Re: Statistics of Low-Order Bits in Images

[Timothy Newsham <newsham@wiliki.eng.hawaii.edu>]

> 
> 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.

Sounds like the simple solution is for people everywhere to replace
the low order bits of all of their pictures with good random noise. Image
quality shouldnt suffer drastically and if the random data is replaced
with output from a good cryptosystem then it would be indistinguishable.
So are any people here influential with the authors of any popular
imaging software?