feat: Add option to detect scanned PDFs with textlayers and run ocr conditionally

[striepemaarten]

What is the proposed feature?

Add some sort of flag/guess score for PDFs indicating if they are result of a scan.

Perhabs it could be done based on some factors like fullscreen images being present on all pages in combination with the already present quality score for the textlayer.

I am dealing with some scanned pdfs where the scanner includes a hiddden, poor quality textlayer.
The issue i am facing with those, is that i have no way to tell beforehand.

force_ocr=true produces much better ouput but on clean pdfs results are obviosuly worse.

Why would this be a good addition?

It would prevent the need for additional pdf analytics steps outside.
Xberg already has all metdata, etc. of the pdf file.
force_ocr=true seems all or nothing

[mits87]

I'm struggling with the same, as far as I understand the ocr.vlm_fallback should works like that but unfortunately it doesn't

[Goldziher]

@tobocop2 is taking over