Harvey Newstrom wrote:
I remain undefeated at detecting steganography. I would love to get a real
challenge that is not trivial to detect. I would even enjoy failing such a
challenge, because it might signal hope for the development of undetectable
steganography. Thus far, however, I see no evidence that undetectable
steganography is even theoretically possible.
### Suppose I use a one-time pad to specify a small number of
positions on a page as significant, randomly distributed throughout the
page. Then if the character filling a position is from "a" to, let's say,
"m" - it will be read as "1", if it's "n" to "z" -it means "0". I write a
text that fills the page, is grammatically and semantically correct, and
contains, in the specified positions, the correct characters to encode my
message. As long as the number of significant characters per page is low, I
will not have any problems with shaping my background text to fit the
message, writing without any irregularities in grammar which could be
detected by a machine, or bizarre semantic errors which might be detected by
a human. Since the one-time pad is different for each new page, there will
be no statistical abnormalities even in very long series of pages.
Alternatively, the one-time pad could specify the position of a single
significant character in each consecutive sentence or clause, thus sparing
me the need to maintain precise spacing between the significant characters,
as in the initial proposal.
Of course, this method is difficult, if not impossible, to fully
automate without human-level AI, and would require you to have something
akin to a talent for poetry, too. The bandwidth would be low, unless you are
writing huge amounts of e-mail which could become suspicious. Still, this
should be an undetectable form of steganography, just like manipulation of
high-level elements of pictures which I mentioned in one of my earlier
What do you think?
Rafal Smigrodzki, MD-PhD
This archive was generated by hypermail 2b30 : Sat May 11 2002 - 17:44:11 MDT