Author Topic: Scanning documents and character recognition  (Read 1346 times)

Scanning documents and character recognition
« on: 22 October, 2023, 06:36:03 pm »
Posting on behalf of Lynn... 
Looking to purchase (council, not me) some effective software/solutions for ocr of documents, scanned or e-copies. We tried various ones before but often they’re not brilliant.
End use would be for braille conversion (we use Duxbury) or large print.
Something that has the capability to pick up on mathematical/scientific formula would be amazing although I expect this doesn’t happen.
I think the newer versions of Duxbury does convert equation editor input from Microsoft but I need to check it out. But scanned documents I’ve never seen anything that works.
We used omnipage once but I wasn’t impressed.
Any ideas?


barakta

  • Bastard lovechild of Yomiko Readman and Johnny 5
Re: Scanning documents and character recognition
« Reply #1 on: 22 October, 2023, 10:17:11 pm »
I wish I knew the answer, most software is a tad cranky and takes a lot of training/expertise to use and the output usually needs tidying up.

I haven't seen an update on equations/maths being doable outside of LaTeX markup language but I am a wee bit out of date.

The council could try asking AbilityNet or Thomas Pocklington Trust for guidance as both of those do assistive tech and IT stuff. They might need to pay for this advice tho.

Re: Scanning documents and character recognition
« Reply #2 on: 23 October, 2023, 09:44:27 pm »
Thanks, good advice have passed your reply onto Lynn

barakta

  • Bastard lovechild of Yomiko Readman and Johnny 5
Re: Scanning documents and character recognition
« Reply #3 on: 23 October, 2023, 11:23:18 pm »
Hope they have success in a solution and the council is willing to invest appropriately = usually orgs refuse to put the tech or humans onto it...

fruitcake

  • some kind of fruitcake
Re: Scanning documents and character recognition
« Reply #4 on: 28 October, 2023, 08:01:21 pm »
My brief experiments with the OCR tools in Linux taught me that the quality of the source image makes a surprising difference - plenty of contrast, a plain background without colour gradient or shadows, and not oversharpened. Free OCR tools I tried really weren't great with text in columns or text boxes.

I'm expecting OCR to get much better very soon with the application of Machine Learning. I might be tempted to try GPT as an OCR engine, though not with anything confidential because who knows what happens with the data you offer up. There's lots of things ChatGPT is thought to be suitable for that it's not; optical character recognition may be one thing it's very good at.

Google is already onto this application: https://cloud.google.com/use-cases/ocr  I haven't used it. The usual caveats apply: don't use it for anything you wouldn't want that network to retain and to associate with your organisation.