PyMuPDF extracts invisible characters

[sergenti]

🐛 Describe the bug

Hey, it seems that in one of my docs, the library is extracting invisible characters/characters of the same colors as the background.
Is there a way to parse ONLY visible chars / all characters that have a different color from the BG?

👨‍🔬 To Reproduce

I'd love to upload the file as well but it's not for general distribution.
If you give me a way to check if it's the former or the latter, I'll gladly check out and tell you more!

🖼️ Screenshots

green area (highlighted)

⚙️ Configuration

Linux - Ubuntu 22.04.2 LTS (Jammy Jellyfish)
3.10.6 (main, Nov 14 2022, 16:10:14) [GCC 11.3.0]
PyMuPDF 1.21.1: Python bindings for the MuPDF 1.21.1 library.
Version date: 2022-12-13 00:00:01.
Built for Python 3.10 on linux (64-bit).

[JorjMcKie]

This is typical "Discussions" item - no bug. So let me transfer this first.

[JorjMcKie]

The concept of "invisible text" may be quite complex.

• PDF directly supports invisibility via the Tr operator (text rendering mode). This is widely used by OCR engines.
• Text may simply be "behind" other objects like (intransparent) images or drawings (vector graphics).
• Text may be covered by an annotation - which in some sense is a "temporary" (= deletable) object.
• Even if text is underneath an image, that image may be transparent ... so what's the answer? Or a drawing / annotation uses a "blend mode" that keeps objects of a darker color underneath visible although it is per se intransparent.
• Your case: a snow rabbit in Norway.

So there is no simple way tro answer the question "Is this text / image / other object visible?"

But PyMuPDF lets you determine if a given rectangle is unicolor. Once you have extracted the bbox of text, make a Pixmap of that bbox and check if it is unicolor - or even if the dominant color is present in more than x% of all pixels.

# bbox is the text boundary box
pix = page.get_pixmap(clip=bbox)
if pix.is_unicolor:
    print(f"text is invisible!")
# the following method even returns the most used color:
pix.color_topusage()