[Cryptography] Best internet crypto clock: hmmmmm...
leichter at lrw.com
Sun Oct 26 18:54:54 EDT 2014
On Oct 26, 2014, at 4:57 PM, Peter Gutmann <pgut001 at cs.auckland.ac.nz> wrote:
>> Along similar lines, small noise in image acquisition is now well enough
>> understood and discernible to say "This camera did take that picture" as it
>> does to say "This rifle did fire that bullet," which extends to "These two
>> pictures/bullets came from the same camera/rifle."
> There is, however, an ongoing battle between the device manufacturers and
> people who use these artefacts, since the manufacturers see them as flaws and
> try and eliminate them once they're pointed out. This makes it really
> annoying for people who use them for image-source authentication purposes.
Quite a few years ago, I argued that it should be possible to identify laser printers by small variations in toner placement. The argument the other way was that manufacturing tolerances would make this impossible. Nothing new here.
Manufacturing tolerances are reduced down to the point where they produce artifacts relevant for the use at hand. For a laser printer, that means visual effects noticeable to the human eye at the closest distance a paper page is likely to be held in normal usage. For a camera, it means visual effects noticed at the largest print sizes viewed at their appropriate ranges.
All of these things have fundamental limits set by human sensory capabilities. Most of our digital technologies are near those limits - most obviously in high-resolution LCD displays (what Apple calls "Retina" displays). Sure, under some circumstances, some well-trained observers can easily spot the remaining variations. But it's getting harder every day, and soon only the "golden ears" (and their analogues in different spheres) will even claim to be able to tell, and they'll consistently fail careful tests.
Once you get to that point, there's no reason to go further in controlling manufacturing processes and such. (Oh, some will for advertising points, but it's a very expensive business to wring out minor variations, so few will try.) And yet it's easy to *measure* details to orders of magnitude finer than you can *control* them. We may yet be in the period where images are getting more controlled fast enough to annoy the authenticators - but that period will end soon.
More information about the cryptography