Text with background missing from output

[rsaite]

Referencing #225 which was incorrectly marked as fixed even though the bug still persists.

PDF for reproduction: https://bmjopensem.bmj.com/content/bmjosem/1/1/e000050.full.pdf

Consider the box on the middle right of the first page with blue and grey background ("Strenghts and limitations of this study * Free weight resistance training [..]"). That text is embedded in the pdf and can be extracted, e.g. using PyMuPDF. However, when doing a simple markdown conversion with PyMuPDF4LLM, that text is missing from the output.

#225 was closed because the bug was supposedly fixed with release 0.0.18. However, the bug can still be observed in release 0.0.18 and also in the current release 0.0.24.

[rsaite]

@JorjMcKie Is there any feedback on this issue? Any idea where this bug might come from and what I should consider if I wanted to fix this myself?

[JorjMcKie]

The problems as it happens on first page is that that text box with shaded background is classified as a "significant" vector graphic - which it isn't. Have a fix here.

You can get full visible output if using write_images=True, but the text will not be extracted.

Independently from all that, force_text=True (the default) should still extract the text. Need to look into this.

[JorjMcKie]

Mostly fixed this issue in v0.0.26 - by improving the detection of irrelevant background.
Improvement is still needed for forced text output.