it'd be nice if this could produce text-overlaid PDFs

[jbothma]

tesseract seems to be able to produce PDFs these days with text overlaid on the image. This is useful for searching int he PDF when viewing that way.

It'd be nice if this could produce nice de-skewed PDFs

[jlsutherland]

Definitely. I think it would relatively straightforward to integrate. Would suggest building the text insertion into the Page class and then put a export_to_pdf() method on the Document class.

[jlsutherland]

Would you be interested in contributing @jbothma ?

[jbothma]

Yup - would love to. Won't get to it before next week but will start a PR when I can :)

It's part of the ocr command as an optional output format so not sure what the right place would be to integrate it with doc2text.

[jlsutherland]

Awesome, thank you!

The method's location in the code would be conditional on the way tesseract embeds that data. Does tesseract insert the data into a PDF, or it in a separate state that contains the text and placement information?

In the first case, we would need the method you mentioned that produces a nicely optimized pdf from the images first, then the embedding second. We need this method regardless, I think. In the second case, we could run the tesseract embed method at any time after we produce the fixed image crop.

Thoughts?

[jbothma]

So this is basically what I was talking about.

• doc2text's existing functionality to straighten and flatten and normalise would run first,
• product a multipage tif or whatever,
• then give to tesseract to OCR with pdf config file (for pdf output).

wget http://mfma.treasury.gov.za/MFMA/Urban%20Development%20Zones/Gazette%20No.%2026866.pdf
gs -dNOPAUSE -q -r500 -sDEVICE=tiffg4 -dBATCH -sOutputFile=test.tif  Gazette\ No.\ 26866.pdf
tesseract test.tif outbase pdf

produces https://www.scribd.com/document/324084564/Out-Base

[jbothma]

Tesseract produces the PDF already, so you'd select that as the output format of the OCR step. There's no intermediate hOCR or anything.