[aphx]

In the US criminal justice system (and likely most others), there is no standard for how to systematically compare fingerprint images--as distinguished from scanning a finger itself. The biometric machines are good at saying, "the surface being scanned has the features I'm looking for." But they can't take a latent print (e.g. lifted from a drinking glass or door knob) and match it to another (including one made "professionally" with an ink blotter) in a defensible way.

It turns out that fingerprint examination depends hugely on examiner judgement. That judgement is quite susceptible to several biases, such as a detective saying, "hey, we've got prints and someone in custody, can you match the prints?"

TV shows like CSI etc. have trained us to think of fingerprint identification of criminals as something like public/private key signing, but it's not nearly that trustworthy. The error rate is in the neighborhood of 15-30%!

The result is that fingerprint-comparison results are presented as facts when they are closer to guesses. This ends up convicting innocent people and letting guilty people go free.

If you would like to work on this, please contact me!