Real-world steganography

Sampo Syreeni decoy at iki.fi
Tue Oct 1 11:24:08 PDT 2002


On 2002-10-01, Ben Laurie uttered to Peter Gutmann:

>Yeah, right - and green felt-tip around the edges of your CD improves
>the sound, too.

I'm not sure about HDCD as a technology, but the principle is sound. If we
can compress sound transparently, we can also transparently embed quite a
lot of data into the part which is perceptually irrelevant. We might also
depart with perceptual equivalence and go with perceptual similarity
instead -- e.g. multiband compress the audio, and embed data which allows
us to expand to a higher perceptual resolution. Whatever the
implementation, putting data in the gap between statistical (i.e.
computed against a Markov model) and perceptual (against a perceptual
similarity model) entropy which compensates for some of the perceptual
shortcomings (like total dynamic range) of a particular recording
technology seems like an excellent idea.

However, applications like these have very little to do with steganography
proper. In this case, we can (and want) to fill up the entire gap between
statistical and perceptual entropy estimates with useful data, leaving us
with signals which have statistical entropies consistently higher than
we'd expect of a typical recording with similar perceptual
characteristics. That is, the encoded signal will appear manifestly random
compared to typical unencoded material from a similar source, and we can
easily see there is hidden communication going on. Such encodings will be
of little value in the context of industrial strength steganography used
for hidden communication.

Steganography used in the latter sense will also have to be imperceptible,
true, but but here the entropic gap we're filling is the one between the
entropy estimates of our best model of the source material vs. that of the
adversary's. Be the models Markov ones, perceptual, something else, or
composites of the above. Consequently the margin is much thinner
(bandwidths are probably at least a decade or two lower), and the aims
remain completely separate.

Consequently, I don't believe encodings developed for the first purpose
could ever be the best ones for the latter, or that HDCD-like endeavors
really have that much to do with the subject matter of this list.
-- 
Sampo Syreeni, aka decoy - mailto:decoy at iki.fi, tel:+358-50-5756111
student/math+cs/helsinki university, http://www.iki.fi/~decoy/front
openpgp: 050985C2/025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2


---------------------------------------------------------------------
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo at wasabisystems.com





More information about the cypherpunks-legacy mailing list