NODE 25b4559fRe: HIDE: embedding msgs into snd & graphics
blojo@sting.Berkeley.EDU (Jon Blow)Fri, 12 Mar 93 00:53:27 PST
> I think color images, as opposed to mapped, would be the way to go
> for steganography. More room, and nobody expects them to compress.
Yeah; but even if we're talking full 24-bit images, we're going to have
problems. Once 24-bit displays become standard, I really doubt that
there will be many images stored in a non-lossy format. And once
you try to hide data in a lossy encoding scheme, you run into a
lot (though not all) of the problems you have with colormaps.
It would be a lot easier to hide data in an image if one had a control-image
(the original) as well as the altered-image (the one holding the message),
but this defeats half the niftiness of trying to do things this way.
> Eli ebrandt@jarthur.claremont.edu
-J.
NODE 5d7d638dRe: HIDE: embedding msgs into snd & graphics
Eli Brandt <ebrandt@jarthur.Claremont.EDU>Fri, 12 Mar 93 21:06:58 PST
> And once you try to hide data in a lossy encoding scheme, you run
> into a lot (though not all) of the problems you have with colormaps.
I think it's even worse... with unlossy compression, you can frob
the uncompressed bits and just lose compression. With lossy, you
can't do that, because your message will be smeared away. And
frobbing the compressed stream will produce ghastly artifacts.
But not all images will be lossily compressed. I find that JPEG,
for example, usually introduces too much gunk to be useful.
Certainly, sending an LZW TIFF should be above suspicion for quite
some time.
> It would be a lot easier to hide data in an image if one had a control-image
> (the original) as well as the altered-image (the one holding the message),
> but this defeats half the niftiness of trying to do things this way.
Yeah, this is just a cheesy OTP. Not much point, really.
I've been thinking about the GIF case; the "optimize for colormap
cyclic continuity" technique looks like it will produce better
images than the "crunch to 128 colors". Since I have to write some
annealing code for a neural-net trainer, maybe I'll repackage it for
colormap optimization and see what the results look like. If you
wanted minimum visible crud, you could start with a true-color pic,
find the colormap, order it, and dither down -- rather than adding
white noise to pre-existing dithering.
You know, I think I've been taking the graphics-weenie approach to
this. Who cares how the image looks? Who cares if it's even an
image? Just take your damned PGP file, ^=0xff it, and rename it
"hotbabe.gif". uuencode and mail. The NSA is not going to be
viewing every picture you send, I hope. This fails on "plausible
deniability", I guess.
> -J.
Eli ebrandt@jarthur.claremont.edu