Can't extract images for this PDF

[bbfrog]

Description of the bug

Monaleesa_full.pdf
Pymupdf can't extract images in page 2 and page 4 of this pdf.

How to reproduce the bug

import pymupdf
doc = pymupdf.open('Monaleesa_full.pdf')

page_num = 0
for page in doc:
  page_num += 1
  images = page.get_images(full=True)
  print(f'page {page_num}: {len(images)} images')

PyMuPDF version

1.24.11

Operating system

MacOS

Python version

3.12

[JorjMcKie]

Except for page 7 (0-based), none of the pages contains an image.
What you see are vector graphics - no images.

[JorjMcKie]

Vector graphics cannot be extracted. All you can do is making a "photo" of the respective page area ...

[bbfrog]

Acrobat API can extract the vector graphics and save as png or svg. How does it do this? Is it hard to support in Pymupdf? THanks!

[JorjMcKie]

You can try this script. Or do this:

import pymupdf

doc = pymupdf.open("input.pdf")
for page in doc:
    for i, bbox in enumerate(page.cluster_drawings()):
        pix = page.get_pixmap(clip=bbox, dpi=150)
        pix.save(f"{doc.name}-{page.number}-{i}.png")

[bbfrog]

Thanks @JorjMcKie very much. It works and can extract the image I want. But it also extracted tables from this pdf as drawing, is there any field can differentiate the tables with other drawing? Thanks!

[JorjMcKie]

No, the recognition of vector graphics is based on syntax algorithms only - not on a sematic understanding of what it is that is wrapped bey lines, etc.

If however you are lucky enough such that PyMuPDF can identify the tables as tables ... then you could check whether those table boundary boxes coincide with identified vector graphic boxes. And thus exclude them from picture conversion.

[bbfrog]

Thanks. Chatgpt works for identifying whether a image is table or plot:)