I trying to create a model for determining whether a questioned hand written signature matches known signature samples, and predict if the signature is genuine or forgeries.

I'm guessing I'll have to use a CNN and probably some variation of PCA.

How would you tackle the "signature verification" problem?
Is there an already trained models for that? or datasets?

All I could find are research articles, but:

  • no GitHub projects.
  • no AWS/Azure/Gcloud service.
  • no Kaggle completions or datasets.

Maybe I not searching for the right thing!

Any dataset of signatures I could use, and ideas on how to approach that kind of problem would be very helpful. tnx :)

  • 1
    $\begingroup$ Here is a demo that shows cs.toronto.edu/~graves/handwriting.html how to recreate a text in a style of a particular persons handwriting. I guess that can also be extended to the Signature (Generation) or Matching. $\endgroup$
    – Kaustubh
    Aug 13, 2018 at 8:59

1 Answer 1


One of the key features in hand-write is the frequency of the signature, i.e. high frequencies - how "shaky" is the signature, and low frequencies - how rounded is the signature etc. For that, you would like to use Fourier transformation on the image, and train the results on labeled data, if available.

  • $\begingroup$ tnx.. that's a good idea! will use it. BTW Wouldn't a deep enough CNN figured it out all by itself? $\endgroup$ Aug 12, 2018 at 9:31
  • $\begingroup$ Upvote for not suggesting neural nets straight away. $\endgroup$ Aug 12, 2018 at 9:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.