I want to find a generalized method that can help me improve accuracy of OCR output. I have tried several methods in OpenCV i.e binary thersholding, adaptive thresholding, blurring, closing etc. After pre processing I get good results for one template but the accuracy for another template decreases. Hence I want to preprocess all templates using one method that can give me 90 to 95 percent accuracy.

The images I am trying to process are here.

OCR used is pytesserect.

  • I checked the 6 images and I don’t think they need processing, because their quality is good. They do contain thousands of colors, because they’re saved as lossy JPEG, but simple binarization should take care of that. What could give you better results is to use a more powerful OCR engine, such as LEADTOOLS (Disclaimer: I work for the vendor). When I tested with our OCR, all 6 images gave more than 95% correct words without any pre-processing. If you like, you can try the free evaluation. – Amin Dodin Feb 20 at 20:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.