Problem with OCR

[SaeidN]

Hi,

I almost have read all the topics about OCR. I can say all of the topics are for 2009 and before.

Some of them doesn't work at all for me, some works but gives error, and one of them was I think just for learning fonts (then it shows words in a small part of screen in 70 sec which is so long), another one needed office 2003 (MODI), I installed it and didn't work again.

The letters that I need OCR to read has different backgrounds.

Can anyone please show me an OCR which works great with no problem?

Also is there any way to take a snapshot of all letters (26 letters and 10 numbers), and save them as image, and then tell OCR that which one matches which one?

Thank you

[MattHiggs]

Have you tried tesseract ocr?  I have been able to use it, both when installed using the built executable and when running it through Cygwin without issue.

https://github.com/tesseract-ocr/tesseract/wiki

Edited October 21, 2016 by MattHiggs

[SaeidN]

6 hours ago, MattHiggs said:

Have you tried tesseract ocr?  I have been able to use it, both when installed using the built executable and when running it through Cygwin without issue.

https://github.com/tesseract-ocr/tesseract/wiki

Yes, I tried it, and gives error.

I found this post and it worked.

Is there any way to tell the script in that post, to look for all white pixel with 10 shades variation (like in pixelsearch parameter) and then convert them to characters.

All I want it to read is only A-Z, a-z, and 0-9

I appreciate if you would answer to this, too: 

Quote

Also is there any way to take a snapshot of all letters (26 letters and 10 numbers), and save them as image, and then tell OCR that which one matches which one?

 

[LarsJ]

Here are a few posts you can study: Tesseract Simple Example, Getting Tesseract to Work and Tesseract doesnt detect the easiest image.

Pixels with 10 shades variation: I don't think you can do this. Tesseract just finds all text in the image.

26 letters and 10 digits: I don't think you can do it and it should not be necessary. Tesseract is able to recognize letters and digits.

[AndyG]

On 22.10.2016 at 3:01 AM, SaeidN said:

Is there any way to tell the script in that post, to look for all white pixel with 10 shades variation (like in pixelsearch parameter) and then convert them to characters.

Absolutely wrong way....

Think some seconds about what you would perceive "better" if YOU would be an OCR-Software!

First, a "picture" with a load of "shi*" informations aka shaded "pixels", or second, a "preprocessed" (aka nicetoOCRing) picture.

If you are serious, you would prefer a preprocessed and easytoOCR picture. So why not make the "picture" better to "read" instead of force the OCR-software to calculate letters from shi*?

Btw. it would be very helpful if you could show one/some of those pictures you want to OCR

[BrewManNH]

This isn't for solving CAPTCHAs is it?