Do modern systems encode a degree of variability? Either by way of slightly varying the parameters of the underlying vectors (beziers?) to mimic the slight variability in speed and curvature of a true signature, or by analysing 100 input samples to generate a wider variability of outputs?
Seems like this would add to the ambiguity over provenance.
You could probably use some form of DMP [0] for this. It basically learns movements by viewing them as the difference from a spring-damper system. This allows all kinds of adaptations.
Seems like this would add to the ambiguity over provenance.