Improve OCR using MRZ knowledge?

[fabwa]

Hi there. Thanks for this great tool. Question / Suggestion:
How would one go about narrowing down the character set of possible outcomes depending on character position within the MRZ?

e.g. names can never have numbers, dates are never alphabetical etc. and their positions are known. This would avoid 0->O confusions for these fields.

Thanks!

best

[stropitek]

I know that the SVM library that is used in this project can give you back probabilities when doing classification (see here), which would allow you to choose the one from your set of interest which has the highest probability.

However I've been away from this project too long now to know if the js version allows it (the js version is based on an emscripten build of the original c version, see https://github.com/mljs/libsvm)