Well I am new to OCR and Emgucv and here it goes. I am having colored ID card that I want to process through tesseract ocr to get details. I have converted color image to gray scale and I have successfully converted to binary image and passed the same to tesseract and worked properly though i have to filter junk data from text received.
Now my issue is I am expecting different kind of images having such as contrast, lighting conditioning from users. I want to know if something is available which I can do to get global threshold to get binary image. I already played with adaptivethreshold, Otsu but did not worked for me.
What I am assuming is images are matrix of pixels so is their anyway to write some function which will work on any image? I am not able to figureout where should I start with.
I am working on csharp and using tesseract for ocr. I have used following code.
double th = CvInvoke.Threshold(source, source2, 0, 255, ThresholdType.Otsu);
CvInvoke.Threshold(source, source, th/2, 255, ThresholdType.Binary);
Sample Image: [1]: https://i.sstatic.net/JleRx.jpg
Please suggest example in csharp to find global threshold.
I am doing following steps for OCR.
Additionally Please let me know whether I am doing expected algorithm for OCR or I am missing something? Also please suggest what should i do to improve ocr accuracy? Any help will be highly appreciated.
You should use Canny Edge detection. Emgu CV Canny
It should help your accuracy. It finds edges by their local differences, not by the brightness/contrast of the whole image.
