Steganography idea: CU-SeeMe

[Nelson Minar <nelson@santafe.edu>]

The thing that bothers me about existing steganography code I've seen
is that it all uses uncommon communication channels to hide data. For
instance, the "hide data in a picture" programs: useful, up to a
point, but how often do I send pictures to other people? I think to be
effective, methods need to be employed that exploit existing, well
used communication channels.

So here's one idea I've had as a place to hide a channel: network
video, in particular CU-SeeMe video streams. CU-SeeMe is a lowtech
network video application, people running Windows and Macs with a $99
camera and a PPP stack can send about 2fps (160x120) to each other
over the Internet. Times I've used it, bandwidth is about 8kbits/sec
(I'm on a modem).

So here we have a reasonably high bandwidth channel, person to person,
that is already being used a lot on the net. How easy would it be to
borrow a few bits to hide the secret message? Because the image
quality is low (4bpp or 6bpp grey), and definitely nonrandom, you'd
need to be more subtle than just borrowing the low bit.

I think the answer could be found in image processing algorithms.
Spreading a bit out across various pixels would nice, as would some
sort of adaptive algorithm that identifies part of the image that's
safe to hide data in. It's not good enough that the image with hidden
data looks the same to the eye: it has to be statistically identical
to undoctored streams.

The biggest barrier to this I see is that the CU-SeeMe protocol itself
is proprietary, and the only existing free code I know of that does it
doesn't work very well (version 3.0 of a Unix reflector: buggy, now
very old). Of course you have to emulate the normal software 100%
correctly, or the Men in Black will see you're using stego tools.


I should remark that the subliminal channel stuff in digital
signatures is a really beautiful example of steganography.