Maybe no stego on eBay afterall

[John S. Denker]

Regarding my remarks about the ubiquity of Poisson
noise in low-light photographs, Hadmut Danisch wrote:
> 
> I disagree for several technical, physical, photographical
> reasons, 

I can't imagine what those reasons are, so I won't respond.

> but we can ignore this for this discussion, since
> it is a standard method of physics to measure for a
> longer time or many samples at once and take the average to
> reduce noise.

Huh?  A camera has a limited exposure time, limiting
the opportunity to perform signal-averaging.  My
original note pointed this out.  Similarly I pointed
out that it is possible in principle to reduce the 
noise to arbitrarily low levels by using very strong
illumination, very wide aperture, and very long
exposure -- but there is no 11th commandment requiring
everyone to do so.  Therefore it is a fact, whether
you like it or not, that pictures have noise in them.

Audio signals have noise in them, too.

Real life does not have the sort of mathemetical purity
that software sometimes has.  Get used to it.

> What do you want to show? If you hide an information this
> way, how should the receiver decode/separate it? 

Receivers have been successfully decoding noisy signals
for some time now.  I think some guy named Shannon had
something to say about this:  how much signal we need
in order to achieve reliable throughput, et cetera.

> The receiver would
> need to know how your picture looked like before you added
> information 

No, it would not.

> or how to separate information belonging to the
> white picture and information belonging to your message.

For starters, a simple method for separating the signal
from the "carrier" image would be to use a high-pass filter.
The pixel noise has all frequencies, including very high
frequencies, while many natural scenes have areas devoid of 
high-frequency content.

> If you substitute just the LSB, you presume that the LSB is
> an industrial-strength random bit sequence. 

I don't know what that is saying.  The LSB is not strongly
random before or after encoding.  (It isn't entirely
non-random, either, but it would be a colossal blunder to
assume that everything is either entirely random or entirely
non-random.)

BTW, I didn't say substitute.  I said add.  

> But if this were
> the case, why the hell should digital cameras bother to include
> it in the stored image? 

Vulgar language does not make the argument more clear.

What's the point?  Use a lossless encoder, go to jail?

> If you have an image with a depth of
> 8 bit, where the LSB is just random nonsense, why not simply
> cut it away?

I say again, it is a colossal fallacy to assume that the
LSB is either completely random or completely non-random.

An application of this idea is called "guard digits" in 
the language of numerical methods.  This idea is also
discussed in the reference I cited previously, specifically
at 
  http://www.monmouth.com/~jsd/turbid/paper/turbid.htm#sec-entropy
and elsewhere.

> And secondly, your transmissions are eye-catching. 

No, they're not eye-catching in the sense of blowing
the cover off the stego.

> Why should someone transmit images of a white wall?

I never suggested it was necessary or appropriate to
photograph _just_ a white wall.  
  1) My mention of a white area was just to simplify
     the discussion.  My stego method is readily 
     adaptable to non-white areas by anyone with
     ordinary skill in the art.
  2) It is a fact, whether you like it or not, that
     many perfectly ordinary images on eBay contain
     white areas.  I just looked.  The very first
     image I came across was
http://members.shaw.ca/angelsofglory/NAUTICAL/HELMS/LARGEHELMS/largehelmfishside600.JPG
     which contains many, many pixels worth of areas close
     enough to solid colors to make the aforementioned
     high-pass filtering easy to carry out.

In response to Peter Wayner, Hadmut Danisch wrote:
> 
> Of course. If the LSB wouldn't correlate to the higher order
> bits in any way, they definitely wouldn't contain image information
> and the camera would do a better job to simply cut them away.

This is diametrically wrong.  Correlated bits are the
ones that do not contain information, and can be cut
away without loss.

Saying "of course" does not make the argument more clear.

> Latest cameras use a higher resolution than just 8bit, so the
> LSB of the image file isn't the LSB of the image anymore.

There is no law requiring images to be filed using only
8 bits.

---------------------------------------------------------------------
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo at wasabisystems.com