Redact phrases of text from a PDF using Python and (only) PyPDF2. Would need to modify (trivial) for multiple pages.

redact_pdf.py

import PyPDF2
from PyPDF2 import PdfFileReader, PdfFileWriter
from PyPDF2.generic import StreamObject, DecodedStreamObject
from PyPDF2 import filters

# You change the phrases/text you want redacted and the file paths:
phrases = ["123-45-6789", "YOUR_SECRET_PASSWORD", "YOUR_SECRET_ID"]
read_path = "/path/to/yourpdf.pdf"
write_path = "/path/to/yourpdf-redacted.pdf"


def replace_w_stars(text, phrase):
    return text.replace(phrase, "*" * len(phrase))


def redact(text, phrases=phrases):
    if isinstance(text, bytes):
        text = text.decode()

    for phrase in phrases:
        text = replace_w_stars(text, phrase)

    return text


class EncodedStreamObject(StreamObject):
    def __init__(self):
        self.decodedSelf = None

    def getData(self):
        if self.decodedSelf:
            # cached version of decoded object
            return self.decodedSelf.getData()
        else:
            # create decoded object
            decoded = DecodedStreamObject()

            decoded._data = filters.decodeStreamData(self)
            decoded._data = redact(decoded._data)

            for key, value in list(self.items()):
                if not key in ("/Length", "/Filter", "/DecodeParms"):
                    decoded[key] = value
            self.decodedSelf = decoded
            return decoded._data

# Overload with redaction version
PyPDF2.generic.EncodedStreamObject = EncodedStreamObject

# Read in the PDFs
f = open(read_path, "rb")
r = PdfFileReader(f)

# Force the redaction by merging with itself
page = r.getPage(0)
page.mergePage(r.getPage(0))

# Write out the result
f_out = open(write_path, "wb")
w = PdfFileWriter()
w.addPage(page)
w.write(f_out)
f_out.close()

Hi, I am trying to use your script but sadly I don't see any effect. I don't understand exactly where your EncodedStreamObject class is actually used. I guess it should produce its effect inside mergePage() method.

I tried to set a breakpoint inside __init__() and the debugger is not stopping anywhere so I think it has never been called inside the script execution.

Could you help me on it? All other resources on the web seems to be quite unuseful and/or outdated.

Thanks in advance
Andrea

Hi there Andrea (@ndricca)! The PyPDF2 package has changed quite a bit since I wrote this. It still works! But you need to use version 1.26.0 (pip install PyPDF2==1.26.0). I just tried it and it works using this version. :)

To confirm you have the right version do this before running the gist:

pip show PyPDF2

Name: PyPDF2
Version: 1.26.0 <--- Should be this
.
.
.

Hope that helps!

Hello,
to be able to use latest pypdf(2) version you'll have to change the overridding code to :

# Overload with redaction version
pypdf2.generic._data_structures.EncodedStreamObject = EncodedStreamObject
the class was in 'generic' and moved to 'generic._data_structures'

'+ some minor deprecated call changes

btw thanks for sharing.

Hi, upgrading to pypdf was quite straightforward, the code runs till the end but withouth any replacement.
I guess I am still not able to decode the text correctly.
By putting a breakpoint on line 21 above on text = replace_w_stars(text, phrase), I see the text variable has a value that starts like this:
'0.00000 0.00000 595.00000 842.00000 re W n'.
I tried changing the decoding bytes with different results but with no result. Do you have any suggestion?

You're replacing line 7 with the actual phrases you want to redact, right? I would check if you can find your phrases within the text variable at your breakpoint (e.g. assert phrases[0] in text).

Also, I would see if adding '595.0000' as a phrase, results in it being redacted (replaced with *) from your particular PDF, since that is a phrase which definitely does appear in the text variable based on your post.

Let me know the results and hopefully we can figure out why it's not working.

Thanks for your quick reply riley, what I am saying is that the content of the text variable is encoded or maybe encrypted.

Just a more comprehensive example:

1. I created a .docx file, wrote "This is a very standard text." inside and saved it as PDF
2. If I apply extract_text() method to the page without overwriting class EncodedStreamObject, I correctly get 'This is a very standard text. \n '.
3. If I overwrite the class and place a breakpoint on line 21 above, the text variable has the following content:

/Span <</MCID 0/Lang (en-GB)>> BDC q
0.000008871 0 595.32 841.92 re
W* n
BT
/F1 11.04 Tf
1 0 0 1 56.64 760.68 Tm
/GS7 gs
0 g
/GS8 gs
0 G
[(T)] TJ
ET
Q
q
0.000008871 0 595.32 841.92 re
W* n
BT
/F1 11.04 Tf
1 0 0 1 62.04 760.68 Tm
0 g
0 G
[(h)3(is is a )8(v)-4(er)10(y)-3( )] TJ
ET
Q
q
0.000008871 0 595.32 841.92 re
W* n
BT
/F1 11.04 Tf
1 0 0 1 116.06 760.68 Tm
0 g
0 G
[(s)11(tand)5(ard)5( t)-3(ex)7(t.)] TJ
ET
Q
q
0.000008871 0 595.32 841.92 re
W* n
BT
/F1 11.04 Tf
1 0 0 1 178.46 760.68 Tm
0 g
0 G
[( )] TJ
ET
Q
 EMC  /Span <</MCID 1/Lang (en-GB)>> BDC q
0.000008871 0 595.32 841.92 re
W* n
BT
/F1 11.04 Tf
1 0 0 1 56.64 738.1 Tm
0 g
0 G
[( )] TJ
ET
Q
 EMC