Original text visible beneath translated text.

[chad130]

I'm having trouble figuring out how to identify and remove the original non-English text from a translated pdf. Is this possible? I can pull the English text using "get_text()", so maybe I have to construct a new pdf somehow?

This occurs after turning a non-searchable pdf to an image(pymupdf), then ocr(Tesseract), then back to pdf. I then run the pdf through Google Translate and the overwritten text is the result.

pdf before translation:
page_4.pdf

pdf after translation:
page_4_tr.pdf

[JorjMcKie]

The translation has converted the original page to an image, upon which the translated text has been written.
So the easiest way probably is to just remove that image:

import fitz
doc=fitz.open("page_4_tr.pdf")
page=doc[0]
print(page.get_images())
[(13, 0, 1681, 2378, 8, 'DeviceRGB', '', 'FXX1', 'DCTDecode')]
page.delete_image(13)
doc.save("english.pdf", garbage=3, deflate=True)

This gives a page with only the English text - nothing else. Note the garbage collection and compression options.

[chad130]

Maybe I'm doing something wrong, but whenever I try this method, the "page.delete_image(13)" function triggers an AtrributeError:
AttributeError: 'Document' object has no attribute 'is_image'

Otherwise, I get the same response from get_images() as you. Knowing it's an image is new info, so I'll play around with that and maybe I'll find another way. Any additional help you could provide would be appreciated.

[JorjMcKie]

there was an error in an earlier release, which used a wrong method name
Either upgrade or do Document.is_image = Document.xref_is_image.

[chad130]

I think I'm using the latest version(1.21.1). I uninstalled, purged cache, and reinstalled to verify. The problem seems to be with the fitz.py file. It contains the old method you referenced.

Is it possible I'm not getting the latest version of fitz.py with my pip install? I receive version 4.0.2

Either way, the fix u suggested did the job. Alternatively I had success changing the xref_is_image method to is_image within the fitz.py.