Digital camera fingerprinting...

[Major Variola (ret)]

At 11:52 PM 8/24/04 -0400, Dave Emery wrote:
> Just a random distraction from the normal topics (but not
>completely irrelevant either)...

Highly relevant sir.

> 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.

Kinda like disk drives and DRAM arrays.  Its all about yield.
Covering up mistakes transparently.

> 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.

Perhaps, but the jpeg-ization might lose these, or at least the
image "unicity distance" might require many more pictures than
a careful steganographer will publish.

> 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...

Very relevant, traffic analysis and fingerprinting (intentional or not)
are
always tasty subjects.  One question for the court would be, how many
*other* cameras have column 67 disabled?   One of every thousand?
And how many thousand cameras were sold?

Pope Major Variola (ret)