What is a digital watermark?
Why should I care about watermarking?
What information can be put into a watermark?
Is there a standard definition of watermark information?
What is the smallest chunk of data that can hold a complete watermark?
What types of data cannot be watermarked?

What is a digital watermark?

A digital watermark is, in essence, a hidden message, within a digitized image, video or audio recording. The watermark is integrated into the content itself, so it requires no additional storage space. It can contain any information that the party writing the watermark cares to embed into a given work. Practically speaking though, space is at a premium, so the embedded message is usually quite small, often a short number. However that identifier can be mapped to any other kind of information - the composer's name, the studio musicians who recorded his composition, or the name and e-mail address of the consumer who purchased a copy of the recording. Unlike a traditional watermark on paper, which is generally visible to the eye, digital watermarks can be made invisible or inaudible. They can, however, be read by a computer with the proper decoding software.

back to top

Why should I care about watermarking?

Digital watermarks make it possible to "establish responsibility" over copies of a given work in the digital domain. The lack of "audit trails" is the most compelling barrier to online commerce for digitized works of art and music. Pirated music and videos can be traded anonymously among parties that do not know each other through online indexing and copying suites like Napster, Gnutella and Freenet. Recovered copies of works pirated through these systems do not reveal anything about the parties that are trading them.

Blue Spike's definition of "digital watermark" and Giovanni® watermarking technologies further address the fact that the information embedded, essentially the watermark message, is independently invaluable to the content because it is used to establish responsibility for the work. Subsequent post-production copies of it that are licensed by distributors, agents and consumers can all be marked with identifying information that ties a given copy to an identifiable party. Consumers can participate in online music trading schemes, but not without revealing their complicity in acts of piracy. Likewise, out-of-channel music files can be traced back to distributors and fulfillment houses that are responsible for their distribution.

back to top

What information can be put into a watermark?

Anything a publisher deems useful can be encoded (The limits are the size of the target media and the size of the data to be encoded). Watermarks can be tailored to downstream users as a class, including information that can be used to enhance the playback of a piece of music on special equipment. Or they can be written to include any kind of consumer data, enabling the kind of one-to-one marketing schemes that are more easily animated online than any other medium. Our advice is that watermark information relate closely to distribution channels of the content to be watermarked and subsequently distributed.

back to top

Is there a standard definition of watermark information?

No, although the debate regarding how universal any digital watermarking system may be is now taking place.

Besides the debates within the content business on Digital Television, copy control techniques initiated by legislative pressures such as the Hollings Bill and several other bills introduced in Congress, it is clear the technology of watermarking suffers most from misguided expectations that any one implementation can handle several even conflicting purposes.

First, the Secure Digital Music Initiative (SDMI), an ad hoc industry group whose membership included Blue Spike and companies from the consumer electronics and recorded music industries, roughly defined a digital watermarking specification to protect recorded music that is vended via digital music distribution. (Players using the SDMI-specified watermark system had been introduced in 2000.)

Still, technically speaking, the specification SDMI defined for digital watermarks only addresses the file header and watermark payload that are set into music files, really strings of bits that will be queried by special chips in consumer players that will use them to control copying and playback of copied songs.

SDMI was unable to objectify a decision, though Blue Spike was the only statistically-inaudible system, as per SDMI's own analysis. Blue Spike also had the smallest computational footprint while not being successfully hacked during the HackSDMI effort. We do not believe that digital watermarks are immune from attack but do believe the a key-based watermark is the most appropriate application for efforts such as SDMI. In fact, HackSDMI was originally initiated by Blue Spike as a condition for its participation within SDMI. We submitted over 130 hacks, with approximately 12 reported successes from the HackSDMI oracle, though our friends in the pure academic community seemed to capture most of the press.

For the audio industry, Giovanni watermarking software uses a variety of methods for enabling support for any identifier. For instance, the Industry Standard Recording Codes (ISRCs) can serve as a baseline for watermark data, allowing for approximately 100 bits per message. The audio industry seeks to encode at a rate of at least one watermark message per 7 seconds, while mastering efforts may require only between 4-5 watermarks per track.

What must be understood are the trade-offs between the quality of the watermark, or rather its inaudibility, and the application in which it is used: automatically readable watermarks for radio monitoring and secreted forensic watermarks to source tag DVDs are two very different deployment scenarios, each with their own trade-offs in security and robustness.

For the still image industry, no standard currently exists.

We believe existing files are best used as watermark messages. An observation: watermark messages may not exceed 100 bits (12 characters) for the vast majority of commonly available images on the Internet. In commercial transactions the 12 characters would likely be a credit number or invoice number to establish responsibility for the paid copy. Unique 32 bit identifiers can easily be generated to serve practically any commercial transaction event.

back to top

What is the smallest chunk of data that can hold a complete watermark?

For audio, probably a few seconds, depending on the signal. Still, that is enough to hold a watermark in a sound snippet of a song used for an advertisement - or appropriated by a musician for his own recordings. For still images, the target image should be at least 100 x 100 pixels in order to hold a complete watermark.

back to top

What types of data *cannot* be watermarked? (psst... they can)

In general, any data that has zero tolerance for error, such as software in executable form, cannot currently be watermarked. Any additional information added to these kinds of data objects would cause users trouble. CPUs require that exact instructions be fed to them in order to process the data properly.

Alternative methods for watermarking such data have been patented by Blue Spike. These processes are designed to provide for watermarking security in software products. While Giovanni static software watermarks may encode licensing information, for example, dynamic approaches retain semantic relationships with the original unencoded software for tamper resistance. These implementations have further ramifications of watermarking a piece of content with its player hidden in the content itself. Instead of needlessly increasing the number of proprietary players, Blue Spike offers a "player-per-copy" approach now prevalent in content protection schemes.

Predetermined keys may be generated to create permanent associations between data, whether structured or unstructured, namely, the file formatting, and the generated key. Less computational complexity vis-a-vis traditional encryption is acheived. Authentication without traditional encryption techniques is an additional benefit. Registration and authorization for several on-line and off-shelf software and services benefit from Blue Spike's proven innovation in this area. Several areas of application deployment has resulted in market successes. These approaches focus on the issue of software's "functional value" instead of the "aesthetic value" used to guide content-based watermarking. When time is the constraint, especially competing for the attention of the market, even a little obscurity can provide, well, necessary.

Please see Blue Spike Patents