Can Mathematical Symbols Be Extracted from Scanned Journal Pages?

[qnach]

TL;DR

Many ancient journals were scanned into PDF files. What I don't know is how could these pages become mark-able and search-able?

They should only be images.

Many ancient journals were scanned into PDF files. What I don't know is how could these pages become mark-able and search-able?
They should only be images.

 

[pbuk]

Optical character recognition (OCR).

 

[jedishrfu]

In addition, PDF tools may allow for edits of the searchable text when OCR fails to recognize some scripts.

 

[sophiecentaur]

Optical character recognition (OCR).

It works quite well these days but I read a lot of books (obviously scanned) on my Kindle and there are one or two mistakes in most of the books. Context usually digs me out of the problem but the Maths in some papers could produce undetectable errors (undetectable by my Maths brain, at least).

 

[sophiecentaur]

In addition, PDF tools may allow for edits of the searchable text when OCR fails to recognize some scripts.

That sounds v intelligent. You mean an improved bolt on when the context reads as garbage? It's a matter of spotting an error in the first place.

 

[robphy]

PDF Editors like Adobe Acrobat (https://www.adobe.com/acrobat/) and ABBYY (https://pdf.abbyy.com/) can show the text layer. I'm sure there are tools for spell-checking and maybe even grammar-testing that layer.

There are also tools that can extract the whole text layer into a text file (e.g. https://en.wikipedia.org/wiki/Pdftotext ).

A new technology is trying to extract mathematical symbols from handwriting and from scanned images
(e.g.
https://mathpix.com/
https://photomath.com/en/
https://webdemo.myscript.com/
https://socratic.org/ (from Google)
http://write-math.com/
https://www.i2ocr.com/free-online-math-equation-ocr
https://www.cs.rit.edu/~dprl/software.html
https://www.xthink.com/mathjournal.html (once promising)
https://www.inftyproject.org/en/software.html InftyReader
)

Maybe AI can help learn the appropriate context to improve recognition
https://www.searchonmath.com/
https://approach0.xyz/search/
https://mathdeck.org/I haven't tried all of these sites.

 

[sophiecentaur]

The weak link is in the OCR though. OCR utilities can spot text and non-text but some old papers give OCR a hard time. I should imagine any system might need to ask a human for advice. This could involve a lot of (specialist) man-hours for millions of documents.

 

[sophiecentaur]

A new technology is trying to extract mathematical symbols from handwriting and from scanned images

That could be very demanding. It could involve parsing an equation and working out its meaning from the context. Sort of thing that only an expert in the field of the paper could do. But never say never, about computing.