4 ms·

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

iancmceachern 3y ago
It doesn't necessarily need to be AI, it could be a simple filter, etc
gield 3y ago
OCR has been around for a few decades by now: https://nl.wikipedia.org/wiki/Optical_character_recognition https://nl.wikipedia.org/wiki/Optical_character_recognition
- hbn 3y ago
  For better or for worse https://www.theverge.com/2013/8/6/4594482/xerox-copiers-randomly-replacing-numbers-in-documents https://www.theverge.com/2013/8/6/4594482/xerox-copiers-rand...
notjulianjaynes 3y ago
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.
- pbhjpbhj 3y ago
  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-expert https://chat.openai.com/g/g-LCtICWzCt-historical-handwriting... [2] https://www.transkribus.org/ https://www.transkribus.org/
- flir 3y ago
  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 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.
dan-robertson 3y ago
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.

iancmceachern 3y ago

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

gield 3y ago

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

hbn 3y ago
For better or for worse https://www.theverge.com/2013/8/6/4594482/xerox-copiers-randomly-replacing-numbers-in-documents https://www.theverge.com/2013/8/6/4594482/xerox-copiers-rand...

notjulianjaynes 3y ago

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.

pbhjpbhj 3y ago
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-expert https://chat.openai.com/g/g-LCtICWzCt-historical-handwriting... [2] https://www.transkribus.org/ https://www.transkribus.org/
flir 3y ago
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 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.

dan-robertson 3y ago

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.