Cryptography Reference
In-Depth Information
1
0
a
0
1
0
1
0
1
c
b
b
b
Figure 6.5: An expanded version of the tree shown in Figure 6.4. The
decisions about which of the dashed branches to take are made by
drawing bits from an extra pseudo-random source. Only the first
decision is made using a bit from the data to be hidden.
inated by one character that appears more than 50% of the time. If
“a”,“b” and “c” were to emerge 75%, 25% and 5% of the time respec-
tively, then it would not be possible to encode information with this
scheme and also produce the letter “a” 75% of the time.
One way around this process is to produce pairs of characters.
This is often feasible if one letter dominates the distribution. That
is, produce the pairs “aa”,“ab”,“ac”,“ba”,“bb”,“bc”, “ca”,“cb”, and “cc”
with probabilities of 56%, 18%, 3%, 18%, 6%, 1%, 3%, 1%, and .2%
respectively.
6.3.2 Regular Mimicry and Images
The regular mimicry algorithms described in this chapter are aimed
at text and they do a good job in this domain. Adapting them to
images is quite possible, if only because the digitized images are just
patterns of the two letters “1” and “0”. But the success is somewhat
diluted.
Chapter 9 shows how to flip the least significant bits to store in-
formation. Chapter 9 doesn't try to mimic the pattern of the least
significant bits. It just assumes that they fall into a standard even
distribution. The regular mimicry algorithms can be used to tailor
the distribution to some model.
The simplest solution is to group together the pixels into a regular
set of groups. These groups might be
2
×
2
or
3
×
3
blocks or they
