Extracting filled polygons and saving as new pdf

[ddurgaprasad]

My task is to separate out straight lines, filled polygons, text into different PDFs for further analysis.
I am successful in extracting straight lines by using '_page.edges'. I presume edges are the ones with x0=x1 or y0=y1 . Now filled polygons are to be saved into separate pdfs. Is it possible to separate out the filled polygons and text ?
Sample.pdf
Also I noticed , out of 5 texts available in the attachment, pdfminer could extract only two correctly.In one particular case 8524 is extracted as 8254. Degree symbol ° is getting extracted as (cid:176). For this reason , I am thinking of separating out the text and make use of OCR.

[jsvine]

Hi @ddurgaprasad, sounds interesting!

I presume edges are the ones with x0=x1 or y0=y1

Essentially, yes. There's an exception, however: If pdfminer.six (the library powering pdfplumber) says that the object is a curve (rather than a line or rect), its edges — even if they are strictly vertical or strictly horizontal — will not be recognized as edges.

Is it possible to separate out the filled polygons and text ?

pdfplumber does not yet provide an automated way to detect polygons, but you can detect them yourself by inspecting each curve's points attribute. E.g., for your sample PDF, pdf.pages[0].curves[0] (the thinner of the two sloping lines) is this:

{'x0': Decimal('219.785'),
 'y0': Decimal('127.699'),
 'x1': Decimal('336.328'),
 'y1': Decimal('406.652'),
 'width': Decimal('116.543'),
 'height': Decimal('278.953'),
 'linewidth': Decimal('0'),
 'stroke': Decimal('0'),
 'fill': Decimal('1'),
 'evenodd': Decimal('1'),
 'stroking_color': None,
 'non_stroking_color': None,
 'object_type': 'curve',
 'page_number': 1,
 'points': [(Decimal('225.258'), Decimal('466.301')),
  (Decimal('336.328'), Decimal('189.578')),
  (Decimal('330.785'), Decimal('187.348')),
  (Decimal('219.785'), Decimal('464.141'))],
 'top': Decimal('187.348'),
 'bottom': Decimal('466.301'),
 'doctop': Decimal('187.348')}

[ddurgaprasad]

@jsvine ,thanks for clarifying about polygons. For extraction of text ,do you suggest any workaround? I am not sure why some characters are juxtaposed.

[jsvine]

I don't know if there's a workaround, unfortunately. For text extraction, pdfplumber depends on pdfminer.six, which in turn depends on the PDF being created correctly. There's some discussion of the issue here:

• pdfminer.six removes strings from .pdf file pdfminer/pdfminer.six#35
• Convert CIDs to Unicode HazyResearch/pdftotree#33

[jsvine]

Closing old issues. Feel free to reopen.