# How reliable is a mathematical model of a human fingerprint for identification?

How reliable is a mathematical model of a human fingerprint for identification?

I am looking for a way to uniquely identify individuals that is very reliable and easy to use that does not require storing actual biometric data. Storing the actual biometric data would seem to violate privacy. The method outlined in this article seems like it might be a good fit. I don't know the false positive / false negative rates.

Mathematical Models of Fingerprints on the Basis of Lines Description and Delaunay triangulation

• If storing actual biometrics violates privacy why do you think that storing a model of the fingerprint containing enough data to identify a human would not? A jpeg image or your mathematical model are just representations of an actual fingerprint, so they should be seen more or less the same. Sep 18, 2021 at 22:12
• @SergeBallesta From a full copy of a fingerprint the fingerprint can be reproduced. A mathematical model of a fingerprint ignores most of the details, thus a full fingerprint cannot be derived from a mathematical model of a fingerprint. Sep 18, 2021 at 22:24
• Then it would be close to a hash. That is a non invertible representation of the fingerprint. Unsure about the legal aspects, but it would be interesting on a technical point of view because it would allow to remotely control a fingerprint without a full image. This is done for passwords for security reasons (less consequences for a user database leakage) and could not be done for fingerprints. But I am far from an expert on this point... Sep 19, 2021 at 8:49
• @SergeBallesta I was looking at a hash initially. It may have too many false negatives. A set of hashes may be made to be nearly as reliable as the actual fingerprint. Sep 19, 2021 at 11:14
• @SergeBallesta The advantage of a mathematical model over a hash is that the model can be stochastic whereas the hash must be deterministic. The model could store the relative position of data points and still match with a slightly less than perfect mapping. Sep 20, 2021 at 1:47