This is a hackathon (StirHack) project developed in less than 24 hours.
The main core was a basic deep neural net written by David Mille, explained in this video: https://vimeo.com/19569529.
We add up a training methodology and tweak some parameters of the net so it would accomplish what we wanted. 
Moreover a Python script was created to transform every signature picture in a standarize array of values to treat them easily.

And what's that, you may ask?

The SignatureAuthentifier is a programm that uses neural networks and a set of signatures to learn how to identify a given signature while not being fooled by forges.
While showing a small version of it, a complete programm would be extremely useful in any area where strong security is required. 
We build it with a combination of Python to code images and C++ to do the computing bulk.