Digital camera fingerprinting...
die at dieconsulting.com
Tue Aug 24 20:52:38 PDT 2004
Just a random distraction from the normal topics (but not
completely irrelevant either)...
I happened to spend a few minutes yesterday talking with an
individual who participated in the development of both low and high end
digital cameras for the commercial mass market.
He told me that especially in the low end camera market NO
sensors used were completely free of anomalous pixels (black, white,
dim, bright etc) and much of the actual processing in digital camera
firmware was related to masking or hiding the inevitable defects which
apparently can include (at least in CMOS sensors) entire rows or columns
that are bad.
This got me thinking - clearly these concealment patches are not
completely undetectable in families of (multiple to many) images taken
with the same exact camera... and for the most part the defects are born
with the sensor and change little over time if at all. And with few
exceptions they are random, and different for each sensor.
Thus it ought to be possible to detect with reasonable
probability that a particular image or (much easier) that a particular
family of images was likely to have originated with a particular camera.
A kind of digital fingerprint if you will...
Cypherpunk relevance (marginal perhaps), but the ability to say
that a particular image or set of images came from a particular camera
COULD have legal consequences for those bent on activities someone
thinks of as unfriendly to their interests...
Of course the headers of jpegs from cameras (and maybe
elsewhere) often contain serial numbers and other identifying
information so to the first order this is irrelevant to average users,
but interesting none the less.
Dave Emery N1PRE, die at dieconsulting.com DIE Consulting, Weston, Mass 02493
More information about the cypherpunks-legacy