From: norm@netcom.com (Norman Hardy)
Date: Tue, 25 May 93 18:08:12 PDT
To: cypherpunks@toad.com
Subject: Steganography
Message-ID: <9305260108.AA22729@netcom3.netcom.com>
MIME-Version: 1.0
Content-Type: text/plain


Here is a pitfall to be avoided in Steganography using
low bits of AD converter output. Such converters may be biased
in their low bit. If you hid 2,000,000 bits in a digitized image
you would probably get from 998,000 to 1,002,000 one bits
if you took no precautions. A real digitizer might well be
expected to produce more than
1,100,000 one bits or less than 900,000. Falling too close to
50% would be a clue that the data was not the yield of a
AD converter. Stuffing a few percent of extra one bits according
to a random number generator known to the receiver makes the
data look more typical.
 
There may be statictical dependencies with the next more
significant bit as well.
Some elementary statistics can be done on the yield of 
a real image scanner to examine this issue.