13.3 Digital copyright labeling

Instead of trying to restrict and control the usage of some copyright protected material, one may also allow its unlimited copying and use, but make sure that some copyright- related information is available to anybody who cares and who is interested in the legitimate ownership of the intellectual property. This is where digital copyright labeling techniques or digital watermarking technologies come into play.

13.3.1 Introduction

In the real world, the term watermarking refers to a technique that can be used to impress into paper a specific text or image mark (called a watermark ). From daily life, we are all familiar with watermarks of varying degrees of visibility that may be added to presentation media as a guarantee of authenticity, quality, and ownership.

Similarly, in the digital world, the term digital watermarking refers to a technique that can be used to impress into digital data a specific text image mark or label. Quite naturally, such a mark or label is called a digital watermark. The aim of a digital watermark is to embed a digital mark into protected material to designate copyright-related information, such as origin, owner, content, or recipient. Unlike usage control, digital copyright labeling does not limit the number of copies allowed, but may deter people from illegal copying by allowing the determination of the legitimate owner of the protected material and the corresponding copyright (in the case of ownership labeling), or by allowing an illegitimately redistributed copy to be traced back to its original recipient (in the case of recipient labeling). Consequently, digital copyright labels may provide evidence for copyright infringements after the event. They may also serve as a kind of deterrent to illicit copying and dissemination by making the misuse of protected material traceable and providing evidence of illegal acts accordingly . Note that the use of digital copyright labeling techniques also requires a legal system that allows the copyright holders to sue people who breach their rights. Also note that the use of digital copyright labeling techniques is not contrary to usage control; it is, rather, complementary by providing another defense against misbehavior on the protected material that may have escaped from the domain of usage control. More specifically , digital copyright labeling techniques may also be used by sophisticated usage control technologies to encode control information into the digital data streams.

In general, there are two types of labels for identifying and protecting copyrights as related to multimedia documents:

A document can be marked with a label that uniquely identifies the copyright holder ( ownership labeling ).
A document can be marked in a manner that allows its distribution to be uniquely traced ( recipient labeling ).

In the literature, ownership labels are often referred to as watermarks, whereas recipient labels are often referred to as fingerprints . Consequently, fingerprinting a document means introducing individual marks into each copy sold or distributed that make the copy unique. This is similar to the way fingerprints make people unique [6]. Once an illegal copy turns up, the content provider can see from the fingerprint which of the original copies was illegally redistributed. Consequently, the corresponding party can be sued for having illegitimately redistributed his or her copy.

Although digital copyright labeling is relatively new as a means of protecting intellectual property rights, the theories and techniques behind it have been around for quite a long time. Refer to [7, 8] for a comprehensive overview about the theories and techniques that can be used for digital watermarking. By applying multiplexing techniques as in data

communications, some digital watermarking techniques can also be used to embed multiple marks (watermarks or fingerprints) and extract them separately. This feature will be important for identifying ownership and other intellectual property rights in works composed of many copyright assets, such as multimedia documents and presentations, as well as groupware and workflow documents.

The specific requirements of each watermarking technique may vary with the application, and there is no universal watermarking technique that satisfies all requirements for all applications. Consequently, each watermarking technique has to be designed within the context of the entire system in which it is being deployed. There are several parameters that are used to categorize watermarking techniques. The resulting categories are briefly overviewed next .

13.3.2 Categories of watermarking techniques

In this subsection we briefly overview some categories of watermarking techniques that apply for images. Other forms of multimedia documents, such as audio or video, generally require watermarking techniques that can be categorized according to slightly different criteria.

13.3.2.1 Visible and invisible watermarks

A digital watermark can be visible or invisible:

A visible digital watermark is intended to be perceptible by the user . As such, it typically contains a visual message or a company logo indicating ownership of the image.
Contrary to that, an invisible digital watermark is intended to be imperceptible but is detected or extracted by an appropriate piece of software. Consequently, an invisibly watermarked image is similar but not identical to the original unmarked image.

Users prefer to have a watermarked document behave no differently and suffer no perceptible quality degradation from the original. Consequently, users generally prefer invisibly watermarked images.

13.3.2.2 Fragile and robust watermarks

A digital watermark can be fragile or robust:

A fragile digital watermark is generally corrupted by any (image-processing) transformation. For example, watermarks for image integrity checks, in which a change must be detected or spatially localized, are necessarily fragile.
Contrary to that, a robust digital watermark resists common (image-processing) transformations. More precisely, the watermark that is embedded in the data must be recoverable despite intentional or unintentional modifications of the image. For example, a watermark technique for images should be robust against such image-processing operations as filtering, requantization, dithering, scaling, and cropping.

Robustness is a key requirement often imposed by applications. For example, watermarks that are used for ownership assertion should be robust. Unfortunately, the requirements of truly robust watermarks are difficult to meet in practice, and the development of robust watermarking techniques is a difficult problem. In fact, a single technique satisfying all requirements imposed on robust watermarking is quite difficult to achieve and is the subject of current research and development.

13.3.2.3 Public and private watermarks

A digital watermark can be public or private:

A public digital watermark can be detected and read by anyone not having access to certain secret information. All the user needs is an appropriate detector software.
Contrary to that, a private digital watermark can only be detected and read by someone who has access to an appropriate detector software and certain secret information, such as a pass phrase, a pseudorandom number generator seed, or the original image.

Obviously, private digital watermarking techniques are superior , from a security point of view. The secret information improves security but also renders detection of the watermark difficult or impossible without the secret information. This information must be communicated and distributed to a user or third party via secure channels if the watermark detection process is not always carried out by the image owner. Thus, a private watermarking scheme cannot be used for annotation or to inform a potential user of its proprietary status; only the content owner has the secret information that is required to detect the watermark. With a private scheme, the watermark can be used only to demonstrate ownership of content once its owner discovers its illicit use. Contrary to that, public watermarking techniques are attractive for many applications. For example, if we want to detect copyright violations in an image archive or in images published on the Web, we can use mobile agents, such as Webcrawlers, to perform identity checks for as many images as we can locate. Private watermarking techniques that require the original or a reference image in the watermarking detection procedure are less suitable for such applications (they require the mobile agents to locally store images).

Depending on whether secret or public keys are used for private digital watermarking, secret key and public key watermarking techniques may be distinguished (similar to secret and public key cryptography):

A secret key digital watermarking technique uses the same user key for watermark insertion and extraction or detection. Consequently, secret key digital watermarking schemes require secure communication channels between the image owner and the image receiver, or user, to pass the keying information.
Contrary to that, a public key digital watermarking technique uses separate keys for watermark insertion and extraction or detection. A private key is known only by the image owner and is typically used for watermark insertion, whereas a public key is known to everybody and is typically used for watermark extraction or detection.

Public key digital watermarking techniques are particularly important for digital fingerprinting and traitor tracing for broadcast encryption [9 “ 11].

13.3.3 Attacks

The proponents of digital copyright labeling have made broad claims regarding the security of their watermarking techniques, often without specifying which attacks they are expected to survive. Many of these claims have been disproved. In fact, some recent analytical results show that for many watermarking techniques, removal is not a difficult problem [12]. Consequently, it is required that a terminology similar to the one used in cryptanalysis is developed for the analysis of watermarking techniques.

According to [13], there are four classes of possible attacks against watermarking techniques:

Robustness attacks aim to diminish or remove the presence of a mark in a watermarked image without harming the image beyond rendering it useless. Typical signal-processing attacks revolve around commonly used operations, such as data compression, filtering, resizing, printing, and scanning. An example is the collusion attack, in which differently watermarked versions of the same image are combined to generate a new image, thereby reducing the overall strength of the watermark.
Presentation attacks are slightly different from robustness attacks in the sense that they don t necessarily remove the mark from the watermarked image. Instead, the image is manipulated so that the detector can t find the mark anymore. An exemplary presentation attack was developed at Cambridge University to foil automated Webcrawlers. The attack involves chopping a watermarked image into small parts that are then reassembled on a Web page with appropriate HTML tags. A Webcrawler sees only the individual image blocks, which are too small to contain any watermark. Obviously, this attack causes no image quality degradation, as the pixel values are preserved.
In some watermarking schemes, the mark s detected presence can have multiple interpretations. Consequently, an attacker can engineer a situation that neutralizes the strength of any evidence of ownership presented. Against this background, interpretation attacks aim to forge invalid or multiple interpretations from watermark evidence. For example, an attacker can attempt to introduce another watermark in an already watermarked image, thereby creating an ownership deadlock. Typically, such an attack requires in-depth analysis of the specific watermarking technique under attack.
Finally, legal attacks go beyond the technical merits or scientific evidence presented by watermarking techniques. As such, they make use of existing and future legislation on copyright laws and digital information ownership, the different interpretations of the law in various jurisdictions, the credibility of the owner and of the attacker, and the ability of an attacker to cast doubt on the watermarking scheme in the courtroom.

Understanding these attacks may help propose watermarking techniques that are more robust not only in the strength of their marks but also in their ability to guard against possible attacks. In either case, it is not known today whether the design and implementation of digital watermarks that are sufficiently robust against attacks is feasible at all.