Digital Watermarking

Here is a brief introduction to digital watermarking. Reference [1] is a very good introductory text on digital watermarking. You can also check out my work report [2] for an evaluation of different watermark embedding techniques. In the report I compared the effectiveness of embedding watermarks in the spatial, frequency and wavelet domains.

What is digital watermarking?

Digital watermarking is defined as "the practice of imperceptibly altering a Work [the image] to embed a message about that Work." [1] Watermarking works in two phases, the embedding phase and the detecting phase. In the embedding phase a message and an image, called the cover work, is passed to the embedder algorithm. The algorithm embeds the message within the image resulting in the watermarked image. In the detecting phase, the watermarked image or any image is passed to the detector algorithm, which returns a message indicating whether a watermark is present in the image.

Watermarking Process

Properties of Watermarks

Watermarks have three main properties: fidelity, robustness, and detection error. These properties determine the effectiveness of the watermarks.

Fidelity

Fidelity refers to the perceptual similarity between the original and watermarked work. A good watermarking algorithm will ensure that the viewer sees no difference between the original and watermarked image.

Robustness

Robustness refers to the ability of the watermark to survive attacks on the watermarked image. Watermark should be robust enough to survive attacks such as JPEG compression, cropping, scaling etc.

Detection Error

Detection error is measured in two ways: false negatives and false positives. False negative is when the detector algorithm detects no watermark when there is a watermark present. False positive is when the detector algorithm detects a watermark when there is no watermark present. The detector algorithm should be able to prevent these types of errors.

Applications

In this section three examples of watermark applications are presented. These applications are broadcast monitoring, content authentication and transaction tracking. The algorithms used in each example will be different because they have different requirements.

Broadcast Monitoring

Consider a broadcasting studio which plays advertisements for various companies. How can the company know for sure that the studio did actually play the advertisement on the appointed time? One way is to pay someone to watch the studio's broadcast. This is not a feasible solution for a big company that wants hundreds of advertisements broadcast from different studios in a day.

Another option for the company is to have all its advertisements watermarked. This way a computer can be set up to scan all TV broadcasting and try to detect the company's watermark. This way the company can tell if their advertisement was broadcast.

A sophisticated watermarking algorithm is not needed in this case because the broadcast signals will only go through natural signal processing. It can be fairly certin that no malicious attacks will be used to remove the watermark.

Content Authentication

There are circumstances when the artist or composer of a media will want the ability to determine if their work has been altered and, if the work has been altered, how the work was altered. In this way the author of the work can track any misuse of their works.

Alteration of media can be determined by first embedding the media with a fragile watermark. A fragile watermark by design will degrade or change as the media is edited. Upon detection of the degraded or changed watermark one can determine that the work has been edited. Also if the fragile watermark is sophisticated enough that the change it goes through is constant under different editing techniques, it can be determined what editing technique was applied to the media.

Transaction Tracking

The artist or composer of work might wish to sell copies of their work to the public. As copies are sold more and more pirated versions of the work will be created and sold. Pirated copies must be tracked and destroyed.

In order to prevent piracy, the artist must track his works. Placing a unique watermark on each copy that is sold can do this. If a pirated copy is detected in the market, the pirated work can be scanned to see what watermark is on it. This way the original copy from which the pirated copies are being created can be tracked. Thus all sources of pirated copies can be eliminated.

In this type of system the watermark must be very sophisticated and robust. It can be guaranteed that pirates will send the work through malicious attacks in an attempt to remove the watermark.

References

[1] I. J. Cox, M. L. Miller, J. A. Bloom, Digital Watermarking. Morgan Kaufman Publishers, 2002.

[2] P. Siva, Effectiveness of Still Image Digital Watermarking Algorithms, University of Waterloo: Work Term Report, 2002.