4

I find text recognition (OCR) has gotten quite well for scanned papers. Such software needs to deal with tilting, noise, dust and dirt, but the high resolution of the scans (150 dpi or more) seems to sufficiently compensate that.

On the other hand, I often have screen shots that have perfect readability, no noise, no tilting etc. but the text recognition badly fails.

Here's an example:

Sample screenshot

Screenshot of a library description from Arduino IDE

This results in the following text when using text recognition of Microsoft OneNote:

OneWire 
Jim Studt, To 
Bertrik Sikken M r 
, a k Ken 
I—wiæ temperatuæ and Other Chips 
m Pollard, Robin Glenn Dangel, Lovato, Paul Scott Roberts, 
Love Nvstmm Vers 
Butcher, Roger Clark, 
23 5 INSTALLED 

That's barely helpful and probably needs more time to fix than a good typer needs to write it down from scratch.

Thus, I'm looking for a OCR software that is optimized for screenshots.

Requirements:

  • runs on Windows 10 x64
  • costs less than 10 €
  • takes an image as input and gives text as output
  • does a great job on small fonts that are typically used on screen (like Arial, Consolas, Calibri, Times, Courier, Mono Sans etc.)
  • works with screenshots in the clipboard (I don't want to save each screenshot before doing the OCR)
  • must support ASCII characters, typical source code characters (#<{[]}>;) and Umlaut characters (äöüÄÖÜß)
  • works offline (no Internet connection required)
  • has a GUI that I can work with
  • its main purpose is OCR. So it is more a small utility rather than some hundreds of MB software which has OCR as a side feature (so Microsoft OneNote would not qualify here)

Optional:

  • it may assume that the text is black, but should work with single color background

Non-requirements:

  • it does not need to perform well on high-resolution scanned documents
  • it does not need to support crazy fonts like brush script, ligature fonts (e.g. FiraCode), old fonts (e.g. Fraktura) or similar
1
  • Tesseract does a better job here, but it confuses 8 and B. I think a screenshot optimized version could still do better and correctly distinguish 8 and B. Also, Tesseract does not have a GUI. May 28 '20 at 9:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.