6 ms·

Show HN: Character Recognition with OpenCV

52 points by brakmic 11y ago

jozan 11y ago
This looks cool. I didn't try the examples but that receipt example is what I really need. I keep track of how much money do I spend on what things by going through receipts collected over a month. While the information is useful it's too much manual work. I wish I could just scan or take a picture of my receipts and extract text for further processing.
- archimedespi 11y ago
  Evernote can do that if you don't mind trusting them with your reciepts (I don't personally use this since I don't really trust Evernote that much).
- uberneo 11y ago
  You can try https://www.expensify.com/ https://www.expensify.com/ .. they have a really nice App and they are nice people as well .. i read their startup story once and it was really interesting
derwildemomo 11y ago
The title is wrong – it actually uses Tesseract to do the OCR. Call me negative, but I don't see the big leap here, it's just a simple demo to recognize text in an almost perfect input picture. It uses the Emgu CV, self description: "Emgu CV is a cross platform .Net wrapper to the OpenCV image processing library.".
- brakmic 11y ago
  Hi, Actually, the input picture is anything but "perfect". For example the German letters ö,ä,ü and 'ß' are often hard to scan. Of course it's a demo. That's the subtitle of the project. :) Kind regards,
  - derwildemomo 11y ago
    It's perfect from an OCR point of view: - Perfectly upright, no visible rotation - Sharp, not remotely blurry - No visible brightness/color gradient (as is often the case with pictures taken by mobile phones.) It contains umlauts (and i'm not sure whether the tesseract training data for german includes those, but i'm quite certain it does), but that doesnt disqualify the image. I was just saying hey, it's not using openCV but tesseract, so the title is wrong.. and i stand by that statement. I appreciate the work you do and i was exclusively referring to the title here. have a good day ;)
    - brakmic 11y ago
      I was learning a bit about OpenCV (bought a book about OpenCV with Python) and found EmguCV while surfing some "computer vision" pages. And because my company uses .NET heavily, I thought it would be cool to let it work with .NET. And this is it...just wanted to share my experiences. :) Maybe it can be of some use to others. Cheers, Harris
- pfarnsworth 11y ago
  I did this back in 2012, using Tesseract and OpenCV to scan letters from screenshots. I did it in python, though, but it was relatively straightforward. I actually had to create my own "language" to help with the processing, bringing it from the low 90s to ~98% accuracy. It doesn't diminish OP's work, though, nice job.
jeroen 11y ago
Is there any reason to use this over https://www.nuget.org/packages/Tesseract/ https://www.nuget.org/packages/Tesseract/ ?
- brakmic 11y ago
  A very cool alternative! Didn't know it exists. Well, I'm learning about OpenCV within Python's ecosystem so never checked anything besides EmguCV. Maybe I could write another demo with this Nuget! :) Regards,
- brakmic 11y ago
  So, I checked the code from Nuget's GitHub pages and at the first sight it seems more complex which doesn't have to be 'bad' but its simply more demanding than my small example. Of course, it also seems that you can do more sophisticated things with this Nuget. Regards,
z3t4 11y ago
This is interesting because I have a food project on ice, that got stuck on OCR'ing receipt's. There are proprietary scanners that are OK, but not perfect. There are still a lot of progress to be done is in this area!
scosman 11y ago
For anyone wanting to go deeper, this book on deep neural nets teaches them using character reco as the example. http://neuralnetworksanddeeplearning.com http://neuralnetworksanddeeplearning.com

52 points by brakmic 11y ago

jozan 11y ago

This looks cool. I didn't try the examples but that receipt example is what I really need. I keep track of how much money do I spend on what things by going through receipts collected over a month. While the information is useful it's too much manual work. I wish I could just scan or take a picture of my receipts and extract text for further processing.

archimedespi 11y ago
Evernote can do that if you don't mind trusting them with your reciepts (I don't personally use this since I don't really trust Evernote that much).
uberneo 11y ago
You can try https://www.expensify.com/ https://www.expensify.com/ .. they have a really nice App and they are nice people as well .. i read their startup story once and it was really interesting

derwildemomo 11y ago

The title is wrong – it actually uses Tesseract to do the OCR. Call me negative, but I don't see the big leap here, it's just a simple demo to recognize text in an almost perfect input picture. It uses the Emgu CV, self description: "Emgu CV is a cross platform .Net wrapper to the OpenCV image processing library.".

brakmic 11y ago
Hi, Actually, the input picture is anything but "perfect". For example the German letters ö,ä,ü and 'ß' are often hard to scan. Of course it's a demo. That's the subtitle of the project. :) Kind regards,
- derwildemomo 11y ago
  It's perfect from an OCR point of view: - Perfectly upright, no visible rotation - Sharp, not remotely blurry - No visible brightness/color gradient (as is often the case with pictures taken by mobile phones.) It contains umlauts (and i'm not sure whether the tesseract training data for german includes those, but i'm quite certain it does), but that doesnt disqualify the image. I was just saying hey, it's not using openCV but tesseract, so the title is wrong.. and i stand by that statement. I appreciate the work you do and i was exclusively referring to the title here. have a good day ;)
  - brakmic 11y ago
    I was learning a bit about OpenCV (bought a book about OpenCV with Python) and found EmguCV while surfing some "computer vision" pages. And because my company uses .NET heavily, I thought it would be cool to let it work with .NET. And this is it...just wanted to share my experiences. :) Maybe it can be of some use to others. Cheers, Harris
pfarnsworth 11y ago
I did this back in 2012, using Tesseract and OpenCV to scan letters from screenshots. I did it in python, though, but it was relatively straightforward. I actually had to create my own "language" to help with the processing, bringing it from the low 90s to ~98% accuracy. It doesn't diminish OP's work, though, nice job.

jeroen 11y ago

Is there any reason to use this over https://www.nuget.org/packages/Tesseract/ https://www.nuget.org/packages/Tesseract/ ?

brakmic 11y ago
A very cool alternative! Didn't know it exists. Well, I'm learning about OpenCV within Python's ecosystem so never checked anything besides EmguCV. Maybe I could write another demo with this Nuget! :) Regards,
brakmic 11y ago
So, I checked the code from Nuget's GitHub pages and at the first sight it seems more complex which doesn't have to be 'bad' but its simply more demanding than my small example. Of course, it also seems that you can do more sophisticated things with this Nuget. Regards,

z3t4 11y ago

This is interesting because I have a food project on ice, that got stuck on OCR'ing receipt's. There are proprietary scanners that are OK, but not perfect. There are still a lot of progress to be done is in this area!

scosman 11y ago

For anyone wanting to go deeper, this book on deep neural nets teaches them using character reco as the example. http://neuralnetworksanddeeplearning.com http://neuralnetworksanddeeplearning.com