Make some fake fingerprints?? Make some real fingerprints. One can then treat it as either an anomaly detection problem, or supervised classification problem.

Anomaly detection:

Having trained on a set of only real finger-prints can we identify the fake ones as anomalies. (Perhaps check out one-class svm see http://rvlasveld.github.io/blog/2013/07/12/introduction-to-one-class-support-vector-machines/)

Supervised detection:

You train on both the fake and real finger prints. In this case you have labels for both classes. Its a fairly standard classification problem then I guess.

You may have to investigate some different feature extraction methods to feed into your classifiers in either of the above. I knew someone who was looking at using Gabor type filters for fingerprint analysis. 

Do you already have some form of training/test data?

Best

Alex