Structure Of A Typical Watermarking System

Published Date: 02 Nov 2017

Chapter 1

Digital Watermarking

1.1 Introduction

With the rapid development of digital multimedia techniques and wide spread distribution of digital information over the internet, protection of intellectual property rights have become increasingly important. The information which include still images, video, audio or text is stored and transmitted in a digital format. Information stored in digital format can be easily copied without loss of quality and distributed efficiently because of easy reproduction, transmission and manipulation. It allows a pirate to violate the copyright of real owner. Therefore some work needs to be done to develop security systems to protect the content of digital data.

In recent years, digital watermarking is one of the best potential techniques for multimedia authentication and copyright protection by embedding some information called watermark into the digital multimedia content. This information later can be extracted from or detected in multimedia content to make an assertion about the data authenticity and copyright protection etc. Digital watermarks remain intact under transmission /transformation allowing us to protect ownership rights in digital form and are robust to several unintentional as well as deliberate attacks. Absence of watermark in a watermarked image would lead to the conclusion that the data content has been modified [1].A watermarking algorithm consists of watermark structure, an embedding algorithm and extraction or detection algorithm.

1.2 Structure of A Typical Watermarking System.

The process of digital watermarking is divided into three basic steps which are shown in Figure1.1. They are watermark selection, watermark embedding and watermark extraction processes.

Watermarking

Watermark Extraction

Watermark Selection

Watermark Embedding

Figure 1.1 Steps in Digital Watermarking

In the first step, an appropriate watermark is selected for watermarking of an image/video. Basically there are three types of watermarks that can be embedded in an image/video such as Pseudo Random Gaussian Sequence (PRGS) watermark, binary logo watermark and gray scale or color image watermark. A Pseudo-Random Gaussian Sequence watermark is a sequence of 1’s and -1’s. This kind of watermark possesses a unique statistical property that the watermark has zero mean and unit variance. These watermarks only give a binary response on detection i.e.,1 or 0 for presence or absence of watermark respectively. Such watermarks do not reveal any copyright information. All first generation watermarking algorithms were relied on PRGS watermarks.

binary lena lena_gray color lena

Figure 1.2 Binary, Grayscale and Color watermarks

A binary logo watermark is a kind of watermark which embeds meaningful data in the form of organization’s logo or owner’s image. Such watermarks are used for subjective detection i.e. the detection is based on human observer’s cognitive recognition power. The advantage of using binary watermarks is that it has some contextual information which can be perceived by human observer. Compared to PRGS which only offers a statistical measure to detect watermark, binary watermark offers visual detection measure which can be used as evidence to convince the jury in the court of law. A gray scale or color logo watermark is similar to a binary watermark but with a greater contextual details. The difference in contextual information is evident if one compares the watermarks shown in Figure 1.2. The watermark embedding and extraction processes are described in the following subsections.

1.2.1 Watermark Embedding Process

In Figure 1.3(a), the watermark embedding process for still images and videos is presented. The original image is denoted by I, a watermark by W, the watermarked image by IW and K is the embedding key. The embedding function Emb takes on its input as the original image I, watermark W and key K and then generates a watermarked image IW. Introduction of the embedding key K is necessary for enhancing the security aspect of the watermarking system. The embedding process can be performed either in spatial domain or transform domain. The domain selection depends on the watermarking application.

The inverse transform must be applied in order to obtain the watermarked image if the embedding is performed in transform domain. Mathematically, the embedding function for the spatial domain techniques can be represented as follows:

Emb (I, W, K) = IW……………. (1.1)

For the transformed domain, the following expression is valid.

Emb (f, W, K) = IW……………. (1.2)

Where f represents the vector of coefficients of the transformation applied.

Key K

When the watermarked image is obtained and placed on internet or transmitted over the communication channel, possibly, the attacks occur. The attacked /altered or modified watermarked image is denoted by Ir [2].lena_gray

Watermark Embeddinglena_graylena_gray

Watermarked Image Iw

Original Image Iimages

Watermark W

Key K (a)

Detection

Watermark Extraction

images

Extracted Watermark We

Yes/No

(1 or 0)

Watermarked Image Irlena_gray

Original Image I

(b)

Watermark W

Figure 1.3 Schematic Block Diagram of Watermarking (a) Embedding,

(b) Extraction/Detection

1.2.2 Watermark Extraction/Detection Process:

The watermark extraction/detection process for still images is shown in Figure 1.3(b). A detector function Dtc takes a watermarked image Ir whose ownership is to be determined. The detection function either recovers a watermark We from the watermarked image or checks the presence of the watermark W in a given watermarked image Ir. In this procedure the same key K is used. Depending on the selection of watermarking scheme, the original image I may or may not be used. Mathematically, the extraction procedure for blind extraction (extraction without using the original image I) can be expressed as follows:

Dtc (Ir, K) = We……………. (1.3)

For non-blind extraction (extraction using the original image) the following holds:

Dtc (Ir, I, K) = We……………. (1.4)

The blind watermark detection generates at its output, a binary value indicating the presence or absence of the watermark W. By this, the following can be assumed:

A watermark must be extractable or detectable. In the watermarking extracting schemes the watermark is being extracted in its exact, original form. On the other hand, detecting a specific given watermarking signal is present in an image or not, then the scheme is called the watermark detection scheme. The ownership can be proved from watermark extraction where as the verification can be done by watermark detection [2].

1.3 Applications of Watermarking

Digital watermarking is described as a viable method for the protection of ownership rights of audio, image, video and other data types. It can be applied to different applications such as copyright protection, fingerprinting, copy protection, broadcast monitoring and authentication etc [3][4].

Copyright Protection: For the protection of the intellectual property, the data owner can embed a watermark representing copyright information in the original data. The embedded watermark can be used as a proof against someone in a court who intentionally infringed the copyrights [5].

Fingerprinting: To trace the source of illegal copies, the owner can use the fingerprinting technique. In this case, the owner can embed different watermarks in the copies of the data that are supplied to different customers. Fingerprinting can be computed by embedding a serial number that is related to the customer’s identity in the data. It enables the intellectual property owner to identify customers who have broken their license agreement by supplying the data to third parties.

Copy protection: The information stored in watermark can directly control digital recording devices for copy protection purposes. In this case, the watermark represents a copy-prohibit bit, and watermark detectors in the recorder authenticate the data offered to the recorder [6].

Broadcast monitoring: By embedding a watermark in commercial advertisements, an automated monitoring system can verify whether the advertisements are broadcasted as contracted. Broadcast monitoring can protect not only the commercials but also the valuable TV products.

Data authentication: The fragile watermarks can be used to check the authenticity of data. A fragile watermark indicates whether the data has been altered or not. Further it gives the information in which part the data is being altered [7].

1.4 Properties of Watermarking

In the literature, various watermarking techniques have been proposed in the past. In order to be effective, a watermark / watermarking system should have the following properties [8][9].

Imperceptibility: The embedding algorithm must embed the watermark in such a way that this does not affect the quality of the original image. If the human cannot distinguish the original data from the watermarked data with the inserted watermark, the watermark embedding procedure is considered to be truly imperceptible.

Capacity: Capacity or data payload is defined as the number of watermark bits encoded in original image within a unit of time or work. This property describes how much data should be embedded as a watermark to successfully detect during extraction. Watermark should be able to carry enough information to represent the uniqueness of the image. Different applications have different payload requirements.

Robustness: The robustness is defined as the ‘ability to detect the watermark after common signal, image and video processing operations. Watermarks could be removed intentionally or unintentionally by simple image processing operations like compression, cropping, rotation, contrast adjustment, gamma correction, Gaussian noise, linear and nonlinear filtering, frame dropping, frame averaging and frame swapping etc., hence watermarks should be robust against a variety of such operations.

Unambiguousness: The extracted watermark must identify the ownership unambiguously, which means the errors in an extracted watermark must be as low as possible. Craver et al. [10] showed that an attacker can watermark a watermarked image and hence claim ownership which is called a counterfeiting attack. So a watermarking algorithm should be able to resist such attacks.

Security: According to Kirchhoff’s principle, the security of a cryptosystem depends on the secrecy of the key and not on the cryptographic algorithm. Same rule applies to watermarking algorithms, i.e. the watermarking algorithms must be public but watermark embedding should be based on a secret key [11].

Computational efficiency: The watermarking algorithm should be effectively implemented by hardware or software. Especially, the watermark detection algorithm should be fast enough for multimedia data monitoring in the distribution network.

There exists a complex trade-off in digital watermarking between three parameters: robustness, capacity and imperceptibility as shown in Figure1.4.

Robustness

Capacity Imperceptibility

Figure 1.4: Trade-off in digital watermarking

It is quite easy to see that these three parameters are conflicting. One may want to increase the watermarking strength to increase the robustness but this results in a more perceptible watermark on the other hand. Similarly, one can decrease the data capacity by decreasing the number of samples allocated to each hidden bit but this is counterbalanced by a loss in robustness. As a result, a trade-off has to be found and it is often depends on the targeted application.

1.5 Types of Watermarking

Based on visual perception, image / video watermarking can be classified as visible or invisible. A visible watermarking typically contains a visual message or a company logo indicating the ownership of the image / video. An invisible watermarked image is visually very similar but not necessarily identical to the original unmarked image / video. The invisible watermark’s presence can be determined only through a watermark extraction or detection algorithm. There are three categories of the invisible watermarking according to the robustness

against various attacks as shown in Figure 1.5. [12]

Watermarking

Visible

Invisible

Robust

Semi - Fragile

Fragile

Figure 1.5: Types of Watermarking

Fragile Watermarking: The watermark should not tolerate any tampering that modifies the complete integrity of the image / video. Fragile watermarking is used for image authentication and integrity verification.

Semi-fragile Watermarking: The watermark should tolerate intermittent noise and common image processing operations such as lossy compression, but should be fragile to any malicious tampering that modifies image/video content. Semi-fragile watermarking is used for soft image authentication and integrity verification.

Robust Watermarking: The embedded watermark should be resistant to any signal and image processing attacks that do not seriously affect the quality and value of the original image/video. Robust watermarking is used for copyright protection.

1.6 Watermarking schemes

Watermarking schemes are categorized according to the original data required to extract or detect the watermark presence in a digital

medium as follows [2][13].

Non-Blind (private) watermarking: This scheme requires the original image for watermark detection and extraction. There are two types of private watermarking schemes:

Type I systems, which extract the watermark from the watermarked, possibly distorted image and use the original image to find the location of the watermark in distorted image.

Type II systems, which requires an additional copy of the embedded watermark for watermark detection and they are only able to tell whether a given watermark is present or not in the watermarked image. In both schemes knowledge about the private/embedded key is required. Here the private key is a secret data used to embed the watermark [8]. This type of watermarking system is expected to be more robust than the other types as it conveys very little information and requires access to original or watermark image.

Semi-Blind (semi-private) watermarking: This scheme does not use the original image for detection. It gives binary information 1 or 0 if watermark is present or absent respectively.

Blind (public) watermarking: This scheme requires neither the original image nor the embedded watermark in watermark extraction procedure. Public watermarking schemes have more applications than either private or semi-private schemes in the realm of digital media copyright protection and authentication [14][15][16].

The selection of scheme is application dependent i.e copy control application should offer blind detection where as copyright protection can use non-blind detection.

1.7 Attacks on Watermarks

Watermarked object may be altered either intentionally, or accidentally. In both cases the watermarking system should be able to detect and extract the watermark after attacks. The aim of attacks is not always to completely remove or destroy the watermark but usually to disable its detection. The best-known watermarking attacks which may be intentional or unintentional depending on the applications are: geometrical attacks, signal processing attacks, removal attacks, cryptographic attacks and protocol attacks etc [17][18].

Geometrical attacks:

Geometrical attacks attempt to desynchronize the watermark at the decoder. When desynchronization occurs, the decoder cannot find the pixels that have been embedded and thus cannot detect the watermark. The common geometrical attacks which include rotation, resize, cropping, shearing and row column removal etc [19] [20].

Signal processing attacks: Signal processing attacks reduce the watermark energy within an image/video. After such an attack, a decoder can locate the pixels that have been embedded but it cannot necessarily detect the watermark correctly (due to the low energy of the watermark).The common signal processing attacks which include linear and nonlinear filtering, compression, sharpening, Gaussian noise, salt and pepper noise, gamma correction, histogram equalization, contrast adjustment etc.

Removal attacks: Removal attacks achieve complete removal of the watermark information from the watermarked data without cracking the security of the watermarking algorithm. This category of attacks includes denoising, quantization, demodulation, averaging and collusion attacks etc.

Cryptographic attacks: Cryptographic attacks aim at cracking the security methods in watermarking schemes and thus find a way to remove the embedded watermark information or to embed misleading watermarks. One such technique is brute-force search for the embedded secret information. Practically, application of these attacks is restricted due to their high computational complexity.

Protocol attacks: Protocol attacks aim at attacking the entire concept of the watermarking application. One type of protocol attack is the copy attack. The main idea of a copy attack is to copy a watermark from one image to another image without knowledge of the key used for the watermark embedding to create ambiguity with respect to the real ownership of data [21].

1.8 Limitations of Watermarking

Digital watermarking techniques are effectively used in copyright protection, authentication, fingerprinting and broadcast monitoring systems etc. In combination with digital rights management frameworks, they can solve the limitations of the intellectual property dilemma in audio, image and video-related business areas. However, the main intellectual property problems cannot be solved by all existing watermarking methods. Watermarking techniques have employed differently in different attack operations or applications. Simple, noncomplex methods described in [22] are not very resistant to JPEG and JPEG2000 compression, but are resistant to normal image operations. Complex and difficult watermarking techniques based on Discrete Fast Fourier Transforms, Discrete Cosine Transforms or Wavelet Transformations are by contrast very robust against compression techniques, but lack resistance in normal image operations. Today, most watermarking methods cannot reach the main approach. It is still a wide and attractive field for further research in which innovative methods and techniques may be established.

1.9 Performance Measures & Benchmarking

Generally, there are three basic performance measures that are used to evaluate the performance of the watermarking scheme. They are peak signal to noise ratio, normalized correlation and bit error rate.

Peak Signal to Noise to Ratio (PSNR): The PSNR of an image/video frame is a typical measure used for assessing image quality by considering that the just noticeable distortions are uniform in all coefficients in a specific domain such as spatial, frequency, or other transform domain. Since, PSNR is easily computable and it can also be used to provide a generic bound for the watermarking capacity. In this thesis, the PSNR is used to analyze the watermark embedding distortions on watermarked image/video. Higher values of PSNR indicate more imperceptibility of watermarking. It is expressed in decibels (db) as follows.

…………….. (1.5)

where MSE (Mean Square Error) between the original and watermarked

image/video frame is defined as follows.

………………….. (1.6)

where M,N are the size of the original and watermarked image and I(i, j), Iï‚¢(i, j) are the pixel values at location (i, j) of the original and watermarked image/video frame [23].

Normalized Correlation Coefficient (NCC): It gives the similarity between extracted watermark and original watermark embedded. It takes values between 0 and 1. The NCC is 1, when extracted watermark is similar and exact to the embedded watermark. The NCC is 0, when extracted watermark is not at all similar to the embedded watermark. The NCC is between 0 and 1, when extracted watermark is partially similar to embedded watermark.

……… (1.7)

Here w is original watermark and we is extracted watermark., wï‚¢ and weï‚¢ are mean of original watermark and extracted watermrk image. Let n x n is the size of original and extracted watermark image [24].

Bit Error Rate (BER): Bit rate refers to the amount of watermark data that may be reliably embedded within an original image per unit of time or space, such as bits per second or bits per pixel. A higher bit rate may be desirable in some applications in order to embed more copyright information. In this study, reliability was measured as the bit error rate (BER) of extracted watermark data. For embedded and extracted watermark sequences of length B bits, the BER (in percent) is given by the expression:

………………. (1.8)

BER is zero (0%) ,when extracted watermark and embedded watermark are same. BER is one (100%), when extracted watermark is totally different from embedded watermark [25]. The following benchmarks are used to evaluate the robustness of the watermarking system.

Stirmark is a benchmarking tool designed to test the robustness of the digital watermarking schemes. For a given watermarked input image, stirmark generates a number of modified images, which can be used to verify whether the embedded watermark can still be detected. The following image alterations have been verified in stirmark: cropping, flip, rotation, rotation-scale, sharpening, Gaussian filtering, random bending, linear transformations, aspect ratio, scale changes, line removal, color reduction and JPEG compression [26].

Checkmark is a benchmarking suite for digital watermarking which is developed on Matlab under UNIX and Windows. It has been recognized as an effective tool for evaluation and rating of watermarking systems. Checkmark offers some additional attacks that are not present in stirmark. The following image alterations offered here are: Wavelet compression (JPEG 2000), Projective transformations, warping, copy, template removal, denoising (midpoint, trimmed mean, soft and hard thresholding, wiener filtering), denoising followed by perceptual demodulation, non- linear line removal and collusion attack [27].

Certimark is a benchmarking suite developed for watermarking of visual content and a certification process for watermarking algorithms [28].

1.10 Organization of Thesis

The general objective of this thesis is to advance the research in the area of digital image & video watermarking techniques and it is divided into six chapters.

The first chapter gives an overview of watermarking and basic terms in the field of watermarking.

Chapter 2 presents watermarking techniques and literature review on spatial, transform domain watermarking techniques and hybrid watermarking techniques for gray scale images, color images and videos are illustrated. The limitations and challenges of existing work are studied. The objective of the thesis is also given.

Chapter 3 introduces the concepts of singular value decomposition and contourlet transform. The properties of SVD and contourlet transform also briefly explained. A hybrid watermarking algorithm based on singular value decomposition and contourlet transform is given with experimental results.

Chapter 4 provides the preliminary concepts of the non-subsampled contourlet transform with its merits over contourlet transform. Another hybrid watermarking algorithm using non sub sampled contourlet transform and singular value decomposition is presented with experimental results.

Chapter 5 gives an overview on video watermarking. Two hybrid video watermarking algorithms based on contourlet transform along with singular value decomposition and non-subsampled contourlet transform with singular value decomposition are presented with experimental results.

Finally, Chapter 6 provides the conclusions and future research scope.