In message <[EMAIL PROTECTED]>, David Honig writ
es:
>1. Measure at the stats on the least significant bits from your cover
>source.
>2. Shape the uniform distribution you get after encryption into
>the form you observe from your coverbits.
>3. Replace.
This approach is flawed. The distribution of the least significant
bits depends on words from which the LSBs originate. If you consider
only the statistics of the least significant bits, you will distort
statistics based on the whole data set.
You can find a paper that I wrote about the topic at
http://www.citi.umich.edu/u/provos/cv.html#papers
OutGuess, http://www.outguess.org/, is an implementation that will
preserve such statistics. It also ensures that the results from, e.g.
Maurer's Universal test do not change significantly.
Greetings,
Niels.
