Cryptography Reference
In-Depth Information
might be broken into linear groups of pixels from the same row. Now
the least significant bits of each of these pixels can be treated as
characters. One image might be used as the model used to compute
the distribution table that generates a Huffman tree. Then data can
be hidden in another image by using this Huffman tree to generate
blocks of bits to replace the least significant bits in the image.
More sophisticated solutions could be based on the color of the
pixels themselves, but they are probably too complicated to be prac-
tical. The advantage of this system is that it could detect and imitate
any statistical anomalies introducedwhen an image was created. Or-
dinarily, CCD arrays have slight imperfections that affect how each
sensing cell reacts to light. High-quality arrays used by people like
NASA are tested and corrected. Most civilian arrays never receive
this individual treatment. The system might pick up any low-level
incongruities if they happen to fall in a pattern that is reflected in the
statistical distribution of the pixel groups.
6.4 Summary
This chapter described how to produce mimic text that looks statis-
tically similar to the original text. The mechanisms in this chapter
treat letters as the individual element, something that allows the data
to pass some statistical tests but fail others. The count of letters like
'e' and 't' might be consistent, but there are often large numbers of
words that can't be found in a dictionary. Another approach taken by
some experimenters is to treat words as the individual elements for
the statistical models. This requires more text to create the model,
but it provides excellent, if rambling, results. There are nomisspelled
words that aren't found in the source.
Chapter 7 describes how to use a more sophisticated grammar-
based method to achieve a better result. Chapter 8 goes even further
and shows how a Turing machine can be made to run backward and
forward to produce the most complicated text.
The Disguise
The text produced by these regular mimic functions
can be quite realistic. The results are statistically equivalent.
First-order text will have similar first-order statistics. Second-
order text will have the same occurrence of pairs. This can be
quite realistic in the higher orders, but it will rarely pass the
reading test. Humans will quickly recognize it as gibberish.
How Secure Is It?
There is no reason to guess that this system offers
any more security than hiding the information. How hard it
