From: Pierre Uszynski <pierre@shell.portal.com>
Date: Mon, 19 Sep 94 18:32:40 PDT
To: cypherpunks@toad.com
Subject: Re:  art encryption...
Message-ID: <199409200132.SAA19126@jobe.shell.portal.com>
MIME-Version: 1.0
Content-Type: text/plain


> From: Lile Elam <lile@art.net>
> 
> [...] a way an artist can insure a viewer that the art 
> they are viewing is really that of the artist, the artist can *sign* their
> art with their private key and others can check it with their public key.

The problem is that's pretty much all that can be done so far: The viewer
can *verfify* that the copy is a full un-molested copy of what the artist
released, or for physical works, maybe that the object being scanned *is*
the original object (maybe). But for pictures released in bitmap form, and
then lossily compressed, cropped, scaled, dithered, and whatnot, the
(ahem) authenticity of the copy can't be checked anymore. At least not if
the signature was, for example, a signed message digest as has been
mentioned.

Message digests, in the form we now know them, do not resist lossy
compression and such injury. That's what they are designed for. So, in
particular, a magazine can still take a signed copy of a picture, creatively
modify it slightly, and print it claiming this is original work by somebody
else. In this case, the printing process itself is most likely enough to
dispell any hope of recomputing the same message digest. Handling this case
relies back on conventional evidence and juries.

Other formats such as CAD drawings, text, and pictures in primitive forms
(such as Postscript), and source and binary code, are susceptible to more:
It is possible to hide some identifying info in them. But they also are
susceptible to mechanical modification (lossless in terms of the
"artwork", but lossy in terms of the steganography.) For example, by
rewriting "for the same functionality" mechanically or manually. For a
mechanical translation, the plaintiff may now have to identify
which mechanical process was used before, maybe, having a case again
(in addition of conventional legal wranglings).

What would be ideal in all these cases is a steganographic process that
would map to the whole as much as to the parts. Fractal steganography.
In the case of architectural drawings, a signature of the original
architects would be embedded in small sets of dimensions, in small sets
of proportions, in ways the CAD language is used, in the background
details of artwork, in the text included in drawings as well as in 
CAD files comments and variable names... etc... In summary in sufficiently
numerous and different places that mechanical modification, or even
extensive manual rework could still forget some instances of the signature.

The signature still has to be specific enough to identify certainly the
author, as opposed to being attributable to random luck (as in DNA matching).
This means this signature requires a rather large number of bits (fewer
if the signatures are registered precisely and provably before the works
are released), but still enough bits that it is not so easy to hide many
of them (depending on the medium: rather hard in text, rather easy in
complex pictures.)

The signature becomes part of the picture, it is not an afterthought
anymore.

Fractal message digests, whose verifiability would degrade gracefully
as the original is more and more mutilated, would serve the same purpose
for verification (you could check that a decompressed, cropped
part of a picture is really from the artist who fractally signed
the whole.) Conceivably it would be the same if the part was printed
in weird colors or something: "some" of it still "is" from the same artist.

But, for legal issues, how would that be different from copyright
registration? Doesn't copyright registration rely on the same
principle: a set of jurors will determine what the chance is that
this is the same work or not. This just provides tools to fortify a
court case.

Enough ramblings :-) I'll stop now,
Pierre
pierre@shell.portal.com