Looking for more straightforward way to extract svg

[Detsudetsu]

I'm trying to extract drawings from a PDF file and save them as SVG.
Thanks to a previous work found in here, I managed, but the implementation feels a bit roundabout to me.
What my current implementation below does is extract drawings, get their rect size, crop the source pdf to that size, and finally get_svg_image; such a long way!

I want to find any better way than that. Any ideas?

import fitz
import sys
from pathlib import Path

doc = fitz.open(sys.argv[1])
path = Path(sys.argv[1])
dst_dir = Path(path).with_suffix("").with_stem(path.stem + "-drawings")
dst_dir.mkdir()
out = fitz.open()
for page in doc:
    new_rects = []  # resulting rectangles

    for p in page.get_drawings():
        w = p["width"]
        r = p["rect"] + (-0 * w, -0 * w, 0 * w, 0 * w)  # enlarge each rectangle by width value
        for i in range(len(new_rects)):
            if abs(r & new_rects[i]) > 0:  # touching one of the new rects?
                new_rects[i] |= r  # enlarge it
                break
        # now look if contained in one of the new rects
        remainder = [s for s in new_rects if r in s]
        if remainder == []:  # no ==> add this rect to new rects
            new_rects.append(r)

    new_rects = list(set(new_rects))  # remove any duplicates
    new_rects.sort(key=lambda r: abs(r), reverse=True)
    remove = []
    for j in range(len(new_rects)):
        for i in range(len(new_rects)):
            if new_rects[j] in new_rects[i] and i != j:
                remove.append(j)
    remove = list(set(remove))
    remove.sort(reverse=True)
    for i in remove:
        del new_rects[i]
    new_rects.sort(key=lambda r: (r.tl.y, r.tl.x))  # sort by location

    for i, r in enumerate(new_rects):
        if r.height <= 5 or r.width <= 5:
            continue  # skip lines and empty rects

        newpage = out.new_page(-1, r.width, r.height)
        newpage.show_pdf_page(page.rect, doc, page.number, clip=r)
        svg = newpage.get_svg_image()
        file_name = "drawing-%03i-%02i.svg" % (page.number, i)
        dst_path = dst_dir / file_name
        dst_path.write_text(svg)

        print("saved: ", dst_path)

[JorjMcKie]

I actually like your code 😎
The most complex part is identifying the relevant drawing rectangles, removing containments, duplicates and all that.
I see no way how to simplify this.

But you probably need no dummy out PDF and those newpages:
Why not setting the CropBox of the original page to the identified rectangle and directly make an SVG from the original page?

    ... 
    page.set_cropbox(r)
    svg = page.get_svg_image()
    ...

[Detsudetsu]

I'd like to accept your suggestion, which makes the code much clearer!
But let me share a minor issue I found.

I found that some applications wrongly display even outside of the identified rectangle of the SVG created via the CropBox method, but not via the newpage one.
That's because what the former method does is just set viewPort and viewBox, so it doesn't do any clipping actually.
On the other hand, the latter method creates a <clipPath> element corresponding to the rectangle besides what the former does.

For example, the following screenshot shows SVG preview presented by PowerToys File Explorer Add-ons.
We can see the previews of rectangles (not a square) include irrelevant stuff in the upper directory, containing SVG created via the CropBox method.

In that cases, I have to clip the page, not crop.
So show_pdf_page() is still an option, perhaps?

sample.pdf

[JorjMcKie]

Ah, ok, interesting. Didn't know you tried that already. You must be a PyMuPDF power user 😉.
Yes, the show_pdf_page() is the only option to "set" the MediaBox of a page.
I tried setting TrimBox, ArtBox and BleedBox to the value of CropBox after updating the latter (via doc.xref_set_key()), but the SVG sources contain a lot of unexpected (text) stuff in any case - don't know if they would show up under PowerToys (haven't installed that one) when all the other PDF boxes also have been set ...

[Detsudetsu]

Alright now I tried setting TrimBox, ArtBox and BleedBox and yet it didn't work so might as well go with show_pdf_page().
I really appreciate your help!

[Detsudetsu]

By the way, I did a little experiment to find which bounding box (like MediaBox) can clip as I wanted.
I added the following snippet to the very first code.

for page in out:
    xref = out.page_xref(page.number)
    print("page:", page.number)
    for key in doc.xref_get_keys(xref):
        print(key, "=" , doc.xref_get_key(xref, key))

Then ran the code with the sample.pdf that was attached in my previous comment (containing 3 blue quadrangles)
The result was below.

page: 0
Filter = ('name', '/FlateDecode')
Length = ('int', '3106')
page: 1
Type = ('name', '/ObjStm')
N = ('int', '24')
First = ('int', '172')
Filter = ('name', '/FlateDecode')
Length = ('int', '551')
page: 2
P = ('xref', '24 0 R')
S = ('name', '/P')
Type = ('name', '/StructElem')
K = ('array', '[27 0 R 28 0 R]')
Pg = ('xref', '3 0 R')

It's chaotic 🤯
How can they be so different?

[JorjMcKie]

It's chaotic 🤯
How can they be so different?

Don't know what you have been doing ... 🤔
When I use this script, all works fine and looks decent:

import fitz

doc = fitz.open("sample.pdf")
page = doc[0]
paths = page.get_drawings()
src = fitz.open()
for i, path in enumerate(paths):
    r = path["rect"]
    spage = src.new_page(width=r.width, height=r.height)
    spage.show_pdf_page(spage.rect, doc, page.number, clip=r)
    svg = spage.get_svg_image(text_as_path=False)
    out = open("%i-new.svg" % i, "w")
    out.write(svg)
    out.close()

for page in src:
    keys = src.xref_get_keys(page.xref)
    print("\nTempfile page", page.number, "\n")
    for k in keys:
        print(k, src.xref_get_key(page.xref, k))

Result:

Tempfile page 0 

Type ('name', '/Page')
MediaBox ('array', '[0 0 147.4 147.40003]')
Rotate ('int', '0')
Resources ('xref', '3 0 R')
Parent ('xref', '2 0 R')
Contents ('array', '[20 0 R]')

Tempfile page 1 

Type ('name', '/Page')
MediaBox ('array', '[0 0 147.4 73.70001]')
Rotate ('int', '0')
Resources ('xref', '21 0 R')
Parent ('xref', '2 0 R')
Contents ('array', '[24 0 R]')

Tempfile page 2 

Type ('name', '/Page')
MediaBox ('array', '[0 0 73.69999 147.39997]')
Rotate ('int', '0')
Resources ('xref', '25 0 R')
Parent ('xref', '2 0 R')
Contents ('array', '[28 0 R]')

[Detsudetsu]

Oops I've found it 😑
I should have used out.xref_get_key() instead of doc.xref_get_key().
And I've got the same result as yours.

Thank you for your help, and sorry for the late reply.