digital imaging equipments can be printed on films or papers. Moreover, in these equipments images with patient data in DICOM format can be stored on different types of storage media such as CD or DVD. Insurance companies, hospitals and patients may want to change this data for various reasons. Therefore, protecting medical images against this threat is necessary. Watermarking can be used as a solution.

1.2. DIGITAL IMAGES

A digital image is composed of a number of elements, each having a particular location and value. These elements are referred as picture elements, image elements and pixels. Pixels is term used to denote elements of a digital image. A image can be defined as a two-dimensional function f(x,y) where x and y are spatial coordinates and the amplitude of f at any pair of coordinates is called the intensity or the gray level of the image at that point. When x, y a nd the amplitude values of f are all finite and discrete, the image is known a Digital Image. A digital image can be represented naturally as a matrix. 1.2.1 Types of Images 1. Bi-level Image : This is the black and white image having only two values. Each pixel in such type of image requires only one bit for representation i.e. either 0 (for black) or 1(for white). 2. Gray-Scale Image : In Gray Scale Image, any pixel can have any of n values between 0 to n-1 where n is the maximum number of bits required to represent any pixel value. There can be 2n shades of Gray. 3. Continuous- Tone Image : In continuous- tone images, the adjacent pixel values differ just by one or few units. For eyes, it is very hard to distinguish the difference between the adjacent pixel values. Any pixel in such images can be represented by either a single large image (Gray-scale) or by three components (in the case of a colour image). Continuous tone images are generally natural images which are not having any sharp edges. 4. Discrete Tone Image : These are generally graphical or synthetic images. These types of images have sharp edges and no blurring effect. Artificial objects, lines, text have sharp and well-defined edges and therefore vary in contrast from the rest of the background. These images are called artificial images. 5. Cartoon- like Image : These types of images consist of uniform areas, having uniform colors. The adjacent areas have different colours. 6. DICOM Image : DICOM (Digital Imaging and Communications in Medicine) is a standard for handling, storing, printing, and transmitting information in medical imaging. It includes a file format definition and a network communications protocol. It is the international standard for medical images and related information (ISO 12052). It defines the formats for medical images that can be exchanged with the data and quality necessary for clinical use. DICOM is implemented in almost every radiology, cardiology imaging, and

radiotherapy device (X-ray, CT, MRI, ultrasound, etc.), and increasingly in devices in other medical domains such as ophthalmology and dentistry. With tens of thousands of imaging devices in use, DICOM is one of the most widely deployed healthcare messaging standards in the world. There are literally billions of DICOM images currently in use for clinical care. Since its first publication in 1993, DICOM has revolutionized the practice of radiology, allowing the replacement of X-ray film with a fully digital workflow. Much as the Internet has become the platform for new consumer information applications, DICOM has enabled advanced medical imaging applications that have changed the face of clinical medicine. From the emergency department, to cardiac stress testing, to breast cancer detection, DICOM is the standard that makes medical imaging work for doctors and for patients.

1.2.2

Formats of images

1.2.2.1 TIFF is a very flexible format that can be lossless or lossy. In practice, TIFF is used almost exclusively as a lossless image storage format that uses no compression at all. Most graphics programs that use TIFF do not compression. Consequently, file sizes are quite big. And TIF is the most versatile, except that web pages don't show TIF files.

1.2.2.2 PNG is also lossless storage format. However, in contrast with common TIFF usage, it looks for patterns in the image that it can use to compress file size. The compression is exactly reversible, so the image is recovered exactly. Feature of PNG is transparency for 24 bit RGB images. PNG is slightly slower to read or write.

1.2.2.3

GIF creates a table of up to 256 colors from a pool of 16 million. If the image has a fewer than 256

colors, GIF can render the image exactly. When the image contains many colors, software that creates the GIF uses any of several algorithms to approximate the colors in the image with the limited palette of 256 colors available. Better algorithms search the image to find an optimum set of 256 colors. Sometimes the GIF uses the nearest color to represent each pixel, and sometimes it uses error diffusion to ad just the color of nearby pixels to correct for the error in each pixel. Thus, GIF is lossless only for images with 256 colors or less. For a rich, true color image, GIF may lose 99.998% of the colors. It is very good for web graphics.

1.2.2.4 JPEG is optimized for photographs and similar continuous tone images that contain many colors. It can achieve astounding compression ratios even while maintaining very high image quality. JPG works by analyzing images and discarding kinds of information that the eye is least likely to notice. It stores information as 24 bit color.

1.2.2.5 RAW is an image output option available on some digital cameras. Though lossless, it is a factor of three of four smaller than TIFF files of the same image. The disadvantage is that there is a different RAW format for each manufacturer, and so you have to use the manufacturers software to view thw images.

1.2.2.6 BMP is an uncompressed proprietary format invented by Microsoft. There is really no reason to ever use this fomat. The BMP file format (Windows bitmap) handles graphics files within the Microsoft Windows OS. Typically, BMP files are uncompressed, hence they are large; the advantage is their simplicity and wide acceptance in Windows programs.

Fig1.1 File formats

1.3 DIGITAL IMAGE WATERMARKING

1.3.1 Introduction
Digital watermarking is the process of inserting a digital signal or pattern (indicative of the owner of the content) into digital content. The signal, known as a watermark, can be used later to identify the owner of the work, to authenticate the content, and to trace illegal copies of the work. The concept of digital watermarking is driven by the need to caption, and control copyrights for digital media including images and video. Early work in the same identified redundant properties of an image or its encoding that can be modified to encode watermarking information. The early emphasis was on hiding data, since the envisioned applications were not concerned with signal distortions or intentional tampering that might remove a watermark. However as watermarks are increasingly used for purposes of copyright control, robustness to common signal transformations and resistance to tampering have become important considerations. Researchers have recently recognized the importance of perceptual modeling and the need to embed a signal in perceptually significant regions of an image, especially if the watermark is to survive lossy compression. However this requirement conflicts with the need for the watermark to be imperceptible. There are several approaches that address these issues. Recently there has been significant interest in watermarking. This is primarily motivated by a need to provide copyright protection to digital content such as audio, images and video. Digital representations of copyrighted material such as movies offer many advantages. However the fact that an unlimited number of perfect copies can be illegally produced is a serious threat to the rights of content owners. Watermarking can be used for owner identification, to identify the content owner, fingerprinting, to identify the buyer of the content, for broadcast monitoring to determine royalty payments, and authentication, to determine whether the data has been altered in any manner from its original form. The latter purpose is somewhat different from those of copyright control and the characteristics thereof may be different. A number of technologies are being developed to provide protection from illegal copying. Two complimentary techniques are encryption and watermarking. Encryption protects content during the transmission of the data. Watermarking compliments encryption by embedding a signal directly into the data. Thus the goal of a watermark is to always remain present in the data. There are several properties that a watermark must exhibit. These include that it must be difficult to notice, robust to common distortions of the signal, resistant to malicious attempts to remove the watermark, support a sufficient data rate commensurate with the application, allow multiple watermarks to be added and that the decoder be scalable.

1.3.2 Structure of a digital watermark

The structure of a digital watermark is shown in the following figures.

Fig.1.2 Watrmark Embedding The material that contains a digital watermark is called a carrier. A digital watermark is not provided as a separate file or a link. It is information that is directly embedded in the carrier file. Therefore, the digital watermark cannot be identified by simply viewing the carrier image containing it. Special software is needed to embed and detect such digital watermarks. Kowas SteganoSign is one of these software packages. Both images and audio data can carry watermarks. A digital watermark can be detected as shown in the following illustration.

Fig.1.3 Watermark Extraction

1.3.3 Principle of Watermarking

In general, any watermarking algorithm consists of three parts: The watermark (payload) The encoder (marking insertion/embedding algorithm) The decoder and comparator (verification or extraction or detection algorithm)

Fig.1.4 Principle of watermarking 1.3.3.1 Watermark insertion: Watermark insertion involves watermark generation and encoding process

Fig.1.5 Watermark Embedder (Encoder)

1.3.3.2 Watermark Generation: Each owner has a unique watermark or an owner can also put different watermarks in different objects the marking algorithm incorporates the watermark into the object. The verification algorithm authenticates the object determining both the owner and the integrity of the object. The watermark can be a logo picture, sometimes a binary picture , sometimes a ternary picture ; it can be a bit stream or also an encrypted bit stream etc. The encryption may be in the form of a hash function or encryption using a secret key . The watermark generation process varies with the owner.

1.3.3.3 Encoding Process: In the encoding process both the original data and the payload data are passed through the encoding function. The payload signal and the original host signal now together occupy space, which was previously occupied only by the host signal. For this purpose either the original data is compressed or redundancy in digital content is explored to make space for the payload. 1.3.3.4 Watermark Extraction: Extraction is achieved in two steps. First the watermark or payload is extracted in the decoding process and then the authenticity is established in the comparing process. 1.3.3.5 Decoding Process: The decoding process can be itself performed in two different ways. In one process the presence of the original unwatermarked data is required and other where blind decoding is possible. Fig.1.6 and Fig.1.7 show the two processes. A decoder function takes the test data (the test data can be a watermarked or un-watermarked and possibly corrupted) whose ownership is to be determined and recovers the payload.

Fig1.6 Simple Decoding Process 1.3.3.6 Comparison Process: The extracted payload is compared with the original payload (i.e. the payload that was initially embedded) by a comparator function and a binary output decision is generated. The comparator is basically a correlator. Depending on the comparator output it can be determined if the data is authentic or not. If the comparator output is greater than equal to a threshold then the data is authentic else it is not authentic. fig.1.8 illustrates the comparing function. In this process the extracted payload and the original payload are passed through a comparator. The comparator output C is the compared with a threshold and a binary output decision generated. It is 1 if there is a match i.e. C >= and 0 otherwise. A watermark is detectable or extractable to be useful . Depending on the way the watermark is inserted and depending on the nature of the watermarking algorithm, the method used can involve very distinct approaches. In some watermarking schemes, a watermark can be extracted in its exact form, a procedure we call watermark extraction. In other cases, we can detect only whether a specific given watermarking signal is present in an image, a procedure we call watermark detection. It should be noted that watermark extraction can prove ownership whereas watermark detection can only verify ownership .

Fig.1.7 Comparing Process 1.3.4. Requirements of Watermarking Digital watermarks can be measured on the basis of certain characteristics and properties that depend on the type of vapplication. These characteristics and properties include the difficulties of notice, the survival of common distortions and resistance to malicious attacks, the capacity of bit information, the coexistence with other watermarks, and the complexity of the watermarking method. In general, they are described as fidelity, robustness, fragility, tamper resistance, data payload, complexity, and other restrictions. Digital watermarks must fulfil the following requirements. Robustness It may not be possible without knowledge of the watermark algorithm or secret key to remove the watermark or to make it illegible. Robustness means the resistance ability of the watermark against the watermark attacks or modifications made to the original file. After modifications, resizing, file compression, rotation, and common operations, the watermark can still be detected and demonstrate a good quality. Non-perceptibility It means that the brought bit sample of the watermark does not produce perceptible changes acoustically or optically. A perfect non-perceptible bit sample is present if data material marked with watermark and the original cannot be distinguished from each other. Non-detectable It is always true that brought watermark information in data material is non-detectable if it is consistent with the origin data. Undeletable The watermarking must be hard to remove or even unable to remove by any attackers. Complexity Complexity describes the cost to detect and encode the watermark information. One of measurement technique could be the amount of time. It is a good design to make watermarking algorithm and procedure as complex as possible.

Capacity Capacity refers to the amount of information that can be stored in a data source. Unambiguous The extracted watermark is equivalent to the embedded watermark. A trade-off has to be taken between the above-mentioned criteria for an optimal watermarking application.

Fig 1.8. Primary requirements of Watermarking Algorithms

1.3.5 Applications of watermarking

1.3.5.1 Copyright Protection: This is by far the most prominent application of watermarks. With tons of images being exchanged over insecure networks every day, copyright protection becomes a very important issue. Watermarking an image will prevent redistribution of copyrighted images.

1.3.5.2 Authentication: Sometimes the ownership of the contents has to be verified. This can be done by embedding a watermark and providing the owner with a private key which gives him an access to the message. ID cards, ATM cards, credit cards are all examples of documents which require authentication.

1.3.5.3 Broadcast Monitoring: As the name suggests broadcast monitoring is used to verify the programs broadcasted on TV or radio. It especially helps the advertising companies to see if their advertisements appeared for the right duration or not.

1.3.5.4 Content Labeling: Watermarks can be used to give more information about the cover object. This process is named content labeling.

1.3.5.5 Tamper Detection: Fragile watermarks can be used to detect tampering in an image. If the fragile watermark is degraded in any way then we can say that the image or document in question has been tampered.

1.3.5.6

Digital Fingerprinting: This is a process used to detect the owner of the content. Every fingerprint will be

unique to the owner.

1.3.5.7 Content protection: In this process the content stamped with a visible watermark that is very difficult to remove so that it can be publicly and freely distributed .

1.3.6 Attacks on watermarks

Fig 1.9 Watermark attacks

A watermarked image is likely to be subjected to certain manipulations, some intentional such as compression and transmission noise and some intentional such as cropping, filtering, etc. They are summarized below : Lossy Compression: Many compression schemes like JPEG and MPEG can potentially degrade the datas quality through irretrievable loss of data. Geometric Distortions: Geometric distortions are specific to images and videos and include such operations as rotation, translation, scaling and cropping. Common Signal Processing Operations: They include the followings. Resampling, Requantization, Dithering distortion, Recompression, Linear filtering such as high pass and low pass filtering, Non-linear filtering such as median filtering, Color reduction, Addition of a constant offset to the pixel values, Addition of Gaussian and Non Gaussian noise, Local exchange of pixels Printing and Rescanning Watermarking of watermarked image (rewatermarking) Collusion: A Number of authorized recipients of the image should not be able to come together (collude) and like the differently watermarked copies to generate an un-watermarked copy of the image (by averaging all the watermarked images). Forgery: A Number of authorized recipients of the image should not be able to collude to form a copy of watermarked image with the valid embedded watermark of a person not in the group with an intention of framing a 3rd party.

10

IBM attack: It should not be possible to produce a fake original that also performs as well as the original and also results in the extraction of the watermark as claimed by the holder of the fake original.

1.3.7 Types of Watermarking

Fig 1.10 Types of watermarking Invisible Watermarks The watermark is embedded into the image in such a way that it cannot be perceived by human eye. It is used to protect the image authentication and prevent it from being copied. Invisible watermark can be further divided into three types: i. Robust Watermarks

Invisible watermark cannot be manipulated without disturbing the host signal. This is by far the most important requirement of a watermark. There are various attacks, unintentional (cropping, compression, scaling) and unintentional attacks which are aimed at destroying the watermark. So, the embedded watermark should be such that it is invariant to various such attacks. They are designed to resist any manipulations that may be encountered. All applications where security is the main issue use robust watermarks. ii. Fragile Watermarks

They are designed with very low robustness. They are used to check the integrity of objects. A watermark is said to be fragile if the information hidden within the host data is lost or irremediably altered as soon as any modification is applied to

11

the host signal. Such a loss of information may be global, i.e. no part of watermarking can be recovered, or local i.e. only part of the watermark is damaged. The main application of fragile watermarking is data authentication, where watermark loss or alteration is taken as evidence that the data has been tampered with. The recovery of the information content within the data demonstrates authentic un-tampered data. Robustness against signal distortion is better achieved if the watermark is placed in perceptually significant parts of the signal. This is particularly evident in the case of lossy compression algorithms, which operate by discarding perceptually insignificant data. Watermarks hidden within perceptually insignificant data are likely not to survive compression. Achieving watermark robustness, and, to a major extent, watermark security is one of the main challenges watermarking researches are facing with. iii. Semi-fragile Watermarks

Watermark is semi-fragile if it survives a limited well specified, set of manipulations, leaving the quality of the host document virtually intact. In some applications robustness is not a major requirement, mainly because the host signal is not intended to undergo any manipulations, but a very limited number of minor modifications such as moderate lossy compressions, or quality enhancement. This is the case of data labeling for improved actual retrieval, in which the hidden data is only needed to retrieve the host data from archive, and thereby it can be discarded once the data has been correctly assessed. Usually data is archived in compressed format, and that the watermark is embedded prior to compression. In this case the watermark needs to be robust against lossy coding.

On the basis of method of extraction of watermark, watermarking algorithms can be classified as: Non-Blind (Private) Use the original signal/image to extract the embedded Watermark. Semi-Blind (Semi Private) Dont use the original signal, use side information and/or original watermark for extraction of watermark. Blind (Public or oblivious) Dont use original signal or side information for extraction of watermark.

1.3.8 Watermark embedding Techniques a. b. c. Spatial Domain Frequency Domain (Transform Domain) Contourlet Domain

1.3.8.1 Spatial Domain techniques These methods based on direct modification of the values of the image pixels, so the watermark has to be embedded in this way. Such methods are simple and computationally efficient, because they modify the color, luminance or brightness values of a digital image pixels, therefore their application is done very easily, and requires minimal computational power. Spatial domain processes are expressed as G(x,y) = T [F(x,y)]

12

G (x,y) : processed image F (x,y) : input image T : operator on F 1.3.8.2 Frequency Domain techniques These methods are based on the using of some invertible transformations like discrete cosine transform (DCT), discrete fourier transform (DFT), discrete wavelet transform (DWT) etc. to the host image. Embedding of watermar is made by modifications of the transform coefiicients, accordingly to the watermark or its spectrum. 1.3.8.3 Contourlet Domain Techniques In the contourlet transform (CT), the Laplacian pyramid (LP) decomposes an image into a low-frequency (LF) subband and a high-frequency (HF) subband. The LF subband is created by filtering the original image with 2-D low-pass filter. However, the HF subband is created by subtracting the synthesized LF subband from the original image but not by 2-D high-pass filtering the original image. A contourlet-based image adaptive watermarking (CIAW) scheme, in which the watermark is embedded into the contourlet coefficients of the largest detail subbands of the image. The transform structure of the LP makes the embedded watermark spread out into all subbands likely in which the LF subbands are included when we reconstruct the watermarked image based on the watermarked contourlet coefficients. Since both the LF subbands and the HF subbands contain watermarking components, our watermarking scheme is expected to be robust against both the LF image processing and the HF image processing attacks.

13

Chapter 2 Bio-Medical Images Watermarking

14

2.1 INTRODUCTION
Digital information management in hospitals, HIS (Hospital Information System), and its special cases of RIS (Radiology Information System), PACS (Picture Archiving and Communication System) forms the information infrastructure of modern health care. Recently the advent of multimedia has boosted the potential of telemedicine applications ranging from teleconsulting, telediagnosis etc. to cooperative working session and telesurgery. These advances in information and communication technology provide in fact new ways to store, access and distribute medical data, and introduces new practices for the profession, as well as the patient themsleves by accessing to their own medical files. With these benefits there are concomitant risks for electronic patient records (EPR) and strictly personal documents circulating in open networks, and being accessible, e.g., via Internet. Thus it is a widely shared point of view that there is an urgent need for network level security measures and protocols in medical information.

Fig 2.1 Example of Medical image tampering

It is becoming easier and easier to tamper with digital image in ways that are difficult to detect. For example Fig2.1 shows two nearly identical images using readily available software (eg. Adobe Photoshop) the cyst was removed from the image by using the healing brush tool. It is difficult if not possible to tell which picture is the original and which has been tampered with. If this image were a critical piece of evidence in a legal case or police investigation, this form of tampering might pose a serious problem.

2.2 SECURITY AND MEDICAL INFORMATION

Medical information record of a patient is a complex of clinical examinations, diagnosis annotations, prescriptions, histological and other findings, and images in various modalities. In the digital format they are centered in the EPR (Electronic Patient Record). This information is gathered over years by a number of health professionals and used as well for different purposes (patient care but also clinical research, epidemiological studies or insurance companies). All patient records, electronic or not, linked to the medical secrecy, must be confidential. The digital handling of EPR on network requires a systematic content validation which is aimed at quality control: actuality (precise interest of the information at a given instant) and reliability (authentication of the origin and integrity). Security of medical information, derived from strict

15

ethics and legislatives rules, gives rights to the patient and duties to the health professionals. This imposes three mandatory characteristics: confidentiality, reliability, and availability. Confidentiality means that only the entitled users, in the normally scheduled conditions, have access to the information. Reliability has two aspects; i) Integrity: the information has not been modified by non-authorized people, and ii) Authentication: a proof that the information indeed belongs to the correct patient and is issued from the correct source. Availability is the ability of an information system to be used by the entitled users in the normal scheduled conditions of access and exercise.

Security

Confidentiality

Reliability

Availibility

Firewall

Encryption

Software Accreditation

Access Control

Integrity

Authentication

Firewall

Access Control

Antivirus

Digital Signature

Fig 2.2 Security Tools Security tools also have their limits. Regarding the information system access, firewalls provide a certain level of isolation between the intra-net and Internet, but are easily bypassed by hackers. For storage and transmission, cryptography is probably a very efficient tool, but once the sensitive data is decrypted, the information is not protected anymore. Furthermore the file headers are in the plain-text format and can be usurped by a pirate. Cipher text, on the other hand, unless protected by error correction facility, is very sensitive to bit errors occurring during storage and transmission. Once the images are in the open (plain-text form), the major threat is the violation of the access rights and of the daily logs by the intruder. Breaking of the confidentiality implies that integrity and authenticity cannot be guaranteed. Finally, a surprisingly large proportion of authenticity problems are not due to any intrusion, but due to errors in the manual entry of patient data. Watermarking is made to introduce identifiers which, by construction, are inseparable from the document they are attached to. They may be seen as ultimate ramparts against usurpation and fabrication. It can be claimed that, in the medical domain, watermarking is an additional tool in the repertoire of security measures, specifically adapted to images, which can be used to thwart certain attacks. The number of studies in the literature dedicated to watermarking is not very expensive. One of the earliest techniques involved insertion of an encrypted version of the EPR in the LSB of the gray levels of medical-image levels. Although the damage to the diagnostic image quality is minimal, the limitations and fragility of LSB watermarking can be catastrophic. Another similar LSB technique was developed in which the image carrier authenticates the origin of the

16

transmission (hospital), and the message to be embedded is composed of an ECG record, the diagnosis report and the doctors seals. Attention is given to the trusted header by watermarking the root part in the image data.

2.3 WATERMARKING MEDICAL APPLICATIONS

To address the specific application needs of watermarking in the context of biomedical images, the three main objectives foreseen are: 1. Data Hiding for the purpose of inserting metadata, annotations, and other information that just makes the image more useful or easier to use. 2. Integrity Control, which is the verification that the image is intact, in that it has not been modified in an unauthorized manner. 3. Authenticity, which is the verification that the image really is what the user supposes it is.

Within the medical domain there are two extremities that may be experimented and a large variety of intermediary cases which may be more or less interpolated from these cases. At one extremity, one family of applications covers the transmission of medical documents over public networks, like in telemedicine, remote or collaborative telediagnosis or telesurgery, distance learning and several applications dealing with database consulting. In this case the demand is very close from the desiderata of image watermarking as expressed for e-commerce or multimedia applications over open networks. The images will face transmission errors and lossy compressions; the protocols will probably be heterogeneous; secure modalities (firewalls, accredited software) will be rare and the end-terminals will not be secured. Under these circumstances we find watermarking solutions very similar to the many developed in other domains of image communication, the basic constraints of medical images being taken as additional guidelines for selecting the watermarking method. On the other extremity, within the hospital network and under the complete security system, very different problems will be faced. Transmissions are done most of the time without loss and specific workstations with adequate protocols and software may be available for handling local security problems. Under these conditions, security problems may only arise from either malicious attempts to break the security protocol, or human negligence or mistakes. The usual constraints of watermarking are: 1. 2. 3. 4. Invisibility of the mark Capacity (expressed in bit per host pixel) Secrecy to unauthorized persons Robustness to attempts to suppress the mark

These demands also exist in the medical domain but additional constraints are added.

17

2.4 CASE STUDIES ON WATERMARKING FOR BIOMEDICAL IMAGES

2.4.1 Authentication and Tracing Medical images may go through several services and receive different processing and annotations. These transformations are recorded in a historical resume that is attached to the image as metadata along with the patient references and the acquisition data. If a watermark is introduced in the image carrying an identifier also present in the resume, it guarantees that no error of falsification has been committed in the simultaneous processing of the image and the resume. The automatic handling of the watermark and the resume by authorized workstations is an additional security for the documents and may allow for instance the practice of reversible watermarking. If non reversible watermarking is used, incremental marks may hold track of the different services that handled the document. 2.4.2 EPR Diffusion If the EPR is kept by the patient, or distributed on several sites, and transmitted to the different services in charge of further considerations or treatments, some mechanisms should be introduced to guarantee the integrity of the document. These mechanisms not only will tell the medical persons whether the document is the same as the original one, but also if some differences exist, their location and importance. The decision could then be taken to consider the document is valuable or not. As the checking of the watermark may be heavy, it is probably not likely that it is done every time the image is displayed. The resume file attached to the image or the image header will be the prime identifier, and the control of the adequacy between the image and the resume or header be made when necessary for instance before entering the documents in a database or before a diagnosis or when a conflict happens. In this scenario the watermark is the ultimate security proof most of the time ignored by the system, which is only used when important security issues arrive.

18

2.5 REQUIREMENTS FOR MEDICAL IMAGE WATERMARKING

2.5.1 Reversible Watermarking Medical tradition is very strict with the quality of biomedical images, in that it is often not allowed to alter in any way the bit field representing the image. Thus the watermark must be reversible, in that the original pixel values must be exactly recovered. This limits significantly the capacity and the number of possible methods. It also constrains to have dedicated routines to automatically suppress and introduce the mark in order to prevent the transmission of unprotected documents.

2.5.2 Defining Regions of Interest and Regions of Insertion The watermark protects the regions of interest while being inserted in the rest of the image plane. One could be more tolerant in the regions of non interest as they do not contribute to the diagnosis. For example to increase capacity and robustness, one can allow the watermarking signal to be somewhat perceptible, provided its level does not disturb the radiologist. It has been shown that judicious alterations such as those occurring in image compression do not interfere with the diagnosis ability. Therefore in time, the attitude demanding strict preservation of the images as a number field will be relaxed. Thus watermark insertion methods that use the whole image, while bringing out imperceptible alterations in the pixels will creep into the medical field as well. 2.5.3 Integrity Control There is a need to prove that the images, on which the diagnoses or any insurance claims are based, have preserved their integrity. One must define the start point of integrity, as the original captured image often must undergo certain processing, like enhancement and contrast stretching, to be more useful to the radiologist. Thus it must be decided which version of the image, whether the pristine sensor output or the processed and standardized image at a certain stage by the radiologist, is taken as the reference for integrity control. The integrity control based on the exact preservation of all the bit planes of the image may be unnecessarily strict. Thus alternatives, specifically content-based integrity control are still open to discussion. 2.5.4 Authentication A critical requirement in patient records is to authenticate the different parts of the EPR, in particular the images. More often the image is identified by an attached file or a header that carries all the needed information (e.g. the DICOM solution to radiology images). However keeping the meta-data of the image in a separate header file is prone to forgeries or clumsy practices. An alternative would be to embed all such information into the image data itself. Another possible scheme is to have both the DICOM header in a separate file and embed the digest of the same information in the image. An important issue here is that how much information can be embedded. Medical data is more demanding in quality but less prone to degradations, as compared to multimedia content. Hence tens of bits per Megabits of data is achievable within the medical constraints.

19

Another variant application is MRI where the watermarking must satisfy the critical requirement that any arbitrary 2-D slice extracted from this volume, even with the unknown slicing angle must provide sufficient authentication evidence on the patient. 2.5.5 Dual Watermarking Scheme We use a dual watermarking scheme to enhance confidentiality and authentication. We focus on two types of watermark hiding. In caption watermarking, by hiding patients information in ROI, both authentication and confidentiality are achieved and gives a permanent link between the patient and medical data. In signature watermarking, we hide the patients digital signature or identification code in RONI for the purpose of origin authentication. 2.5.6 Adaptive watermarking To achieve better performance in terms of perceptually, invisibility and robustness, we use adaptive quantization parameters for data hiding. Because the energy distribution is an important characteristic for digital image processing, we use a model that employs this parameter for determining the adaptive quantization parameter. The embedding strength is more or less proportional to the value of energy to have better robustness and transparency in this method.

20

Chapter 3 Related studies

21

3.1 GENERAL WATERMARKING SYSTEM

First, a host image (original image) directly embeds watermark in spatial domain or is transformed into frequency domain through the well-known spread spectrum approach, i.e. DFT (Discrete Fourier Transform), DCT (Discrete Cosine Transform) or DWT (Discrete Wavelet Transform). However, the algorithms using transform domain offer more robust than directly embedding watermark into spatial domain. Then, coefficients are passed through a perceptual analysis block that determines how strong of the watermark in embedding algorithm so that the resulting watermarked image is imperceptible. The secret key uses to generate watermark and watermark embedding location more. The watermark is embedded using a specific well-designed algorithm based on mathematical or statistical model. If the coefficients in frequency domain, the inverse spread spectrum approach is then adopted to obtain a watermarked image. The watermark extraction applies the similar operations in embedding processes. It employs the inverse operations or uses the mathematical or statistical characteristic to extract the embedded watermark. Watermark detection decides whether an image has been watermarked and the watermark exists or not.

Fig 3.1 Embedding and extraction of watermark 3.2 WATERMARKING IN SPATIAL DOMAIN
3.2.1 Insertion of Watermark A block based spatial domain algorithm is used to hide copyright mark (invisible logo) in the homogenous regions of the cover image exploiting average brightness. 3.2.1.1 The cover image is partitioned into non-overlapping squares of size (8X8) pixels. A block is denoted by the location of its starting pixel (x,y). If the cover image is of size (NXN), total (N/8,N/8) number of such blocks are obtained for watermark insertion. Next, all such blocks are arranged in ascending order based on their variance values. The variance of a block of size (mXn) is denoted by:

22

Is the statistical average value of the block. The blocks having small variance values may be called as homogenous blocks and, of course, the smallness in variance values depends on the characteristics of image to be watermarked. If the watermark symbol is a (NXN) binary image, only N2 homogenous blocks are sufficient to insert one watermark pixel in each homogenous block. A two level map of size (N/8 X N/8) is constructed based on the location of homogenous blocks in the cover image assigning each homogeneous block of the cover image by value 1 while all other blocks by value 0.Thistwo level map later modified as multi level image, also called as secret image (s), is used for extraction of watermark pixels. 3.2.1.2 One watermark pixel is inserted in each homogenous block. Before insertion, the binary watermark is spatially dispersed using a chaotic system calledtorus automorphism. Basically, the torus automorphism is a kind of image independent permutation done by using pseudo random number of suitable length. This pseudorandom number is generated using Linear Feedback Shift Register. The pseudo random number in the present case is of length 256 and the spatially dispersed watermark data thus obtained is denoted by L1. 3.2.1.3 From the two level image formed in step 2, desired blocks of the cover image are selected and statistical average value of these blocks are used for watermark insertion. Let for one such block this average value and its integer part are denoted by A and Ai respectively. Now one pixel from L1 replaces a particular bit (preferably Least Significant Bit planes) in bit plane representation of A for each homogenous block. The selection of particular bit in bit plane representation may be determined based on the characteristics (busyness /smoothness of regions) of the block. The bit plane selection is also governed by global characteristics of the cover image besides the local property of candidate block, such as mean gray value. For a block of low variance (homogenous zone) higher bit plane may be chosen provided that the mean gray level value of the block is either less than T2 or greater than T1 where T1 and T2 are certain pre-specified threshold values which should preferably be close to 0 (minimum) and close to 255 (maximum). However, the closeness of T1 and T2 to 0 and255 respectively, is relative, and is strongly image dependent. Users may choose the value of T1 and T2 and also the proper bit plane by checking the degradation in the image quality affected by the insertion of the logo. A multilevel secret image is constructed by inserting the value of bit position selected for different homogenous block located in the 1 position of the secret image. This positional information as gray value of the secret image helps to extract watermark pixel from the proper bit position of the mean gray value of the block. 3.2.1.4: The choice of lower order MSB plane (say 3rd or higher from the bottom plane) may result in more robust watermarking at the cost of greater visual distortion of the cover image. Further bit manipulation is done to minimize this

23

aberration and to counter the effect of smoothing that may cause possible loss of embedded information. The process effectively changes those mean gray values of the blocks that have been used in watermark insertion. Implementation is done by estimating the tendency of possible change in mean gray value after the attack like mean filtering. Larger size of spatial mask such as (7X7) is used to adjust suitably the gray values of all pixels of the block. The use of spatial mask reduces visual distortion on and average fifty percent times.

3.2.2 Extraction of Watermark

The extraction of watermark requires the secret image(s) and the key (k) used for spatial dispersion of the watermark image. The watermarked image under inspection with or without external attacks is partitioned into non-overlapping block of size 8x8 pixels. Now from the secret image, position of the homogenous blocks are selected and gray value of the secret image indicates the corresponding bit position in mean gray values where watermark pixel was inserted. Hence from the secret image the mean gray value of the blocks of the watermarked image/distorted watermarked image is calculated and watermark pixel is extracted. The spatially dispersed watermark image thus obtained is once again permuted using the same key (k) (pseudo random number) and watermark in original form is thus obtained. This completes watermark extraction process .A quantitative estimation for the quality of extracted watermark image with reference to the original watermark may be expressed as normalized cross correlation

Maximum value of NCC is unity.

24

3.3 WATERMARKING IN TRANSFORM DOMAIN

3.3.1 Discrete Cosine Transform (DCT) The DCT makes a spectral analysis of the signal and orders the spectral regions from high to low energy. It can be applied globally or in blocks. When applied globally, the transform is applied to all parts of the image, separating the spectral regions according to their energy. When applied in blocks, the process is analogous, only the transform is applied to each block separately. The typical algorithm steps 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; 4) Apply coefficient selection criteria; 5) Embed watermark by modifying the selected coefficients 6) Apply inverse DCT transform on each block. The formulae for DCT transform and inverse DCT transform are given as follows:

The human eyes are more sensitive to noise in lower-frequency band than higher frequency. The energy of natural image is concentrated in the lower frequency range. The watermark hidden in the higher frequency band might be discarded after a lossy compression. Therefore, the watermark is always embedded in the lower-band range of the host image that transformed by DCT is perfect selection.

25

Fig 3.2 DCT Implementation Flowchart

3.3.2 Discrete Wavelet Transform (DWT) The DWT can be implemented as a multistage transformation. An image is decomposed into four sub bands denoted LL, LH, HL, and HH at level 1 in the DWT domain, where LH, HL, and HH represent the finest scale wavelet coefficients and LL stands for the coarse-level coefficients. The LL sub band can further be decomposed to obtain another level of decomposition. The decomposition process continues on the LL sub band until the desired number of levels determined by the application is reached. Since human eyes are much more sensitive to the low frequency part (the LL sub band), the watermark can be embedded in the other three sub bands to maintain better image quality.

Fig 3.3 DWT Implementation

26

WATERMARK EMBEDDING ALGORITHM Step 1. For a given image, apply the discrete wavelet Transform up to 3 rd level, which produce a total of 9 bands

wavelet coefficients. Step 2. The middle frequency sub-bands HL and LH from 2nd level of wavelet decomposition are used to get 3rd level decomposition. Step 3. The watermark of size MM is converted into binary pattern. Step 4. The binary image is scaled to the size of original host image and then duplicated. Step 5. Pseudo random sequence is generated using a secret key and combined with the duplicated watermark to increase robustness. Step 6. The resultant watermark is then embedded into the middle frequency sub-bands of host image. Step 7. Finally, inverse DWT is performed to produce the watermarked image.

Fig 3.4 DWT hybrid watermark embedding(using SVD) WATERMARK EXTRACTION ALGORITHM Step 1.For a given watermarked image, apply the discrete wavelet Transform up to 3 rd level, which produce a total of 9 bands of wavelet coefficients. Step 2. The middle frequency sub-bands HL and LH from 2nd level of wavelet decomposition are used to get 3rd level decomposition. Step 3. Same secret key used in the embedding process enables to generate the random sequence. Step 4. Finally, the watermark is extracted from the selected wavelet coefficients. Step 5. After extracting the final watermark, compares it with the original watermark, to find the any attacks happened in the original data.

27

Fig 3.5 DWT watermark extraction block schematic The DWT technique provides better imperceptibility and higher robustness against attacks, at the cost of the DWT compared to DCT schemes. Each watermark bit is embedded in various frequency bands and the information of the watermark bit is spread throughout large spatial regions. As a result, the watermarking technique is robust to attacks in both frequency and time domains. . However, improvements in its performance can still be obtained by viewing the image watermarking problem as an optimization problem.

The DWT technique provides better imperceptibility and higher robustness against attacks, at the cost of the DWT compared to DCT schemes. Each watermark bit is embedded in various frequency bands and the information of the watermark bit is spread throughout large spatial regions. As a result, the watermarking technique is robust to attacks in both frequency and time domains. . However, improvements in its performance can still be obtained by viewing the image watermarking problem as an optimization problem. 3.3.3 SINGULAR VALUE DECOMPOSITION (SVD) SVD is one of a number of effective numerical analysis tools used to analyze matrices. In SVD transformation, a matrix can be decomposed into three matrices that are the same size as the original matrix. Given a real n n matrix A, this matrix can be transformed into three components, U, D and V, respectively, such that

where the U and V components are n x n real unitary matrices with small singular values, and the D component is an n x n diagonal matrix with larger singular value entries. A is the reconstructed matrix after the inverse SVD transformation.

28

Using SVD in digital image processing has some advantages: i. ii. iii. iv. The size of the matrices from SVD transformation is not fixed and can be a square or a rectangle. Singular values in a digital image are less affected if general image processing is performed. Singular values contain intrinsic algebraic image properties.

SVD Based Watermarking

In 2002, Sun et al. proposed an SVD and quantization-based watermarking scheme (Sun et al., 2002). The D component with a diagonal matrix was explored.

In the embedding procedure, the largest coefficients in D component were modified and used to embed a watermark. The modification was determined by the quantization mechanism. After that, the inverse of the SVD transformation was performed to reconstruct the watermarked image. Because the largest coefficients in the D component can resist general image processing, the embedded watermark was not greatly affected. Also, the quality of the watermarked image can be determined by the quantization. Thus, the quality of the watermarked image quality can be maintained.

To extract an embedded watermark, the SVD transformation was employed and the largest coefficients in the D component were examined. After that, the watermark was extracted.

The watermark embedding and extracting procedures can be described as follows:

The watermark embedding procedure: Step 1. Partition the host image into blocks. Step 2. Perform SVD transformation. Step 3. Extract the largest coefficient D(1, 1) from each D component and quantize it by using a predefined quantization coefficient Q. Let Z = D(1, 1)modQ. Step 4. For an embedded watermark bit valued of 0, if Z < 3Q/4, D(1, 1) modify to D(1, 1) = D(1, 1) + Q/4 - Z. Otherwise, D(1, 1) = D(1, 1) + 5Q/4 - Z. Step 5. For an embedded watermark bit valued of 1, if Z < Q/4, D(1, 1) modify to D(1, 1) = D(1, 1) - Q/4 + Z. Otherwise, D(1, 1) = D(1, 1) + 3Q/4 - Z. Step 6. Perform the inverse of the SVD transformation to reconstruct the Watermarked image.

29

Host image Block Partition

SVD Transformation

Detect D component

Quantize D component

Modify D component according to value of watermark bit value

Inverse SVD Transformation

Watermarked image

Fig 3.6 SVD based watermark Embedding Block Diagram

The watermark extracting procedure: Step 1. Partition the watermarked image into blocks. Step 2. Step 3. Perform SVD transformation. Extract the largest coefficient D(1, 1) from each D component and quantize it by using the predefined quantization coefficient Q. Let Z = D(1, 1)modQ. Step 4. If Z < Q/2, the extracted watermark has a bit value of 0. Otherwise, the extracted watermark has a bit value of 1.
Watermarked image

Block Partition

SVD Transformation

Detect D component

Quantize D component

Examine D component

Extracted watermark

Fig 3.7 SVD based watermark Extracting Block Diagram

30

3.4 WATERMARKING IN CONTOURLET DOMAIN

The contourlet transform, is a relatively new image decomposition scheme, which provides a flexible multiresolution representation for 2D signals. It makes use of the Laplacian pyramid decomposition (LPD) for the multiresolution representation of the image. In the contourlet transform, the Laplacian pyramid decomposes an image into a low frequency subband into a high frequency subband. After this, a directional decomposition is performed on every band-pass image using directional filter banks (DFB). The contourlet transform is unequaled since the number of directional bands could be indicated by the user at any resolution. Finally, the image is represented as a set of directional subbands at multiple scales. Discrete contourlet transform is able to capture the directional edges and contours superior to DWT. Even though other transform domains only conform to grayscale images, this domain can conform to DICOM images as well, which is our area of work. Before the embedding process the following preprocessing steps are carried out: 1. For each row of the image, the left and right edges of the image are recorded, similarly for each column of the image, the top and bottom edges of the image are recorded too. For an image of dimensions MxN, the left and right edges of the image form two vectors L and R of size M, and the upper and lower edge of the image construct two vectors T and B of size N. For each vector we select l = min(L), r = max(R), t = min(T) and b = max(B), and then we define a rectangle of which the left corner has coordinates (t,l) and the bottom right one is (b,r). 2. 3. Watermark is reshaped to binary vector (W={w1,w2,w3,.,wk},wk E [0,1]). In view of the robustness, we choose IL, lowpass subband of decomposed IO, for embedding and W is embedded into I L in contourlet domain. For more invisibility the embed process can be done in the detail subbands.

EMBEDDING PROCESS
IL is divided into non-overlapping blocks Ai of size b x b, i=1,2,.,M, where M is the number of the blocks. The energy value of each block Ai is computer according to:

For each block Ai, the adaptive quantization step value i is computed as follows.

Where o is the basic quantization step that is different in ROI and RONI and served as a secret key, and the function floor represents the round off operation. Using singular value decomposition (SVD), the singular value vectors of each block A are computed. By the singular values of each block Ai, Nsi= ||S|| + 1 is computed (where |||| represents the Euclidean norm) and quantized by adaptive quantization step di that represents the quantization level as follows:

31

Finally using the value

, the modified singular values are computed as follows:

Using these modified singular values, the watermarked block is obtained.

EXTRACTION PROCESS
For watermark extraction, we require only the size of the binary vector (W), and basic quantization step (o). The watermarked image is converted into the contourlet domain and the lowpass subband I L is selected for extraction. At first IL is divided into non-overlapped blocks Ai of size b X b,i=1,2,M where M is the size of the blocks. Then is computed and quantized by adaptive quantization step that is computed similar to embedding process.

32

Fig3.8 Contourlet Domain embedding and extraction block schematic 3.5 TEST STANDARDS 3.5.1 Peak Signal to Noise Ratio (PSNR)

Peak Signal to Noise Ratio (PSNR) is used frequently as an objective image quality metric, but it does not consider characteristics of the human visual system (HVS). It is poor at comparing different watermarking methods, but provides a simple indicator for quantifying the similarity between original and watermarked images. PSNR uses peak power of the original image and the mean squared value of the error signal. PSNR is expressed as follows:

3.5.2 Structured Similarity Measure (SSIM) The second measure used in this paper is structural similarity measure (SSIM) index, which is a region-based numerical metric that places more emphasis on the HVS than PSNR. This metric is ideal for testing the similarities in medical images because It focuses on local rather than global image similarity. Mathematically, for regions, it is expressed as:

SSIM compares the similarity in luminance (LC), contrast (CC), and structure (SC) of image regions for each pair of corresponding blocks. a, b,and l are = 1and are used to weight the importance of each of the three components. 3.5.3 Luminance Comparison (LC) Luminance comparison is a function of corresponding blocks mean intensity and is given by

33

Where Io and Io are the mean values of regions respectively and c1 is a constant. 3.5.4 Contrast Comparison

Contrast comparison is a function of corresponding blocks standard deviation and is expressed as

3.5.5 Structured Comparison The structured comparison is given as the correlation coefficient and is given by:

Where CIoIo is the correlation coefficient between the two regions and c3 is constant. 3.5.6 Bit Error Rate (BER) We use BER to calculate the difference between the recovered EPR data and the original EPR data.

where wi and wi are the original and recovered EPR vectors respectively. In the lack of adverse attacks, BER is fo und to be zero. 3.5.7 Normalized Cross Correlation

It is also calculated to quantitatively analyze the likeness of the extracted watermark and the original watermark (logo) in signature watermark. Where V(i,j) and V(i,j) are the original and extracted logos and M1, M2 are the size of the logo image

34

Chapter 4 Design and Implementation

35

4.1 IMPLEMENTATION TOOLS AND PRE-REQUISITES

We are using MATLAB version 2008b as our platform for working on our project. The image processing toolbox provided inbuilt into the software provides a wide variety of tools to exploit. We can perform image enhancement, image deblurring, feature detection, noise reduction, image segmentation, geometric transformations, and image registration. Many toolbox functions are multithreaded to take advantage of multicore and multiprocessor computers. Image Processing Toolbox supports a diverse set of image types, including high dynamic range, gigapixel resolution, embedded ICC profile, and tomographic. Graphical tools let us explore an image, examine a region of pixels, adjust the contrast, create contours or histograms, and manipulate regions of interest (ROIs). With toolbox algorithms we can restore degraded images, detect and measure features, analyze shapes and textures, and adjust color balance. Besides the toolbox, we also required a rich repository of DICOM images and other grayscale biomedical images to work with. The same is maintained at personal systems for convenience (over 50 images). Test Images:

11.tif

1.tif

9.tif

juet_logo.tif

36

4.2 ALGORITHMS IMPLEMENTED

4.2.1 BASIC DWT ALGORITHM 4.2.1.1 EMBEDDING

1. Read Cover Image. 2. Read Watermark Logo.

3. 4. 5. 6. 7. Apply 1-level DWT to cover image and watermark obtained in step 1 & 2 to obtain LL,LH,HL and HH subbands. Take alpha value (non adaptive method) Applyling the alpha value for each subband separately Apply inverse DWT to obtain final watermarked image. Check for imperceptibility of the final watermarked image obtained above against the original image by evaluating the PSNR values between them for various values of embedding strength (alpha)

4.2.1.2 EXTRACTION 1. 2. Apply DWT to watermarked image to obtain extraction subbands Obtain subband of watermark by using the extraction subband obtained above and the value of alpha ascertained previously. 3. 4. Apply IDWT on above components to obtain extracted watermark. Check for consistency of extracted watermark by applying evaluation parameters (PSNR,NCC) against it.

Fig 4.1 Implementation of pure DWT algorithm (non-adaptive)

37

ALPHA 0.1 0.2 0.3 0.4 0.5 0.6

PSNR 105.9469 92.4836 86.7155 83.7426 81.7426 80.8158

Table 4.1 Variation of PSNR with embedding strength(non adaptive) 4.2.2 DWT DUAL AND ADAPTIVE ALGORITHM 4.2.2.1 EMBEDDING IN REGION OF INTEREST (ROI) 1. 2. 3. Read the cover image and watermark logo. Select n points from the current axes and return x and y coordinates in the column vectors x and y. Obtain the binary mask BW for the ROI by setting the pixels inside the region of interest to 1 and those outside the region to 0. 4. Extract the region of interest from the original image using the mask obtained above, by copying intensities into the region specified by the user. 5. 6. 7. 8. Apply DWT on the original and the watermark image to obtain subbands for both. Obtain the embedding strength(alpha) using the mean from all values in the ROI. Apply the watermark to every subband separately including LH,HL and HH subband. Apply inverse DWT to obtain the watermarked image.

4.2.2.2 EMBEDDING IN REGION OF NON INTEREST (RONI) 1. 2. 3. 4. 5. 6. Read the watermark image. Apply 1-level DWT to both cover and watermark images to obtain subbands for both. Obtain the embedding strength using the standard deviation of all values in the RONI Apply the watermark to every subband separately in LL,LH,HL and HH subband. Apply 1-level inverse DWT to obtain the watermarked image for RONI region. To obtain the final watermarked image, containing watermarks of different quality in ROI and RONI, we replace the intensity values of this said image with intensities of watermarked images of ROI and RONI at their respective positions. 7. We check the robustness and imperceptibility qualities of the images obtained using NCC and PSNR parameters.

38

4.2.2.3 EXTRACTION FROM ROI 1. 2. Using the matrix obtained from algorithm 4.2.2.1, extract the subbands of the image using 1-level DWT Using the value of alpha obtained in algorithm 4.2.2.1, obtain the subbands of the watermark image embedded in ROI region. 3. Obtain extracted ROI watermark by applying 1-level inverse DWT to the subbands obtained above.

4.2.2.4 EXTRACTION FROM RONI 1. 2. Using the matrix obtained from algorithm 4.2.2.2 extract the subbands of the image using 1-level DWT Using the value of alpha obtained in algorithm 4.2.2.2 obtain the subbands of the watermarked image embedded in RONI region. 3. Obtain extracted RONI watermark by applying 1-level inverse DWT to the subbands obtained above.

Fig 4.2 Region of interest and non-interest, before, and after watermarking procedure

39

4.3

ATTACKS TO TEST ROBUSTNESS AND IMPERCEPTIBILITY QUALITIES

A total of five attacks (and one compression test) have been used to determine the robustness and imperceptibility qualities of our proposed scheme. After applying each attack, the corresponding values of NCC and PSNR are recorded and compared with the original DWT scheme, to check for improvements / degradations. 4.3.1 TABLE SHOWING COMPARISION OF PSNR VALUES OF ORIGINAL AND PROPOSED SCHEME We obtain the adaptive values of alpha from algorithms 4.2.1.1 and 4.2.1.2, and use the value of alpha obtained from RONI to obtain PSNR values for the original scheme. This way, we have a precise comparison of values where the images used and embedding strength employed is same in both the schemes. ATTACK Salt and Pepper ALPHA VALUES 0.0953 0.3381 0.0964 0.3411 0.0616 0.3356 0.0473 0.3527 0.0448 0.3540 0.1018 0.3399 0.0578 0.3518 0.0746 0.3477 0.0482 0.3534 PSNR (ORIGINAL) 92.7187 PSNR(PROPOSED) 96.4738

Motion Blurring

92.7127

96.0344

Standard Blur

92.4576

95.0513

Gaussian Blur

92.5213

94.8108

Sharpening

92.4837

94.7878

Compression (75%)

59.5353

96.5642

Compression (50%)

59.1245

94.8507

Compression (25%)

58.7738

94.1959

Compression (15%)

58.6858

93.8947

40

4.3.2 TABLE SHOWING COMPARISON OF NCC VALUES OF ORIGINAL AND PROPOSED SCHEME We applied each attack separately, and then recorded NCC values for both the extracted ROI watermark as well as the RONI watermark. We have applied each attack on the watermarked image, and then extracted the watermarks in ROI and RONI from this attacked image. This way, we can evaluate the effect of each attack on the extracted watermark, as compared to the original scheme. ATTACK ALPHA VALUES (ROI AND RONI) NCC (ORIGINAL) FIGURES (ROI AND RONI EXTRACTED WATERMARKS) NCC (PROPOSED)

0.0953

0.7274

Salt and Pepper

0.5927

0.3381

0.9128

0.0964

0.9128

Motion Blurring

0.6153

0.3411

0.9512

41

0.0616

0.9113

Standard Blurring

0.6174

0.3356

0.9445

0.0473

0.9631

Gaussian Blur

0.9442

0.3527

0.9398

0.0448

0.8707

Sharpening

0.4170

0.3540

0.8956

42

0.1018

0.6878

Compression (75%)

1.000

0.3399

0.4112

0.0578

0.8106

Compression (50%)

1.000

0.3518

0.4205

0.0746

0.7693

Compression (25%)

1.000

0.3477

0.4837

43

0.0482

0.8381

Compression (15%)

1.000

0.3534

0.4277

44

Chapter 5 Conclusion

45

5.1 CONCLUSION Dual Adaptive Watermarking for Bio-medical images is a new watermarking method for DICOM images (various types) based on using different embedding strength for ROI and RONI in order to not affect the interpretation by medical specialists. The algorithm will use an automatically selection for ROI and embed the watermark in the singular values of contourlet subbands that makes the algorithm more efficient, and robust against noise attacks than other transform domains. The watermarked image can still conform to the DICOM format.

46

5.2 APPENDICES 5.2.1 APPENDIX A : ATTACKS DONE ON IMAGES A1: COMPRESSION ATTACK The most used image compression is definitely JPEG. In MATLAB, for compressing an image to different quality factors, the image should be created from a matrix and be reread: imwrite(wc_image,'extracted_DWT.jpg','Mode','lossy','Quality',75); A = imread (extracted_DWT.jpg); A2: SALT AND PEPPER ATTACK We can add a variety of noises into an image using the imnoise command in MATLAB. A = imread('Watermarked_DWT.tif'); A = imnoise(A,'salt & pepper',0.02); imshow(A); A3: MOTION BLURRING Motion blurring can be achieved through the following code that is inbuilt into MATLAB: h = fspecial('motion', len, theta) It returns a filter to approximate, once convolved with an image, the linear motion of a camera by len pixels, with an angle of theta degrees in a counterclockwise direction. The default value of len is 9 and theta is 0, meaning a horizontal motion of 9 pixels. H = fspecial('motion',20,45); MotionBlur = imfilter(wc_image,H,'replicate'); imwrite(MotionBlur,MotionBlur_DWT.tif,tif); imshow(MotionBlur)

A4: STANDARD BLURRING h = fspecial('disk', radius) It returns a circular averaging filter (pillbox) within the square matrix of side 2*radius+1. The default radius is 5. H = fspecial('disk',10); blurred = imfilter(wc_image,H,'replicate'); imwrite(blurred,Blurred_DWT.tif,tif); imshow(blurred) A5: GAUSSIAN BLURRING h = fspecial('gaussian', hsize, sigma) It returns a rotationally symmetric Gaussian lowpass filter of size hsize with standard deviation sigma (positive). hsize can be a vector specifying the number of rows and columns in h, or it can be a scalar, in which case h is a square matrix.

47

A6: STANDARD SHARPENING h = fspecial('unsharp', alpha) It returns a 3-by-3 unsharp contrast enhancement filter. alpha controls the shape of the Laplacian and must be in the range 0.0 to 1.0. The default value for alpha is 0.2. Code: H = fspecial('unsharp',0.5); sharpened = imfilter(wc_image,H,'replicate'); imwrite(sharpened,sharpened_DWT.tif,tif); imshow(sharpened)

5.2.2 APPENDIX B : EXTRACTING REGION OF INTEREST AND NON-INTEREST The following MATLAB code demonstrates how to select a region of interest from an image. clc; clear all; close all; a=im2double(imread('11.tif')); imshow(a); [r,c]=ginput(4); BW=roipoly(a,r,c); [R C]=size(BW); %......Extracting ROI....................% for i= 1 : R for j = 1 : C if BW(i,j)==1 out1(i,j)=a(i,j); else out1(i,j)=0; end end end %............Extracting RONI..............% for i= 1 : R for j = 1 : C if BW(i,j)==1 out2(i,j)=0; else out2(i,j)=a(i,j); end end end %......Printing original image, ROI and RONI......% subplot(1,3,1), image(a), title('Original image'); subplot(1,3,2), image(im2uint8(out1)), title('ROI'); subplot(1,3,3), image(im2uint8(out2)), title('RONI');

48

Fig 5.1 Region of Interest and non-Interest

49

REFERENCES 1. A Survey of Digital Image Watermarking Techniques, Vidyasagar M. Potdar, Song Han, Elizabeth Chang, School of Information Systems, Curtin University of Technology, Perth, Western Australia, 2005 3rd IEEE International Conference on Industrial Informatics (INDIN) 2. A Benchmark for Medical Image Watermarking, Navas K. A, Sasikumar M and Sreevidya S, Electronics and Communication Engineering Department, College of Engineering, Trivandrum, Kerala, India, 695016 3. A hybrid watermarking scheme using contourlet transform and singular value decomposition, C.Venkata Narasimhulu1 & K.Satya Prasad, IJCSNS International Journal of Computer Science and Network Security, VOL.10 No.9, September 2010 4. Relevance of Watermarking in Medical Imaging, G. Coatrieux(1,2), H.Maitre(1), B. Sankur (3), Y. Rolland(2), R. Collorec(2), (1) ENST, Department TSI- Paris, France; (2) LTSI, INSERM, Universitede Rennes I, France; (3) Bogazici University, Bebek, Istanbul, Turkey. 5. A Review Study on Image Digital Watermarking, Charles Way Hun Fung, Antonio Gortan, Walter Godoy Junior, ICN 2011 : The Tenth International Conference on Networks 6. A dual adaptive watermarking scheme in contourlet domain for DICOM images, Farhad Rahimi1 and Hossein Rabbani1,2*, Rahimi and Rabbani BioMedical Engineering OnLine 2011, 17 June 2011 7. Robust image watermarking in the spatial domain, N. Nikolaidis!, I. Pitas",*, Signal Processing 66 (1998) 385 403, Received 22 February 1997; received in revised form 23 September 1997 8. Combined DWT-DCT Digital Image Watermarking, Ali Al -Haj, Department of Computer Engineering, School of Electrical Engineering, Princess Sumaya University for Technology, PO Box 1928, Al-Jubeiha, Journal of Computer Science 3 (9): 740-746, 2007, ISSN 1549-3636 2007 Science Publications 9. Robust feature based watermarking process, International Journal of Computer Applications (0975 8887Volume 4 No.5, July 2010,N.R.Nantha Priya PG Student Anna University Tirunelveli Tirunelveli, India S.Lenty Stuwart Assistant professor Anna University Tirunelveli Tirunelveli, India 10. An Efficient Non-blind Watermarking Scheme for Color Images using Discrete Wavelet Transformation,
International Journal of Computer Applications (0975 8887) Volume 2 No.3, May 2010, Nagaraj V. Dharwadkar Computer

Science & Engineering National Institute of Technology Warangal, India. B. B. Amberker Computer Science & Engineering, National Institute of Technology Warangal, India. 11. A robust watermarking scheme using self-reference image, Jiang-Lung Liu, Der-Chyuan Lou *, Ming-Chang Chang, Hao-Kuan Tso Department of Electrical Engineering, Chung Cheng Institute of Technology, National Defense University, Tahsi, Taoyuan 33509, Taiwan Received 7 April 2005; received in revised form 1 July 2005; accepted 1 July 2005 Available online 25 July 2005 12. Watermarking Techniques Spatial Domain Digital Rights Seminar Mahmoud El-Gayyar Instructor Prof. Dr. Joachim von zur Gathen Media Informatics University of Bonn Germany May 06 13. STUDY OF THE EFFECTED GENETIC WATERMARKING ROBUSTNESS UNDER DCT AND DWT DOMAINS International Journal on New Computer Architectures and Their Applications (IJNCAA) 2(2): 353-360 The Society of Digital Information and Wireless Communications, 2012 (ISSN: 2220-9085)

50

14. Dr.M.A.Dorairangaswamy, A Novel Invisible and Blind Watermarking Scheme For Copyright Protectionof Digital
Images, IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.4, April 2009

15. Shikha Tripathi1, Nishanth Ramesh2, Bernito A3, Neeraj K J, A DWT based Dual Image Watermarking
Technique for Authenticity and Watermark Protection, Signal & Image Processing : An International Journal(SIPIJ) Vol.1, No.2, December 2010

16. Zhang Guannan, Wang Shuxun and Nian Guijun, (2004) A Blind Watermarking Algorithm Based onDWT Color Image, Intl. Symposium on Multi-Dimensional Mobile Communications, Vol. 2,pp.634- 638.

51

BIBLIOGRAPHY 1. 2. 3. 4. 5. 6. http://www.unisourcelink.com www.elsevier.com/locate/csi http://ieeexplore.ieee.org http://link.springer.com http://www.scientific.net http://www.academia.edu/2694741/Study_of_the_effect_DCT_and_DWT_domains_on_the_imperceptibility_and_ robustness_of_Genetic_watermarking I. J. Cox, M.L. Miller and J.A. Bloom. Digital Watermarking. Morgan Kaufmann, San Francisco, USA, 2002. http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=4457752 http://link.springer.com/chapter/10.1007%2F978-90-481-2311-7_13#

7. 8. 9.

52

PERSONAL DETAILS NAME MANJARI TYAGI ENROLMENT NUMBER 091237 PHOTOGRAPH

PALLAVI JAIN

091310

TAPAS TRIVEDI

091324

53