Only Convert PDFs with embedded OCRed text to EPUB?

[Geremia]

Is there a way to find only PDFs that have embedded text (such as OCRed PDFs) and only convert those PDFs to EPUB? Thanks

[theducks]

OCR'd is text.

Image only PDF's may need to be OCR'd.
So is your question: "how do I differentiate Text PDF's from Image PDF's?"

I would expect that a Image filled PDF file might be quite a bit larger.

[Geremia]

Originally Posted by theducks

OCR'd is text.

Image only PDF's may need to be OCR'd.
So is your question: "how do I differentiate Text PDF's from Image PDF's?"

Exactly!

Originally Posted by theducks

I would expect that a Image filled PDF file might be quite a bit larger.

Yes, that's certainly one way to guess, but I have text-only PDFs in my library that are 16 MB or more, and image ones that are less than that, so, since it seems Calibre can't do this itself, I wrote a BASH script in Linux that uses the command pdfimages:

#!/bin/bash

# This script will find all PDFs lacking images in a Calibre library
#
# Run it with this:
# find ~/Calibre\ Library/ -iname "*.pdf" -print0 | xargs -0 -I{} ./pdf_no_images.bash {} 2> /dev/null > "PDFs lacking images.txt"

images=`pdfimages -list "$1" | awk '{print $2}' | grep 0`

if [ -z "$images" ]; then
    echo "$1"
fi

It works well. The only problem is that I have to manually go through the "PDFs lacking images.txt" file and convert those PDFs to EPUB, but this would certainly be an easy task to code into a Calibre add-in. I should learn how to do that.

[dgatwood]

Better to use something like pdftotext and see if it returns nothing. PDF files might contain both images *and* text, and I'm assuming you probably want to convert those as well.

[Geremia]

Originally Posted by dgatwood

Better to use something like pdftotext and see if it returns nothing. PDF files might contain both images *and* text, and I'm assuming you probably want to convert those as well.

Oh yes, that's a good point. What I wrote above only finds pure-text PDFs, not mixed text/image ones like the PDFs from Archive.org. I don't think I have many, if any, mixed text/image PDFs, but all my DJVU books are that way. PDFs from Google Books or HathiTrust are mostly images, but they do have a small amount of text for copyright, etc., so making a script ignore that would be more complex.