This is just pet project.

Task: Each fingerprint is written in individual file. Part of file name is id of person. Each fingerprint is 400 float vector. Input is two fingerprints. Output is 1 if fingerprints belong to one person, 0 otherwise.

Implementation: Keras (tensorflow) network based on Dense layers is used. Two stage learning:

1. First network provides classification of one fingerprint. Output is id of person.
2. Second network use part of first network. It takes 2 fingerprints, each is encoded by first network, then both representations are combined and binary classification is provided.