Inline images extraction

[maiiabocharova]

There are some small pictures in my PDFs and I am not able to extract them with page.getImageList() (it returns the empty list). Can you please recommend me what I can do to extract those images, maybe another library? I need those images very much)

[JorjMcKie]

If this happens, you

1. either are not looking at images, but at so-called drawings
2. or are looking at PDF "inline" images.

Drawings can be extracted via page.get_drawings() / page.get_cdrawings(). This is a list of dictionaries. Each dict represents a number of interconnected elementary draw commands: lines, curves, rectangles or quads.

Inline images are only contained in the internal page command source (the /Contents object(s)) - nowhere else. Method page.get_text("dict") (or "rawdict") contains all images on page - inline and others. Each image occurence occurs as many times as it is shown.

[maiiabocharova]

Thank you for the reply!
In the page.get_text("dict") I am getting the thing which I need like text "\nd\n~\n\uf093\n",
page.get_cdrawings() gives me something, but can I make a png or get numpy array out of it? I want to do a template matching afterwards.

[JorjMcKie]

page.get_text("dict")["blocks"] is a list of blocks on the page. Each one is either a text block or an image block - see documentation for TextPage. Image blocks have block["type"] = 1. The image binary is contained in block["image"]. More info is contained in the other dict keys.

The list of drawings dicts page.get_drawings() can be used to re-draw each on some other page - see the docu here.
Each "path" dict therein has a path["rect"], which is the rectangle containing all the elementary draws in it.
You could also do a page.get_pixmap(..., clip=path["rect"]) to create an image of the path. Of course there is the risk, that other things (not belonging to the path) are also part of that picture by accident.

[JorjMcKie]

😎 They are no image whatsoever, they are no drawings, so I tried to select them with my cursor using a PDF viewer, and voilà: they are text!!!
Then I looked at the used fonts (page = doc[0])

>>> pprint(page.get_fonts())
[(47, 'n/a', 'Type1', 'Times-BoldItalic', 'R10', ''),
 (48, 'n/a', 'Type1', 'Helvetica', 'R11', ''),
 (49, 'ttf', 'TrueType', 'VGAVKN+DRSymb21', 'R12', ''),
 (51, 'n/a', 'Type1', 'Times-Roman', 'R14', ''),
 (45, 'n/a', 'Type1', 'Courier', 'R8', ''),
 (46, 'n/a', 'Type1', 'Times-Italic', 'R9', '')]

so these are probably characters from the DRSymb21 font. Therefore I did this:

>>> spans=[]
>>> for b in blocks:
	for l in b["lines"]:
		for s in l["spans"]:
			if s["font"] != "DRSymb21": continue
			spans.append(s)
>>> for i,s in enumerate(spans):
	pix=page.get_pixmap(clip=s["bbox"])
	pix.save("symb-%i.png" % i)

This gave me a number of PNGs, some of them the same or nonsense, but the above were among them:

[maiiabocharova]

This is so smart, would have never guessed, that there is such a solution!

Thank you!!! Just can't express how I am grateful!

[maiiabocharova]

I am so sorry to bother you once again, but it seems that other PDF documents which have the pictograms sometimes they are not not even text:

page = pdf_doc[1].get_textpage().extractDICT()

spans = [span for block in page['blocks'] 
         for line in block['lines'] for span in line["spans"]]
sorted(spans, key = lambda x: (x["bbox"][1], x["bbox"][0]))

I did this and went through all the spans near the position which I need and the pictogram was not there (In the previous document with which you helped me text corresponding to the span which contained pictogram was "\uf093",'\uf07f' -- so unicode). But for example with this PDF (second page)
it's not the case (I tried extracting images -- page.getImageList() extracted the logo of the company. but not the pictogram, drawings and text also have not extracted those pictograms)

This is such a complicated issue...

[JorjMcKie]

this time the PDF page contains images, 257 (!!!) in total. Only one is reachable via an xref (xref 9). This is how to show them:

>>> page.clean_contents()  # required for this page: sloppy PDF creator, read the docu
>>> img=page.get_image_info(xrefs=True)
>>> len(img)
257
>>> # going to draw rects around each image, use a shape because of the large number
>>> shape=page.new_shape()
>>> for i in img:
	shape.draw_rect(i["bbox"])

	
Point(30.479999542236328, 56.64000701904297)
Point(78.95999908447266, 306.7200012207031)
Point(78.72000122070312, 306.96002197265625)
... and so on 257 times ...

>>> shape.finish(color=(0,1,0))  # surround all boxes with green border
>>> shape.commit()  # write shape to page
>>> doc.ez_save("x.pdf")  # save under new name
>>> 

Gives you this:

So except for the logo, all those images are just single 1 or 2 pixels to define your pictograms. An expert at work, obviously!

[maiiabocharova]

Thank you so much, so I'll need to look at some other approaches to match pictograms with the templates which I have, definitely those are not to be extracted with standard approaches.

Thank you for spending time helping me!