Digital watermarking is the best way to protect intellectual property from illicit copying.
Jian Zhao
Why is it so difficult to find a needle in a haystack? Because of the size of the needle relative to the size of the haystack. Also, because once the needle falls out of your hand, it is not predictable where it will land in the haystack. Both principles -- inconspicuousness and randomness -- help conceal information in
digital watermarks
to protect intellectual property of multimedia documents.
In contrast to a traditional watermark on an invoice, for example, a digital watermark can be detected only by appropriate software. Rather than ensuring the authenticity or integrity of documents, as a digital signature or a digital seal does, a digital watermark aims to identify the origin, author, owner, usage rights, distributor, or authorized user of an image, video clip, or audio clip, even if the image or clip has been processed and distorted (via analog-to-digital conversion, low-pass filtering, resampling, lossy compression, cropping, or rotation).
Although digital watermarking is relatively new as a means of protecting intellectual property, the theories and technologies behind it have been around for a while: computer-based steganography (cryptography), spread-spectrum communications, and noise theory.
Traditional steganographic methods vary from simple invisible ink, microdots, and character arrangements to covert channels in digital communications. Earlier computer-based steganographic tools are inadequate to meet the stricter requirements of digital watermarking. Modern steganographic systems use spread-spectrum communications to transmit a narrowband signal over a much larger bandwidth so that the spectral density of the signal in the channel looks like noise.
The two different spread-spectrum techniques these tools employ are called
direct-sequence and frequency hopping. The former hides information by phase-modulating the data signal (carrier) with a pseudorandom number sequence that both the sender and the receiver know. The latter divides the available bandwidth into multiple channels and hops between these channels (also triggered by a pseudorandom number sequence). Co
Noise is not just an awkward companion of analog communications, it is also an ingredient in any digital information and a fundamental premise of digital watermarking. The
process of
watermarking encodes the hidden information as additional noise and incorporates it in the document. Modifications of the original's noise signal caused by moderate levels of
wideband
noise or controlled reduction of noise are not visible or audible.
Most common watermarking methods for graphics and audio signals work in the spatial, time, or frequency domains. The advantage of frequency-domain watermarking is that the watermark is spread throughout the whole video or audio clip and hence is resistant to cropping or cutting.
However, a standard frequency filter, or a lossy compression algorithm, which usually filters out the less significant frequencies, could damage the watermark.
Watermarks can also be embedded in an image's luminance and color bands, or in the contour and texture of an image. Common watermarking methods use the luminance band since it contains the most significant information of a color image.
Watermarking Tools
There are two different categories of watermarking tools available. The first is based on
fingerprinted binary information
(FBI), as exemplified by an eponymous product from the U.K. company HighWater FBI. The other, based on watermarking techniques developed at NEC Research and the University Catholique de Louvain (Belgium), identifies documents by hidden numbers (
fingerprints
). Other approaches, such as SysCoP (System for Copyright Protection), developed at the Fraunhofer Institute for Computer Graphics; Digimarc, from Digimarc (Portland, OR); and Argen
t, from DICE, can encode additional identification information such as the author's name or the ISBN number of a book.
Direct-sequence and frequency-hopping spread-spectrum techniques are the major watermark embedding methods used in existing tools. Both modify the noise value of the target documents. The direct-sequence technique adds noise to every element of the document, whereas the frequency-hopping method selects a pseudorandom subset of the data to be watermarked. Digimarc and FBI, for example, use direct-sequence methods to superimpose a watermark over an image by modulating a noise pattern of the same size onto the image. SysCoP, however, uses a secret key to pseudorandomly select blocks and frequencies that are modulated within the block.
Other systems use secret keys to determine which lines or words of a text will be slightly shifted vertically or horizontally. Hiding secret messages in the least-significant bits of some pseudorandom frequencies or pixels of an image, which is a common app
roach employed in many steganographic tools, can also be considered a simple example of frequency hopping. Because frequency hopping modifies only a subset of pixels or other elements of a document, it tends to be much faster than direct-sequence methods. It is, however, less robust and more vulnerable to attack.
Watermark Extraction
A watermark must be extractable even from degraded documents that might have been photocopied, scanned, or manipulated by imaging programs. If a degraded document does not have the same format, resolution, or physical size as the original, it has to be normalized to the original format before the watermark can be extracted. Typical normalization processes include format conversion, resampling, enlarging a cropped part to full size, and scaling of the signal level.
Watermark extraction includes two main steps: selecting the locations where the watermark has been inserted (only in frequency hopping) and retrieving the watermark from those locations. The
retrieval process normally needs either the original, unwatermarked data or the added noise for comparison with the watermarked document. It is also possible to extract the watermark without the original data. In this case the algorithm detects specific properties and patterns from the watermarked document. These patterns can be represented as signal shapes or the
cross-correlation
between certain document elements. This retrieval method is generally more efficient and enables, for example, SysCoP to retrieve watermarks in real time.
ries to overwrite the previous identification. However, uncontrolled hierarchical watermarking can cause accumulation of noise and hence create perceptual defects such as patterns or blocky images or, in the case of audio streams, hiss. Some watermarking systems -- HighWater FBI, for instance -- prevent this overwriting attack by permitting only one-time watermarking. In addition, watermark registration or deposit of watermarked documents in trusted agencies may provide more effective means to prevent watermark overwriting.
Counterfeiters Beware
or removing watermarks a time-consuming and costly task.
Where to Find
Aliroo
Kefar-Sava, 44442, Israel
Phone: +972 9 7677732
E-mail: support@aliroo.com
Internet: http://www.aliroo.com
DICE Company
Palo Alto, CA, U.S.
Phone: +1 415 326-0367
E-mail: info@digital-watermark.com
Internet: http://www.digital-watermark.com
Digimarc
Portland, OR, U.S.
Phone: +1 503 223-0118
E-mail: Info@digimarc.com
Internet: http://www.digimarc.com
HighWater FBI
Cheltenham, U.K.
Phone: +44 1242 221390
Fax: +44 1242 251600
E-mail: fbi@hwuk.demon.co.uk
Internet: http://www.highwaterfbi.com
Fraunhofer Institute for Computer Graphics
Darmstadt, Germany
Phone: +49 6151 155-147
Fax: +49 6151 155-444
E-mail: syscop@igd.fhg.de
Internet: http://syscop.igd.fhg.de
illustration_link (40 Kbytes)
Digital watermarks hide the identity of an image or audio file in its noise signal.
Dr. Jian Zhao is a senior system analyst and project leader at the Fraunhofer Center for Research in Computer Graphics (Providence, Rhode Island, U.S.). He can be reached by sending e-mail to
jzhao@crcg.edu
.
Look, It's Not There
