How can pdfplumber process pdf on char level

[NewUserHa]

Here I have a PDF with non-Unicode custom font (issue 1), and it has the wrong bbox (like 2 times than real) that already overlaps with each character (issue 2).

Currently, I'm using OCR the characters from each font to get correct maps:

to translate them (issue 1):

• in pdfminer.six, I achieved via high_level.extract_pages() + walking each object recursive via custom function to match to replace to the correct ones.
  (or can also use PDFPageInterpreter with a custom pdfminer.converter.TextConverter class with an override render_char() function)
• in pdfplumber, I didn't find the ability.

• is there an easier way in pdfplumber?

to resolve the overlaps for adding space between characters (issue 2):

• in pdfminer.six, I tried to divide the width and height of LTChar by 2 or 3 in that custom render_char() with self.cur_item._objs[-1] to try to trigger LAParams(word_margin=0.1, all_texts=1), but it didn't work.
• in pdfplumber, x_tolerance=-2 worked, but it added space between every character.
  I'm not familiar with how the LAParams algorithm works, but in its code, it seems that it only adds space in word_margin level, and char_margin is only for group texts.

• since both didn't work. how can solve the space issue easily?

Thanks

[jsvine]

Hi @NewUserHa. I'm not sure I fully understand the questions here; it might help to provide the PDF, the code, and/or screenshots. But at least for the first part, you might try something like this:

def adjust_bbox(char):
  [your custom logic here]
  return new_char

page = pdf.pages[0]
new_chars = list(map(adjust_bbox, page.chars))
text = pdfplumber.utils.extract_text(new_chars)

[NewUserHa]

Thanks for your reply.

pdfplumber.utils.extract_text looks great, but seems it's not in the document.
But I haven't tried it so not sure if it will work.

there's the pdf: WS_T 483.8-2016.pdf
only the last page is the one needs to be looked at,
the issues of the pdf are:

• used a custom font with a not unicode custom mapping
• faked bboxs for some characters of some fonts.

[jsvine]

Thanks for providing that PDF, which helps clarify the question. Let us know if the approach suggested above works for you.

And, correct, pdfplumber.utils.extract_text is not in the documentation. It can be useful, but it's not the main method I'd suggest people to use. I could see adding some more information about it in an "Advanced" section of the documentation.