search

justadudewhohacks / opencv4nodejs

Nodejs bindings to OpenCV 3 and OpenCV 4
MIT License
4.97k stars 827 forks source link

How about the Scene Text Recognition library? #38

Closed JonesWest closed 7 years ago

JonesWest commented 7 years ago

Hi,

Thanks for this library :)

Do you know if you're going to support the Scene Text Recognition any time soon? We need to extract and decode text (or hex numbers) from a webcam picture.

justadudewhohacks commented 7 years ago

Are you talking about https://docs.opencv.org/trunk/d8/df2/group__text__recognize.html?

I can add the bindings, which are independent of tesseract or are you asking for the OCRTesseract detector specifically?

JonesWest commented 7 years ago

Yes, that was exactly what I was talking about :)

To be honest, I am not sure that I understand the difference (I am fairly new to OpenCV) - but I need to detect and decode hex numbers that are printed on boxes?

I tried other libs like: https://github.com/desmondmorris/node-tesseract but it doesn't perform very well because of its language dependencies (I think). So I would like to do it all in OpenCV instead (also because the application needs OpenCV anyway).

justadudewhohacks commented 7 years ago

I am not familar with the text recognition module of OpenCV either, but It seems there is an HMM classifier and a tesseract one.

I can implement the bindings to the HMM one, not sure if this one will give you better results than node-tesseract, but you can try it out.

Implementing the tesseract classifier would mean adding the dependency of a third party lib (tesseract) to the package, which I rather would not do.

carlosen14 commented 7 years ago

I don't know much about bindings, but i know they use tesseract i did use that software weeks ago, it was kind of trikky to make all work.. Maybe it give you an idea..

https://github.com/openalpr/openalpr

JonesWest commented 7 years ago

@carlosen14 Thanks. That could be useful as well.

@justadudewhohacks Would be great! Thanks 👍 :)

justadudewhohacks commented 7 years ago

I added some stuff from the contrib text module and also two basic usage examples for character recognition and word recognition from the trained BeamSearchDecoder CNN model. The features will be available when I published v2.14.

Good luck playing around with it, let me know if this was useful for your project =)

justadudewhohacks commented 7 years ago

New version is now published.