Cryptography Reference
In-Depth Information
two bits per pixel. Basic compression algorithms like JPEG can eas-
ily save a factor of 10 without significantly distorting an image. Most
digital cameras, for instance, now come with built in JPEG compres-
sion chips to save space and allow people to take more pictures. 24-
bit color may be slightly more accurate, but no one wants to waste
the space on it.
The same holds true for music. Today, MP3 files are much more
common than files that record the intensity at each time slice. Com-
pression algorithms like MP3 can easily save a factor of 10 over raw
digitized data. Newer algorithms can save even more. This is great
news if you're storing your CD collection on your computer, but not
if you want an easy channel to exploit for steganography.
This effect, incidentally, is what leads some steganographers to
hide information in the most “perceptually significant” parts of a file.
That is, they want to ignore the noise and hide the information in
the part that the humans can perceive. The noise will eventually ex-
tracted and removed by some compression algorithm, but the per-
ceptually important parts will live on. [CKLS96] Instead of hiding in-
formation in subtle changes of the intensity, hide the information in
the position of a person's nose or the length of the hair.
This is a good point, but it is more of a challenge for researchers
and a loose design principle. Even if basic mechanisms for exploiting
the noise in a filemay not be as robust as possible, they are still worth
exploring. The rest of this chapter is devoted to noise. Following
chapters attempt more robust solutions.
9.2.2 Good noise?
A practical problem is finding good noise. Most image and sound
files include enough natural noise
Chapter 17 discusses
how some cameras don't
provide good enough
noise to mask hidden
bits.
to hide a 3% change, but this noise is rarely as pure as can be.
Figure 9.1 shows a black-and-white scanned image of a photograph
taken of a computer on a desk. Figure 9.2 shows just the least signif-
icant bits. It is obvious that there is a highly random pattern to them
caused by the noise in the digitizing circuit on the scanner. It is ran-
dom, but it is not as random as it could be.
Many images and sound files probably have enough inherent
noise to hide data. The image in Figure 9.1 has plenty of junk so small
variations don't show up. But there are some images that do not han-
dle the imposed noise as well. Many images are created entirely on
the computer in applications like Adobe Illustrator. These produce
pure, consistent fields of color. Even modifying them a bit can stand
