I have a very low resolution, almost pixelated images for processing. The characters can be recognized if zoomed in and a quick eye scan is performed with the aid of basic English grammar and vocabulary set. I have a tried many ICRs and OCRs, but none proved to be useful. Is/Are there any OCR/s that I might have possibly missed on?enter image description here P.S.The cursive font seems to be the problem.

  • 4
    And its color. First thing I'd try is converting it to b/w, increasing contrast, and "blowing it up" to 4+ times its size. The latter might sound strange, but for me increased recognition rate remarkably in several cases. As for the software question: What OS should it run on, and what is your price limit in case of paid products? – Izzy Mar 22 '17 at 11:25
  • If you have access to a photocopier or image manipulation program, try reversing the image, e.g. make it negative (black to white and white to black). I sometimes scan in till receipts that have been printed on thermal paper which over time lose their clarity and this has helped to at least make them readable to the human eye. The font, as you say, might also be an issue – essexboyracer Apr 22 '17 at 18:01

Tesseract 4.0 is probably your best bet for getting accurate results from images like that.

It's open source and is available for Linux (As well as binaries for Windows).

I suggest giving it a shot with the image resized to about 300DPI and in black and white, as per Izzy's Comment.

If the results aren't as accurate as you would like, you could always Train the Neural Network for that image so you can get a more accurate OCR for future images of that relative type.

  • How do I resize an image to 300 dpi? – Dhruva Mar 24 '17 at 4:51
  • Is there an ocr trainer that I can use? – Dhruva Mar 24 '17 at 4:52
  • @Dhruva To resize an image to 300DPI, just use your favorite image editor and there should be an option to scale or resize the image/canvas to a specified DPI (I use GIMP). For training the Neural Network, you'd have to look at the linked documentation and go from there (I can't help you much with that). I would only train for that specific typeface if you do not receive an accurate result from the OCR in the first place AND have more images with that same typeface that you will want to OCR later. – MrPublic Mar 24 '17 at 16:22

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.