Remove text from a PDF

[samuelbradshaw]

Hi! Is there an efficient way to remove/delete all text from a PDF with pypdf?
Also, is there a way to remove all text that uses a specific font from a PDF?

[stefan6419846]

To remove all text which is based upon regular text operations, use writer.remove_text(): https://pypdf.readthedocs.io/en/stable/modules/PdfWriter.html#pypdf.PdfWriter.remove_text

Removing all text of a specific font requires a custom implementation and more effort - see the internal implementation of writer.remove_objects_from_page() for how writer.remove_text() works. An advanced implementation would have to look into the font definitions as well.

[samuelbradshaw]

Thank you! With writer.remove_text() I'm able to remove text correctly in some PDFs, but not others. I think it has something to do with text generated by OCR.

Example PDF where text gets removed correctly:
https://assets.churchofjesuschrist.org/6e/4e/6e4ecd14ebb511ee863aeeeeac1eaa7887580260/come_thou_fount_of_every_blessing.pdf

Example PDF where text does not get removed correctly (OCR text):
https://ia902808.us.archive.org/23/items/conferencereport1880a/conferencereport151chur.pdf

from pypdf import PdfWriter

writer = PdfWriter(clone_from='/Users/sbradshaw/Downloads/conferencereport151chur.pdf')
writer.remove_text()
writer.write('/Users/sbradshaw/Downloads/conferencereport151chur_notext.pdf')

I've been trying to get this to work using various tools – PyMuPDF seems to have the same issue (see pymupdf/PyMuPDF#4221).

[stefan6419846]

For the second PDF file, running your code removes the text layer correctly, which you can verify by trying to select the text layer using Ctrl + A or similar before and after. There just is no way to remove the text from the actual scanned image in the background as well - except if you additionally remove all images from your PDF file.

[samuelbradshaw]

Thanks for your patience. I checked again, and you're right – my Mac was being smart and making the text from the image selectable (Live Text). It was responding so quickly that I assumed there was still a text layer. This works great!