Well, they have to store fuzzy hashes rather than cryptographically secure ones since they're going to get a different section of the finger and slightly different features within that finger each time. There's a good chance that whatever form of fuzzy hashing they're using is reversible in the sense that, given a hash, you can create a fingerprint that isn't necessarily exactly the same as the original but will match that hash.

For example, the obvious approach is to store fingerprint features, which will be then matched by any print that has the same features in the same positions. If you do a good enough job of generating the new print you might even be able to fool police investigations, since they compare prints the same way.