Statistics of Low-Order Bits in Images

[Timothy C. May]

Jim Choate writes:

> Some other factors one needs to consider when analyzing images are:
> 
> * The lsb is going to be random if the image comes from any kind of a/d
>   process. This is because all convertors have a error of +/- 1 bit.

Nope. Not true. Some ADCs digitize with _more_ than the final
resolution and then do rounding or noise-shaping. And ADCs even at the
LSB can still have structure caused by other things, such as the image
itself (a binary image with thresholding will have the "LSB" certainly
not random noise! Q.E.D., by induction.) 

This can give the LSBs in the final product (image, DAT, CD) nonrandom
noise characteristics. This is what we're talking about.

> * If an image is processed by software then the last significant bit will be
>   on or off consistantly for a given color.

Huh? What if the "processing" is "do nothing"? Why will LSBs be
changed consistently? I must be misunderstanding your statement.


> * By the use of exlusive-or operators it is possible to mask the data in a
>   suitable manner. This does require that both sender and reciever have a
>   copy or method of calculating the correct key.

Key generation and sharing (if necessary) is separate from the stego issues.

> The real question is how do you generate keys?

In various ways, depending on what you're doing. Separate from the
main analysis of stego and image statistics/

--Tim May


-- 
..........................................................................
Timothy C. May         | Crypto Anarchy: encryption, digital money,  
tcmay at netcom.com       | anonymous networks, digital pseudonyms, zero
408-688-5409           | knowledge, reputations, information markets, 
W.A.S.T.E.: Aptos, CA  | black markets, collapse of governments.
Higher Power: 2^756839 | Public Key: PGP and MailSafe available.
Note: I put time and money into writing this posting. I hope you enjoy it.