https://github.com/hintak/pymupdf-jbig2-extract

Convert masks in pdf to images
https://github.com/hintak/pymupdf-jbig2-extract

Last synced: 17 days ago
JSON representation

Convert masks in pdf to images

• Host: GitHub
• URL: https://github.com/hintak/pymupdf-jbig2-extract
• Owner: HinTak
• Created: 2024-02-22T01:41:17.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2024-05-11T23:40:24.000Z (6 months ago)
• Last Synced: 2024-05-12T23:24:22.072Z (6 months ago)
• Language: Python
• Size: 7.81 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
Some pdf's (those from the Internet Archive, apparently) consist of mostly

scans. Structurally, every page is a background image with a mask, occasionally with

an invisible OCR text layer, and sometimes a lower-res preview/thumbnail image.

This is mostly a script trying to convert the mask into an image losslessly.

The background image (often just yellowing paper) makes the document less-readable.

Smaller pdf with single image per page (without requiring compositing) also loads faster, too.

The final `extract-all-mask.py` script keeps the original front and back cover, but converts every

page from 2 to N-1 from image+mask to just the mask.

Out of about 400 such pdf's:

* There is one (also apparently the oldest by its creation/modification date)

which has this strange problem (indented and spaced for readability from original)

of every page of 308 refering to every image and mask. Hence the `page.clean_contents()` line

and `mutool clean -sg ...`.

```

4668 0 obj

<< /Type /Page

   /Parent 924 0 R

   ...

   /Resources<<...

               /XObject<>

               >>

   /MediaBox[0 0 487 832]

   /StructParents 0

  >>

```

* There is one other pdf which seems to have (blank) pages with mask only.

* The script under `gist.github.com` has been modified in small ways. The original can be found at that location.

  (migrated to python 3, and changing resolution and color inversion; see below).

* For about 90% (i.e. about 40) of the 400, the heuristics (some masks have a "DeviceGray" color space name, and need to

be inverted to be black-text-on-white-background. Such images also have resolution 360 dpi instead of 300 dpi) seems to work.

But 5 pdfs appear white-text-on-black-background, and about 40 have the wrong resolution with values 150dpi, 300dpi, 350 dpi, 360dpi, 500dpi,

and ~642dpi being seen. Script `convert-page2.py` is intended to be run on the result pdf to quickly see the color of page 2 of result.

`resolution-page2.py` is intended to be run on the source pdf quickly to tell the resolution, plus the

every-page-references-every-image and mask-only-page problems.