Improve OCR with training file?

[drnicolas]

I am using ocrmypdf and use the ocred pdfs to search for text.
This text is mainly last names, first names and birth dates.

Is there an option to help ocrmypdf with these words? Names are typically not used as training objects.

The names in question are all in my database. So, my idea is to provide a list o these names as buzz words to help improving OCR

Good idea?

[jbarlow83]

The --user-words parameter is indeed to supply to Tesseract a list of words you consider spelt correctly. It's a plain text file with one word per line, include spaces.

The word list biases tesseract to prefer words that that are in its dictionaries (including user dictionaries) when it is not confident. So it's not absolute. If words have special capitalization that is considered too

Example word list file

drnicolas
OCRmyPDF

There is also a --user-patterns which could help with birth dates, depending on how birth dates are formatted.