Is there a library or algorithm to identify forgery in images of documents?

I am trying to implement a deep learning solution to identify if there is any forgery in images of documents. This solution for example, https://github.com/ISICV/ManTraNet, is capable of identifying forgery in images of people or objects, however I have tried it with images of documents and it fails to do the job with such level of accuracy. For a better idea of what I am trying to achieve see the example below. Mind that this example has been constructed for explanation purposes only.

This is a dummy check:

I have forged this check by changing the check number at the top right, the name, and the first part of the number at the bottom. See below:

I would like to develop a solution that can classify the image above as forged, taking into account that often times indications of forgery are very subtle and that real examples are images of these documents and are not digitally created such as the ones above. Is there any paper, algorithm, or library that can help in this case? I have looked at multiple forgery detection solutions and papers and tried implementing a couple of CNNs with transfer learning, however they fail at getting this task right with robust results.