For the roll-your-own option, perhaps you could look into more conventional image duplicate-finding approaches, some of which have been discussed here, e.g. by BrowserUk, though you'd need access to the raw scanner images, and have to experiment with the comparison to see if it can be tuned to be sensitive enough to find a reasonably small number of matches without deciding that all fingerprints in your database are matches. I'd think you'd use the fingerprint only as a first identifier, then prompt for a password, so the fingerprint tool need only narrow down your ID to a small number of potential matches in your database, not a single person.

Finally, as interesting as I find the problem, I must say I have serious doubts about biometrics, both in ethical and practical senses. For starters, even in my personal circle of friends and relatives I can count half a dozen missing digits. Then there is the recent suggestion that fingerprints are not so easy to accurately compare as has been thought (and indeed haven't really been properly investigated), and the now famous gelatine hack. Perhaps something like a hardware key would be easier and safer to implement in your instance?