So, it's actually really funny. All biometric hashes are roughly reversible.
The reason why, is that they are similarity metrics: They describe a series of vectors in multidimensional space, and the input is distance-checked against those vectors. If the input is close enough to the hash, it's treated as a match. The thing is, the return value is not a binary "match" or "no match", as you might get from a cryptographic hash. Instead, it's "hot" or "cold" -- and you can keep retrying, attempting to get hotter or colder. So, the way _all_ these biometric systems get broken, be they fingerprint or faceprint, is to generate a random input, and see how close you got. Then perturb randomly. Either you get hotter or colder. If colder, revert the change. If hotter, do more like that. Twenty or thirty thousand rounds later, you've got something that roughly looks like the fingerprint or faceprint. Now, the relevance to both of these document fingerprinters? Both are _very_ likely retrieving a fuzzy fingerprint of the target. An attacker with a fingerprinter and the document can retrieve the print, and start ginning up more and more samples to attempt to match. Those processes that lead to a similar substrate, he can duplicate, those that do not he can throw away. Now, things get interesting, because real matter is involved. Ginning up a million fake digital faces is easy, ginning up a million fake sheets of paper is not. If it's possible to "perturb" a surface, such that it can reversibly be made more or less like a given print, the game is lost. If however any modifications have effectively unpredictable effects on the print (possible!), then a security system could be developed. However, such a system would have to yield fairly radically different signatures across the range of the scanned surface. The upper right corner of the same page would have to yield a very different signature than the lower right. With the laser scanning system returning similar prints despite wettening, scorching, etc, I'm a little doubtful. But, *if* the system had that characteristic, then the problem would become matching a specific region of the document (with the material fingerprint) to the rest of the document. I suppose you could do this by hashing the fingerprint of a given region, with a dump of all the actionable bits on the page, signing that mash with a private key, and stamping the signature into a QR barcode on the page. So, ultimately, if the stuff works, we could actually use it to do cool things. But it depends pretty seriously on the nonlinearity of modifying a given material to match a particular signature. Most systems of this type have fallen, but there's loads of entropy at the micro scale of materials, far more than there is in your fingerprint or your face. So things are a little different here. Could be fun to play with! On Thu, Mar 18, 2010 at 7:10 PM, Wim Lewis <[email protected]> wrote: > On Thu, 18 Mar 2010, Gadi Evron wrote: > > Now, this is cool: > > http://nanotechwire.com/news.asp?nid=2254 > > I was also impressed by this other research, since it requires no > special equipment: > http://citp.princeton.edu/paper/ (or doi:10.1109/SP.2009.7) > > _______________________________________________ > Fun and Misc security discussion for OT posts. > https://linuxbox.org/cgi-bin/mailman/listinfo/funsec > Note: funsec is a public and open mailing list. >
_______________________________________________ Fun and Misc security discussion for OT posts. https://linuxbox.org/cgi-bin/mailman/listinfo/funsec Note: funsec is a public and open mailing list.
