...

3. The shop or customer wishing to authenticate the part takes the number stamped on the part, runs it through the *public* key of the manufacturer (widely available, not kept secret, of course) and gets back the feature vector, which he can then compare to what he actually sees on the object. ....

This seems to have a tricky dependence on the tolerance. The forger can get a valid plaintext and signed feature vector. So, if the tolerance for error is too low, you get false positives, but if it's too high, a forger could create something starting from the feature vector. An interesting CAD/CAM problem.

To keep black market forgery part off the market, a 30% tolerance is way more than enough. There should be no false negatives (making a real part look fake), but if 1/3 of the forgeries slip through (i.e. 2/3 don't), this has the affect of driving the forgery price up by a factor of 3, effectively pricing them out of the market. (Unless the real goods are overpriced a factor of 3...:-) ************************************************** * Allen J. Baum tel. (408)974-3385 * * Apple Computer, 20525 Mariani Ave, MS 305-3B * * Cupertino, CA 95014 baum@apple.com * **************************************************