ABSTRACT

With the advances of technology, security of the digital media has become a major challenge. One way to ensure safety and control over the unauthorized distribution of the media is to effectively hide some classified information in it so that the intellectual property rights of the owner can be proved as and when required. Watermarking is the technique to embed some data into the signal for identifying the copyright ownership and preventing piracy. By pre-processing the image watermark and hiding it in a carrier stream, a robust watermarking technique is presented, resisting all the signal processing manipulations and deliberate attacks. In digital watermarking, the signal may be audio, pictures, or video. If the signal is copied, then the information also is carried in the copy. A signal may carry several different watermarks at the same time. Digital watermarking is a technology that opens a new door for authors, producers, publishers and service providers for protection of their rights and interest in multimedia documents.

CHAPTER 1 Introduction
With the rapid growth of Internet, it is easier for digital data owners to transit multimedia files across the Internet. Thus, there is a huge increase in concentration overcopyright protection of media. Traditionally, encryption and control access techniques were used to protect the intellectual property rights. These techniques do not protect against unauthorized copying after the media have been successfully transmitted and decrypted. Cryptography guarantees confidentiality, authenticity, and integrity when a message is transmitted across the network. It does not protect against unauthorized copying after successful message transmission. Watermarking is an efficient way to protect copyright of media. Watermarking is a technique of embedding a special pattern, watermark, into a multimedia document so that a given piece of copyright information is permanently tied to the data. This information can later prove the ownership, identify a misappropriating person, trace the marked documents dissemination through the network, or simply inform users about the rights-holder or the permitted use of the data.

Motivation The technological advances have led to the multiple copying of digital media without incurring a loss in quality. These unlimited copying causes a considerable loss to the copyright holders. To enforce intellectual property rights and to restrain the unauthorized distribution of the media, some information is hidden in it for proving the ownership. Hence, the main motivation for taking watermarking as a research topic is to develop an effiective algorithm for embedding and extraction of the hidden information (watermark) for data files. The goal in this project is to embed the watermark in the audio signal imperceptibly by the exploitation of the statistical and the objective properties of the cover signal. The low frequency region and the energy of the signal segments are considered in this regard. Lastly, the robustness of the implemented strategy is evaluated against signal processing and intentional attacks which mainly include the desynchronization attacks and this defines the final aim of the undertaken project.

1.1 Objective
1. To develop an effective algorithm for embedding and extraction of the hidden information (watermark) for files. 2. To embed watermark imperceptibly by exploiting the statistical and the objective properties. 3. To evaluate robustness of the implemented strategy against de-synchronization attacks.

1.2 Digital Watermarking

With the advancements in technology and increased access to the digital media, the copyright protection of media has gained a lot of importance. To prevent piracy of media and its unauthorized distribution, the owner or the creator of the media needs to prove his intellectual property rights. This is where watermarking comes into picture. Watermarking is a strategy to embed some classified information into the media perceptually or non-perceptually to prove the ownership. Watermark is the data which is hidden into the digital media for proving the intellectual property rights. The main properties of the watermark include: 1. Imperceptibility 2. Undetectability 3. Resistance to all signal manipulations 4. Extractability to prove ownership 5. Unambiguity The watermark should be imperceptible in the manner that the changes made to cover object for embedding the watermark should not create a very high order of distortion. That is the quality of the cover object should remain unaffected despite the changes done in it for hiding the watermark thereby, making the watermark imperceptible. The watermark should also be undetectable when searched by any malicious user with an aim to destroy or remove it. Another property that the watermark needs to exhibit is its resilience to the signal manipulations and other deliberate attacks. It should be so embedded in the cover work that it is resistant to all changes made in the cover work. Lastly, the watermark should be extractable to some acceptable extents to prove the copyright ownership whenever required to do so. These attributes of the watermarks have to be taken into account before developing the algorithm for its embedding and extraction. Also, the watermark should be unambiguous so that no two parties may claim for the same

media work. Watermark, when extracted, should be able to prove the copyright ownership most efficiently.

1.3 Classification of Watermarking Techniques

The watermarking techniques used to embed information are dependent on the requirements of the applications that the cover work is to be used for. So it is designed according to the type of cover media, resilience to attacks and environment in which it is used. The taxonomy is shown in figure 1. The division of the watermarking strategies has been done in following main streams on the basis of: 1. Cover Media 2. Domain 3. Extraction strategy 4. Resistance to attacks

Figure 1.1: Taxonomy of Watermarking Techniques On the basis of cover media used: The watermarking algorithms are broadly classified into 5 categories depending on the cover work that is to be watermarked. These are: 1. Text 2. Image 3. Audio 4. Video 5. Software

In text watermarking algorithms, spacing between the letter, words and lines or the typing style is manipulated in a manner to embed the watermark. The changes reflect the embedding of 5

watermark in the document. Image watermarking scheme has been the most popular research area in watermarking. Exhaustive research has been carried out in this direction. There are two domains in which the watermarking is done namely: 1. Spatialdomain 2. frequency domain In spatial domain, changes are made to the pixel values of the image in order to embed the watermark which can be another image or any other signal. In frequency domain, transform is applied to the image and then coefficients are changed to embed the watermark depending on the frequency region in which the watermark needs to be hidden. This embedding of watermark is dependent on the Human Visual System (HVS). Audio watermarking is another domain of work where the watermark is embedded in the audio signal to curb the unlimited piracy of the media. Audio watermarking is comparatively difficult to implement due to the sensitivity of the Human Auditory System. The modifications are made to amplitude values of the samples or to the transform coefficients when any transform is applied to the audio signal. Video watermarking involves the embedding to be done in the continuous image frames of the video or to the audio part of the signal or to both. Since, human auditory system is more sensitive to changes in amplitude so the watermark embedding is carried out in the pixel values of image frames. Software watermarking is done in order to discourage software piracy. Watermark can be embedded in the code part or in the data of the software. Watermark can be some classified textual information which when extracted can prove the copyright ownership. On the basis of domain: The watermarking algorithms can be broadly classified into two categories: 1. Time/ spatial domain 2. Frequency/ Transform domain In spatial domain, the watermark is embedded by directly modifying the values of cover object. For example, in images the pixel values can be modified and in audio the amplitude values can be manipulated. In transform domain, a transform is applied to the cover object and then manipulations are done to the transform coefficients in the required frequency band. The main transforms in common use are: Discrete Cosine Transform (DCT), Discrete Fourier Transform (DFT), Discrete Wavelet Transform (DWT) etc. 6

On the basis of extraction strategy: This classification is done on the basis of the fact whether the original cover object or some extra information is required for watermark extraction. It is divided into three broad categories: 1. Non blind techniques 2. Semi blind techniques 3. Blind techniques In the non blind strategy the original cover work is required to extract the watermark. The main disadvantage of this method is that extra bandwidth is required to send the original cover work as well along with the watermarked work. In semi blind strategy, some information derived from the original cover work, not exactly the original media is required for extracting the watermark. This information can be any data derived from the media file. In the blind strategy, no extra information is required for the extraction of the watermark. In comparison to blind strategy, non blind watermarking algorithms are more resistant to attacks On the basis of resistance to attacks: In this category of classification, watermarking scheme is again classified into three types: 1. Fragile 2. Semi fragile 3. Robust The watermarking strategies come under the category of being fragile when they cannot sustain the signal manipulation and intentional attacks. Semi fragile techniques can sustain some attacks but they also fail against some of them. Robust strategies are those which are resistant to all the deliberate and non deliberate attacks. They are capable of sustaining most of the attacks. That is, the watermark can be successfully extracted or detected even if the cover work is corrupted with attacks.

1.4 Digital watermarking life-cycle phases

Figure 1.2 : Digital watermarking life-cycle phases General digital watermark life-cycle phases with embedding-, attacking-, and detection and retrieval functions The information to be embedded in a signal is called a digital watermark, although in some contexts the phrase digital watermark means the difference between the watermarked signal and the cover signal. The signal where the watermark is to be embedded is called the host signal. A watermarking system is usually divided into three distinct steps, embedding, attack, and detection. In embedding, an algorithm accepts the host and the data to be embedded, and produces a watermarked signal. Then the watermarked digital signal is transmitted or stored, usually transmitted to another person. If this person makes a modification, this is called an attack. While the modification may not be malicious, the term attack arises from copyright protection application, where pirates attempt to remove the digital watermark through modification. There are many possible modifications, for example, lossy compression of the data (in which resolution is diminished), cropping an image or video, or intentionally adding noise. Detection (often called extraction) is an algorithm which is applied to the attacked signal to attempt to extract the watermark from it. If the signal was unmodified during transmission, then the watermark still is present and it may be extracted. In robust digital watermarking applications, the extraction algorithm should be able to produce the watermark correctly, even if the modifications were strong. In fragile digital watermarking, the extraction algorithm should fail if any change is made to the signal. 8

1.5 Purpose of Digital Watermarking

Watermarks added to digital content serve a variety of purposes. The following list details six purposes of digital watermarking (Memon & Wong, 1998). Ownership Assertion to establish ownership of the content (i.e. image) Fingerprinting to avoid unauthorized duplication and distribution of publicly available multimedia content Authentication and integrity verification the authenticator is inseparably bound to the content whereby the author has a unique key associated with the content and can verify integrity of that content by extracting the watermark Content labeling bits embedded into the data that gives further information about the content such as a graphic image with time and place information Usage control added to limit the number of copies created whereas the watermarks are modified by the hardware and at some point would not create any more copies (i.e. DVD) Content protection content stamped with a visible watermark that is very difficult to remove so that it can be publicly and freely distributed Unfortunately, there is not an universal watermarking technique to satisfy all of these purposes (Memon & Wong, 1998). The content in the environment that it will be used determines the digital watermarking technique. The following section describes some digital watermarking techniques.

1.6 Properties
For better activeness, watermark should be perceptually invisible within host media, statistically invisible to unauthorized removal, readily extracted by owner of image, robust to accidental and intended signal distortion like filtering, compression, resampling, retouching, crapping etc. For a digital watermark to be effective for ownership, it must be robust, recoverable from a document, should provide the original information embedded reliably and also removed by authorized users. All these important properties of digital watermarks are described as1. Robustness :The watermark should be robust such that it must be difficult to remove. The watermark should be robust to different attacks. The robustness describes whether watermark can be reliably detected after performing some media operations. 2. Perceptual transparency:This property describes that whether watermark is visible or 9

invisible to human sensor organ. Perceptible watermarks are visible to human while imperceptible are not. Imperceptible watermarks are such that content remains same after applying digital watermarking technique. 3. Security: Security property describes that how easy to remove a watermark. This is generally referred to as attack on watermarking. Attack refers to detection or modification of watermark. 4. Complexity: This is important property which is to be consider in real time applications like video. Complexity property is concerned with amount of effort needed to extract or retrieve the watermark from content. 5. Capacity : Capacity property of digital watermarks refers to amount of information that can be embedded within the content. The important point is that more data is used in watermark, watermark will become less robust. In addition to these properties, watermarks are having some extra properties as unambiguity, tamper resistance, inseparable from the works and able to undergo some transformation as works.

1.7 Why to use it?

First important application come into mind is copyright protection of digital media.It is easy to duplicate digital data exactly without quality loss. Similar to process in which artist signed their painting with a brush to claim their copyrights, artist of today can watermark their work and hide some information say their name in the image. Hence, embedded watermark will allow to identify the owner of work. This concept is applicable to digital video and audio also. Especially, distribution of digital audio over internet in MP3 format is currently a big problem. Digital watermarking may be useful to setup controlled audio distribution and provide efficient means for copyright protection, usually in collaboration with international registration bodies such as IDDN(Inter Deposite Digital Number). In addition with copyright protection, Digital watermarking is playing a important role in many fields of applications such as broadcast monitoring, owner identification, transaction tracking, proof of ownership, fingerprinting, content authentification, copy control, device control. Digital watermarks can also serve as invisible labels and content link. For example, photo development labs may insert a watermark into the picture to link the print to its negative.so, it is becomes easy to find out negative of a print. All one has to do is to scan the print and extract the information from negative. In a 10

completely different scenario, the digital watermarks may be used as a geometrical reference which may be useful for programs such as Optical Character Recognition(OCR) software. The embedded calibration watermark may improve the detection reliability of the OCR software since it allows the determination of translation, rotation and scaling. Digital watermarking also serves as a means of advertising within the digital media. For instance, the user may download and view a digital image, use a watermark reader to extract the digital signature, then access a web based directory to find the companys name and up-to-date address, phone number and web and e-mail address. Digital watermarks also serves the purposes of identifying quality and assuring authenticity. A graphic or audio file bearing digital watermark can inform the viewer or listener who owns to the item. The technique Digital Watermarking is the recent research in the field of multimedia and internet copyright protection field. There are various applications of digital watermarking as broadcast monitoring, owner identification, proof of ownership, transaction tracking,content authentication, copy control, device control and so on. Out of these, some important applications are described as-

1.Broadcast monitoring This application identifies that when and where works are broadcast by recognizing watermarks embedded in these works. There are variety of technologies to monitor playback of sound recording on broadcast. The digital watermarking is alternative to these technologies due to its reliable automated detection. A single PC based monitoring station can continuously monitor to 16 channels over 24 hours with no human interaction. Resulted monitoring is assembled at central server and is now available to interested one .The system can distinguish between identical versions of songs, which are watermarked for different distribution channel. Such system requires Monitoring infrastructure and watermarks to be present in content.Watermarking video or music is planned by all major entertainment componies possessing closed networks.

2. Encoding According to the thinking of major music companies and major video studios, encoding happens at mastering level of sound recording. In such downstream, transactional watermarks are also considered. Each song is assigned with unique ID from the identifier database.After completion 11

of all mastering processes, ID is encoded in sound recording.To enhance encoding of audio or video recordings requiring special processing, the human-assisted watermark key is available.

3. Copy and playback control The data carried out by watermark may contain information about copy and display permissions. We can add a secure module into copy or playback equipment to automatically extract the permission information and block further processing if required.This approach is being taken in Digital Video Disc(DVD).

4. Content authentication The content authentication is nothing but embedding the signal information in Content. This signature then can be checked to verify that it has not been alter. By watermarks, digital signatures can be embedded into the work and any modification to the work can be detected.

12

CHAPTER 2 Image Watermarking

A watermark is A pattern of bits inserted into a digital image file that identifies the file's copyright information (author, rights, etc.). The name watermark is derived from the faintly visible marks imprinted on organisational stationery. Unlike printed watermarks, which are intended to be somewhat visible (like the very light compass stamp watermarking this report), digital watermarks are designed to be completely invisible.

copies watermark Alice Send to Bob

Image 1

Watermark Image
Figure2.1: Image watermarking

The purpose of digital watermarks is to provide copyright protection for intellectual property that is in digital format. As seen above, Alice creates an original image and watermarks it before passing it to Bob. If Bob claims the image and sells copies to other people Alice can extract her watermark from the image proving her copyright to it.

13

The caveat here is that Alice will only be able to prove her copyright of the image if Bob hasnt managed to modify the image such that the watermark is damaged enough to be undetectable or added his own watermark such that it is impossible to discover which watermark was embedded first.

2.1 Technical Details

Digital watermarking technology makes use of the fact that the human eye has only a limited ability to observe differences. Minor modifications in the colour values of an image are subconsciously corrected by the eye, so that the observer does not notice any difference. While vendors of digital watermarking schemes do not publicly release the exact methods used to create their watermarks, they do admit to using the following basic procedure (with obvious variations and additions by each vendor). A secret key (string or integer) produces a random number which determines the particular pixels, which will be protected by the watermarking. The watermark is embedded redundantly over the whole image, so that every part of the image is protected.

Figure2.2 : Working

14

One way of doing this is by Patchwork. This technique uses a random number generator to select n pairs of pixels and slightly increases or decrease their luminosity (brightness level). Thus the contrast of this set is increased without any change in the average luminosity of the image. With suitable parameters, Patchwork even survives compression using JPEG. Although the amount of secret information has no direct impact on the visual fidelity of the image or the robustness of the watermark, it plays an important role in the security of the system. The key space, that is the range of all possible values of the secret information, x must be large enough to make exhaustive search attacks impossible. In the process of extracting the watermark, the secret key is used to identify the manipulated pixels and finally to decode the watermark. As an example of poor engineering, an early version of Digimarcs watermarking software gave each licensed user an ID and a two-digitnumeric password, which were issued when she registers with Digimarc and pays for a subscription. The password checking mechanism could easily be removed by flipping a particular flag bit and the passwords had only 99 possibilities so it was short enough to be found by trial and error. A deeper examination of the image also allowed a villain to change the ID and thus the copyright of an already marked image as well as the type of use (such as adult -> general public content). Before embedding a mark, watermarking software usually checks whether there is already a mark in the picture, but this check can be bypassed fairly easily with the result that it is possible to overwrite any existing mark and replace it with another one. The quality of digital watermarks can be judged in two ways; firstly it must be able to resist intentional and unintentional attacks and secondly the embedded watermark must not detract from the quality of the image. The higher the resistance of a watermark against attacks, the higher the risk of the quality of the image being reduced, and the greater the chance of obvious visual artefacts being created.

15

2.2 Methods used to test Watermark Robustness

These are some of the methods that can be used to test whether a watermark can survive different changes to the image it is embedded in.

Compare this Original Image with the attacked images below, and see if you can spot any changes in quality. Horizontal Flipping

Many images can be flipped horizontally without losing quality. Few watermarks survive flipping, although resilience to flipping is easy to implement.

16

Rotation & Cropping A small rotation with cropping doesnt reduce image quality, but can make watermarks undetectable as rotation realigns horizontal features of an image used to check for the presence of a watermark.

The example has been rotated 3 degrees to the right, and then had its edges cropped to make the sides straight again.

JPEG Compression/Re-compression JPEG is a widely used compression algorithms for images and any watermarking system should be resilient to some degree to compression or change of compression level e.g. from 71% to 70% in quality like the example .

Scaling Uniform scaling increases/decreases an image by the same % rate in the horizontal and vertical directions.

Non-uniform scaling like the example at left increases/decreases the image horizontally and vertically at different % rates. Digital watermarking methods are often resilient only to uniform scaling. 17

Dithering Dithering approximates colors not in the current palette by alternating two available similar colors from pixel to pixel. If done correctly this method can completely obliterate a watermark, however it can make an image appear to be patchy when the image is over-dithered (as in the

elbow area of the image).

Mosaic A mosaic attack doesnt damage the watermarked image or make it lose quality in any way, but still enables the image to be viewed in eg: a web browser by chopping the image into subsections of equal size and putting it back together again.

To the viewer a mosaic image appears to look the same as the original but a web crawler like DigiMarcs MarcSpider sees many separate images and doesnt detect that these separate images are parts of a watermarked image. This means that the watermark cannot be detected, as a problem common to all image watermarking schemes is that they have trouble embedding watermarks into small images, (less than 256 pixels in height or width). Stirmark StirMark is the industry standard software used by researchers to automatically attempt to remove watermarks created by Digimarc, SysCoP, JK_PGS (TALISMAN project .P.F.L. algorithm), Signum Technologies and EIKONAmark.

18

Stirmark attacks a given watermarked image using all the techniques mentioned in this report as well as more esoteric techniques such as low pass filtering, gamma correction, sharpening/unsharpening etc. All vendors of digital watermarks have their products benchmarked by Stirmark and as of August 2001, no watermark from any vendor survives the test, ie: the watermarks are all removed without degradation to image quality occurring.

19

Chapter 3 Audio Watermarking

With the development of the internet and increase in the need for the security and integrity of the data in transit on the network, watermarking has gained a great interest. With this, the protection of intellectual property rights has become one of the areas of main concern. The main work in the field of watermarking has been focused on images and videos. Only a few audio watermarking schemes have been reported till date. Digital audio watermarking is the scheme of embedding some relevant information in the audio signal so as to prove the copy right ownership. Audio watermarking is difficult to implement as compared to image watermarking due to high sensitivity of human auditory system (HAS). The general strategy to implement audio watemarking is diagrammatically given as:

Figure 3.1 : Audio Watermarking: General Strategy The simplest visualization of requirements of the audio watermarking algorithms is the magic triangle. Its vertices are inaudibility, robustness to attacks and watermark data rate. This triangle is a perfect visual representation of the trade-offs between the watermark capacity and robustness, keeping the perceptual quality of the watermarked audio at an acceptable level. It is certainly not possible to attain high robustness and high watermark data rate simultaneously. Hence, to achieve robustness against attacks,

20

Figure 3.2 : Magic Triangle capacity of the watermark has to be compromised. The requirements that have to be satisfied among all the three are very much application dependent. As for an instance in steganography applications, the algorithms have to attain robustness to attacks.

3.1 Human Auditory System

The Human Auditory System is very sensitive to slightest changes in the audio due to its wider dynamic range. Hence watermarking of audio signals is more challenging as compared to watermarking of images or video. The HAS perceives sounds over a range of power greater than 109:1 and a range of frequencies greater than103:1. The HAS is highly sensitive towards Additive White Gaussian Noise (AWGN). The slight perturbations in a sound file are detected even if they are as low as one part in ten million or 80dB below ambient level. This is because the human ear can perceive amplitude distortion but it is relatively insensitive to phase distortion. On the other hand, HAS has a fairly small differential range which means loud sounds can mask out the weaker sounds . 21

The human ear acts as a frequency analyser that maps signal frequencies to locations along the basilar membrane. The HAS is generally modelled as a non uniform filter bank with logarithmically widening bandwidth for higher frequencies. The bandwidth of each filter is set according to the critical band, which is defined as the bandwidth in which subjective response changes abruptly. Hence, HAS is modelled as a bandpass filterbank, containing strongly overlapping bandpass filters with bandwidths around 100 Hz for bands with a central frequency ranging between 500Hz to 5000 Hz for bands placed at high frequencies. The main properties of the HAS mainly used in watermarking algorithms are frequency (simultaneous) masking and temporal masking. Masking properties are exploited to embed additional bit stream into the cover audio signal without generating the audible noise thus keeping the watermark concealed.

3.2 Types of Audio Watermarking Techniques

Digital audio watermarking techniques can be broadly classiffied into 2 categories based on the domain in which the watermarking is done: 1. Time domain 2. Frequency domain

Figure 3.4 : Taxonomy of Audio Watermarking Techniques

22

In time domain audio watermarking schemes, the watermark is embedded by modifying the audio signal itself. In transform domain, modifications are done on the transform coefficients of the components of the audio signal. Time domain techniques can be further classified into two categories (a) when the audio signal samples are modified for embedding the watermarking (LSB Decoding method, Quantization method, phase watermarking method), (b) when inaudible noise (watermark) is added on to the audio signal (addition of pseudo random sequence, echo hiding method, patchwork method). The following sections will discuss various audio watermarking techniques and the affect of various audio signal attacks on the robustness of the watermark when embedded by these techniques.

3.2 Time Domain Techniques

When the watermark is directly embedded into the audio signal then they come under this category and they are broadly classified into two parts on the basis of how the watermark is embedded into the signal. 1. LSB Decoding Method The simplest and most straight forward technique of embedding the watermark is to embed it into the Least Significant Bits of the audio signal. Given the extraordinary high channel capacity of using the entire audio signal for transmission, the watermark can be embedded into it multiple number of times. The audio signal is first divided into segments and a subset of the segments is selected. The LSBs of these segments are modified according to the bit of the watermark that is to be embedded. Extraction of the watermark is performed by extracting the least significant bits of the selected segments. If the extracted bits and the inserted bits match then the watermark is successfully detected. It is a blind algorithm the original audio signal is not required for detection of the watermark. The advantages of this method are: (a) As it is simple to implement, its algorithmic delay is very less. (b) It has very high watermark capacity as stated earlier.

23

Disadvantages being: (a) It is extremely less robust due to the fact that random changes of the LSBs destroy the coded watermark. (b) It is quite unlikely that the watermark would survive the analog to digital conversion and vice versa.

2. Quantization Method This scheme divides an audio signal into samples and then quantizes a value for each sample. Then the quantized value of each sample is modified based on the watermark bit to be embedded. In the scalar quantization method a quantization step is decided on whose basis the quantized value of the sample is generated by using the quantization function. It is with the use of quantization step only that the value of the sample is modifiied with regard to the bit being inserted. The detection process is exactly the reverse of the insertion process. The main advantages of this method are: (a) It is very simple to implement (b) It is a blind algorithm. The disadvantage is: (a) It cannot survive the noise attacks if the additive noise is larger than the quantization step. This method can also be characterized as one which uses the information of the audio signal for embedding the watermark.

3. Addition of Pseudo Random Sequence In this scheme, the watermark is simply considered to be pseudo random sequence.This technique takes into account the characteristics of the psycho acoustic model such that added sequence does not cause any audible effect on the audio signal. Before embedding the watermark, it is shaped in such a manner that its addition to the signal is not heard. The shaping of watermark can be done in many ways, one of them is filtering. In this method, the audio signal is again divided into samples and the shaped pn sequence is added to each of the samples of the signal or some of the selected samples. The method of addition of the watermark can be additive, multiplicative etc. In the extraction process, the audio signal is again

24

divided into samples and the watermark is extracted from the selected ones or all of them by the procedure reverse of what was followed for embedding. Now the extracted and the original and the extracted watermark can be compared to check for the successful detection. The advantages of this algorithm are: (a) The watermark can be repeatedly embedded in the audio signal. (b) It is quite robust against attacks. The disadvantage being: (a) The transmission cost of the audio signal goes very high.

4. Echo hiding method Echo hiding embeds the data into the audio signal by introducing an echo in time domain. The nature of such echoes is to add resonance to the host audio.

x(n) = s(n) + a:s(n - d)

Here only 1 echo sound is being added but multiple echoes can also be added. Here in this method the host audio signal is divided into smaller portions. Each portion can be considered as an independent signal. Now embedding of the watermark into these independent signals takes place by introducing the bit delays. As for an instance, binary messages are added by echoing the original signal with 2 delays, either a d0 sample delay for bit 0 or a d1 sample delay for bit 1. Extraction of the embedded message involves the detection of the delay. Cepstrum or Autocepstrum analysis are used for these purposes. The advantages of Echo hiding method are: (a) It is quite imperceptible and usually makes the sound rich. (b) This is a blind algorithm and is highly robust against desynchronization attacks. The main disadvantages of this method are: (a) Its increased complexity due to the use of cepstrum analysis for detection. (b) As the human auditory system is quite sensitive, so echoes ca be very easily detected and they provide a clue for malicious attack.

25

5. Patchwork method This scheme embeds a special statistic into the audio signal. Two patches are pseudo randomly selected from the original audio signal. These two patches are nothing but the group of some samples each. In the simplest patchwork strategy,a constant value d is added to one patch and subtracted from another. Apart from the additive strategy, multiplicative operations can also be used for the implementation of the patchwork method in which a single bit of the watermark is embedded into the signal by multiplying or dividing the sample values of one patch thus, leaving the other patch intact. The watermark is extracted at the receiver end by the comparison of energies of the two patches so as to say the expected values of the samples of the two patches are compared. In case of the addition and subtraction operated patchwork scheme, the difference of the sample values of the patches is compared with the threshold value to decide whether the samples contain the watermark or not. The threshold value is decided with the constant value that is used while embedding the watermark . Its greatest advantage is: (a) Since the algorithm does not use the original audio signal for watermark extraction, it is a blind algorithm. (b) This scheme can also be implemented in transform domain [26] as well and in that case the modifications would be done on the transform coefficients of the samples in the patches . Disadvantage of this scheme is: (a) It is less robust against attacks if position of patches is known.

6. Spread Spectrum method The spread spectrum technique encrypts information by spreading the encrypted data across a large frequency spectrum. In this technique the watermark is first shaped and then it is spread over the entire audio signal by the use of various spread spectrum techniques. The spread spectrum technique can also be implemented in the transform domain. In Direct Sequence Spread Spectrum techniques, spreading is accomplished by modulating the original signal with a sequence of pseudo random binary pulses known as chips. In the embedding process, the audio data is coded as a binary string. The code is modulated by a carrier wave and multiplied by the pseudo-random noise sequence, having a wide frequency spectrum. It spreads the frequency 26

spectrum [29] of the data over the available frequency band. This spread data sequence is attenuated and added to the original file. For extraction the watermarked signal is again multiplied by the same pseudo-random noise sequence. The watermark is spread over so many components of the signal so that the energy of any component is very small and certainly undetectable. The advantages of this technique are: (a) It is a simple to implement. (b) It is a blind algorithm. Its greatest disadvantage is: (a) This method is highly fragile against de-synchronization attacks.

3.3 Frequency Domain Techniques

Audio signals are transformed from time domain to frequency domain to enable effective embedding of the watermark. Transform domain techniques allow the embedding of the watermark in the perceptually significant components of the audio signal. The transform domain techniques are more robust against attacks as compared to the time domain techniques but the drawback being the high computation requirement. Various time domain techniques can be incorporated with the transform domain techniques to make more robust methods of audio watermarking.

1. DCT based Watermarking Watermark message is embedded into the host audio by modifying the DCT coefficients, which can be regarded as addition of noise to the original audio signal. Audio signals is divided into frames and DCT is applied to them individually. Shaped watermark signal is then added to the DC or AC coe_cients of the transformed audio signal. Then inverse DCT is applied to the frames. The DCT transform divides the signal into three frequency regions namely , low mid and high. The watermark is fragile when embedded in the low or high frequency region so for robustness, it is embedded in the mid frequency range. The other reason for not embedding the watermark in the low frequency region is that it causes audible changes to the signal thereby, destroying its fidelity. In case of embedding the watermark in high frequency region, the

27

problem is of robustness only. The chances of the removal of the watermark are very high by simple signal processing operations. For extracting the watermark, DCT needs to be applied to the signal and then from the selected frames the watermark is extracted by following the procedure reverse to that used for embedding.

2. DFT based Watermarking The methodology to embed the watermark in the audio signal is same as DCT. The audio signal is divided into frames and DFT is applied on them individually. Some or all of the frames can be used for embedding the watermark. In case the watermark is to be embedded in some frames then one way for the selection of these frames is on the basis of a secret key. This key is used at the time of extraction also for taking out the selected frames. The application of DFT divides the signal into two parts of phase and magnitude. The watermark is embedded into any of these parts of the DFT coeffcients depending upon the requirement of the application. Prominently the embedding is done in magnitude part only. Perform the inverse DFT on the frames to reconstruct the audio signal. For extraction again the same DFT is applied to the signal and from the selected frames watermark is extracted. The advantages of this technique are: (a) It is highly robust against various signal processing attacks. (b) It is a blind algorithm. (c) This scheme is more robust to attacks than DCT based watermarking scheme. The disadvantage is: (a) The complexity of the procedure involved for watermarking and extraction is quite high.

3. DWT based Watermarking DWT based watermarking schemes are on the same pattern as DCT based schemes. The audio signal is transformed by wavelet transforms using wavelet filters. The common filters for watermarking are Daubechies Orthogonal Filters, Haar Wavelet Filter and Daubechies BiOrthogonal Filters. The signal is decomposed into several frequencies by these filters. On single

28

level decomposition, the signal is decomposed into two parts, high frequencies and low frequencies. The part containing low frequencies is again decomposed into two parts, high frequencies and low frequencies. The number of levels in which the signal can be decomposed depends upon application and length of signal. The decomposed data thus obtained gives DWT coefficients. The original signal can again be reconstructed by applying the inverse DWT scheme.

3.4 Methods used to test Watermark Robustness

th

The most common method of watermarking audio is to mark every x bit in an audio file depending on the random generator seed calculated from the watermarking key applied to the audio. These are some of the ways watermarks can be removed from audio files MPEG1 Layer III (MP3) audio compression A digital audio compression algorithm that achieves a compression factor of about twelve while preserving sound quality. What this lossy compression does is remove the frequencies not heard by the human ear from the audio. If a raw audio file is converted to MP3 at a bit-rate of 128kbps than roughly 90% of the frequencies are removed. This means that a search for the watermark needs to find an unaltered length of samples that contains at least 2 watermarked bits to prove the watermarks existence. Audio Restoration programs Audio restoration programs are designed to remove hisses, crackles and pops from audio recordings. They do this by searching through the wavelength, removing samples that dont fit in amongst neighbouring samples, and replacing them with an average of the two neighbour samples. Although the removal of digital watermarks is obviously not a purpose of these programs, they work remarkably well at doing so as the sample bits inserted to watermark the audio dont fit in with their surrounding pixels, and are therefore removed.

Echo Hiding Removal Echo hiding relies on the fact that we cannot perceive short echoes, eg: 1 millisecond(ms) and embeds data into a cover audio signal by introducing an echo characterised by its delay and its relative amplitude compared to surrounding samples. The echo delays are chosen between 0.5 ms and 2 ms and the best relative amplitude of the echo is around 0.8 ms. However specialised 29

software which looks for echoes with a length between 0.5 ms and 2 ms (as seen below), can be used to detect and remove these echoes without effecting sound quality.

Zitter The simplest and most effective attack on any audio watermarking scheme is to add jitter to the signal. In our first implementation, we split the signal into chunks of 500 samples, either duplicated or deleted a sample at random in each chunk (resulting in chunks of 499 or 501 samples long) and stuck the chunks back together. This turned out to be almost imperceptible after altering, even in classical music; but the jitter prevents the marked bits from being located, and therefore the watermark is obliterated. In his paper titled Audio watermarking: Features, Applications And Algorithms, Michael Arnold agrees with the Cambridge team stating that one of the greatest challenges [of watermarking] is the robustness against the so-called jitter attack.

30

CHAPTER 4 ADVANTAGES AND DISADVANTAGES

4.1 Advantages of digital watermarking The field of digital watermarking is not restricted to digital images. This technique time and frequency masking properties of human ear to conceal the watermark and make it unaudible. There are some benefits of this technique as1. Uniquely identifies author of copyrighted works. 2. Robust design of digital waetrmark withstand pirating attacks. 3.Embedding watermark is easy. 4. Implementation on PC platform is possible. There are some military and civilian applications of digital watermarking also. Military applications 1. Intelligence activities. 2.Traitor tracing. 3. Image tampering detection. Civilian applications 1. Copyright protection of each digital media in hardcopy or on internet. 2. Intelligent web browsers. 3. Law enforcement for chain of evidence. Some special watermarking technique uses color separation. So watermarks appears in only one of the color bands. Therefore watermarks becomes strongly invisible. Whenever the colors are separated from printing then watermarks becomes visible. This approach is advantageous to journalists to inspect digital pictures from a photo stockhouse before buying an un-watermarked versions. There is an important advantage of invisible fragile watermarks. With such invisible fragile watermarks, implementation of web based image authentication This web based authentication includes watermark embedding and authentication system. In case of watermark embedding system, it is installed in server as application software that any authorized user who access to 31

server can generate watermark image.The distribution can be done through any network as FTP, e-mail etc. Once image is distributed to externally, client can get authentication web page to get verification of images. Digital watermarking techniques is having advantages as we seen earlier.It also have some limits also.

4.2 Disadvantages of digital watermarking Digital watermarking is recent research field, therefore its intrinsic limitations are not understood yet. The blind watermarking algorithm which is really robust is not in existence today. Another disadvantage is that owner can erase the watermark. By knowing the exact content of watermark and algorithms to embeds and retrieve it. It is always possible to make it unreadable without any significant degradation of the data. Again it is not clear that. if this drawback will be got around in future, in the meantime ,the possibility of erasing the watermark or its part ,once its content is known must be taken into account when designing a copyright protection system. As a matter of fact, if anyone is allow to read the watermark, then anyone can erase it by knowing the embedding algorithm. In some researches, the conclusion comes that not all watermarking techniques will be useful in resolving ownership disputes. Watermarking does not prevent image copying as much as it simply makes copied images easier to track down and detect ownership. Some watermarks vanish if someone manipulates the image in a program like Photoshop. The watermarks have been known to weaken or disappear by the time the images were processed for the Internet. Resizing, compressing and converting images from one file type to another may add noise to an image or diminish its watermark in such a manner that the watermark becomes unreadable.

32

CHAPTER 5 FUTURE OF DIGITAL WATERMARKING

The field of digital watermarking is still evolving and is attracting a lot of research interest. The watermarking problem is inherently more difficult that the problem of encryption, since it is easier to execute a successful attack on a watermark. In cryptography, a successful attack often requires deciphering an enciphered message. In the case of digital watermarking, merely destroying the watermark, usually by slightly distorting the medium containing it, is a successful attack, even if one cannot decipher or detect any hidden message contained in the medium.The enormous popularity of the World Wide Web in the early 1990's demonstrated the commercial potential of offering multimedia resources through the digital networks. Since commercial interests seek to use the digital networks to offer digital media for profit, they have a strong interest in protecting their ownership rights. Digital watermarking has been proposed as one way to protect such interests. Though much research remains before watermarking systems become robust and widely available, there is much promise that they will contribute significantly to the protection of proprietary interests of electronic media. Collateral technology will also be necessary to automate the process of authentication, non-repudiable transmission and validation. An exhaustive list of watermarking applications is of course impossible. However, it is interesting to note the increasing interest in fragile watermarking technologies. Especially applications related to copy protection of bills with digital watermarks. Various companies have projects in this direction solutions will soon be available. In addition to technological developments, marketing and business issues are extremely important and require in-depth analysis and strategic planing. It is very important to prepare the industry to the usage of digital watermarks and and it is very likely that fully functioning to convince them of the added value their products can gain if they employ digital watermarking technologies.

33

CHAPTER 6 CONCLUSION
The large need of networked multimedia system has created the need of COPYRIGHT PROTECTION. It is very important to protect intellectual properties of digital media. Internet playing an important role of digital data transfer. Digital watermarking is the great solution of the problem of how to protect copyright. Digital watermarking is the solution for the protection of legal rights of digital content owner and customer.

34

CHAPTER 7 REFERENCE
[1] http://www.wekipedia.org [2] Kim, H.J., Audio Watermarking Techniques, Paci_c Rim Workshop on Digital Steganography, Japan, July 2003. [3] Arnold M., Audio watermarking: Features, Applications and Algorithm, IEEE International Conference on Multimedia and Expo., vol. 2, pp. 1013-101, 2001. [4] Cano, R.G., et al, Analysis of Watermarking Schemes, 2nd International Conference on Electrical and Electronics Engineering (ICEEE) and XI Conference on Electrical Engineering, Mexico, 2005. [5] Arnold M, Audio Watermarking, Dr. Dobb's Journal, vol. 26, Issue 11, pp. 21-26, 2001.

35