How to easy process images previous to OCR?

[Disonantemus]

Problem

I have many documents with images that contain raster images with useful information (like many kinds of charts: organizational, flow, statistics, etc), but images are very different, with many colors/sizes/quality, that get bad results with OCR.

Workaround

Right now, I do:

• Extract the images with pdfimages -all in.pdf _
• Find the correct image
• Process with imagemagick, Example: remove all colors but black (text color) very useful to clean charts/boxes with colors/textures that don't get OCRed): magick in.png -alpha remove -fill white -fuzz 70% +opaque "#000000" -colorspace gray out.png
• OCR with ocrmypdf

I do get a lot better OCR from that (better than use ocrmypdf flags process images).

PROS:

• OCR is better.

CONS:

• Too many steps.
• This is annoying.
• PDF looks worst: the image shown is the one that gets OCRed, not the original one.

Alternative

I know is possible to do a plugin, because #923 , but I don't know how to do that. I'm not a programmer.

Question

There is an easy way to process image(s) inside a PDF with imagemagick (CLI, no python) to manual improve OCR in hard cases?

[fahmiridho07]

There is no built-in ImageMagick hook via CLI only — custom preprocessing needs a plugin (your link #923 is correct), which does require a little Python.

But for your case (colored charts / boxes / textures) you can usually skip the manual pdfimages + magick roundtrip with the built-in flags:

ocrmypdf --remove-background --deskew --clean --oversample 300 --force-ocr in.pdf out.pdf

What each does:

• --remove-background: drops colored backgrounds, similar to your +opaque black step
• --deskew: straightens slightly rotated scans
• --clean: light despeckle
• --oversample 300: re-render at 300 DPI before OCR, helps a lot on low-res charts
• --force-ocr: re-OCR even pages that already have text

Two notes on your CONS:

1. Image quality in output: OCRmyPDF by default keeps the original image for display and adds an invisible OCR layer. If your output looks worse, it is usually --remove-background / --clean rewriting the visible image. Add --output-type pdf and compare with/without --optimize to keep the original appearance closer.
2. If you only need the text, add --sidecar out.txt to get the OCR text without touching the PDF appearance at all.

If those flags still do not match your manual magick result, the plugin route is the only way to inject an arbitrary magick command per page. Minimal example is in docs/plugins. Happy to help adapt your magick ... +opaque line into a 10-line plugin if you want.