How to work with OCR to recognize photographs

Asked

Viewed 2,508 times

7

I am currently working on my Course Completion Work and the theme involves OCR. I did some research and found the Tesseract, I also found libraries in Nuget for him (I’m using C#). The problem is that my work is about reading Vehicular Board, when I pass an image of a car (the ones that the radars take out), Tesseract can’t read the image, says that the image is invalid. I tried with several different images, with images that only had the plate and always the same thing (understand things like error). But if I create an image in Paint with a text and try to read it, it works, but that’s not what I want.

I wonder if anyone has ever had experience with this and knows how to solve, or if you know another way to work with OCR and that meets this need.

If you know another way do in another language, please tell me, I can develop in a language other than C#.

c# java c python ocr
  • 1

    In what format are you passing the image to the Tesseract?

  • I tried with Tif, Jpg and Bmp.

3 answers

6


The complete solution can be seen in this video. For what I need it is necessary to use Emgucv, Opencv and Tesseract. So I can do the reading of vehicular plates.

6

There is a guy named Opencv who solves his problems, he can capture not only Image but also video, including I’m having to study him that has implementation in C, C++, Java and Python (I opted for python).

http://opencv.org/new-opencv-books.html

I am seriously thinking about buying this book even, because I have a lot of interest and needs in the theme.

Link Opencv

  • does it read the board the way I need it? Know if there is any demo that I can do this test?

3

I had the same problem a few days ago. After testing some the one that worked best for my case was the ocrad, which is in javascript. I needed to recognize the image title of a photo taken from a document.

The good that he gives you the recognition score. In my case when he gave a score greater than 60%, I considered as recognized. But for me it only worked well after editing the image, cutting out only the part he needed to recognize.

In your case it would be interesting to evaluate only the part of the characters on the board, I think it would have better results.

  • I saw his demo, I found it too weak for what I need. But thanks for responding!

  • Well, in your case I think it has to be something more robust. Good luck, if you find something put here for us to know.

Browser other questions tagged c# java c python ocr

You are not signed in. Login or sign up in order to post.