Hacker News Reader

Comment by miohtama | Hacker News Reader

Unrelated to excellent article content, do we have today an AI solution to enhance the quality and readability of scanned PDFs?

View on HN

OCR has been around for a few decades by now: https://nl.wikipedia.org/wiki/Optical_character_recognition

For better or for worse

https://www.theverge.com/2013/8/6/4594482/xerox-copiers-rand...

I think the solution here is not so much AI as finding some library which has the magazine and making new scans. That would also help with the hard-to-solve-with-ai problem of the missing page.

It doesn't necessarily need to be AI, it could be a simple filter, etc

I'd also be curious about this as I am presently dealing with some PDFs with both handwriting and poorly scanned type. I can read and understand these documents but the text is not recognized using tesseract OCR.

I've been looking at handwriting recognition offerings and signed up for ChatGPT [1] to trial access for this task against Irish genealogical records, which it does quite well on. However, ChatGPT itself recommended that I use Transkribus [2]. That looks to be quite an esoteric interface, but my initial test suggests it's pretty good -- I've not made a proper comparison yet as I'm wanting to have a workflow of downloading images, transcribing them and maybe compiling them into a PDF with manual corrections.

[1] an 'expert' like this but it wasn't this one I used, can't locate it right now as am on a different computer, https://chat.openai.com/g/g-LCtICWzCt-historical-handwriting...

[2] https://www.transkribus.org/

My use-case was easier than yours (OCRing scans of microfiche of typewritten pages) but I found this https://github.com/Stirling-Tools/Stirling-PDF incredibly handy.

I still had to write a bit of Python, mostly for levelling and centering the text blocks, but it really is a PDF swiss army knife.

Reply on news.ycombinator.com