0

I am trying to read characters from a low resolution image using Tesseract:

enter image description here

Using indications from here, I tried adaptive thresholding, but results are not convincing

[adapt thresholding]

I also tried to adapt code from here without more success.

Would you have any idea to improve pre-processing?

guhur
  • 2,500
  • 1
  • 23
  • 33
  • Honestly I think it is too low res, but at least the image needs to be straightened horizontally, and the border has to go. – user3169 Dec 21 '18 at 06:31
  • Ok thanks. Do you have any idea to help to rectify the skew? Hough lines detector? – guhur Dec 21 '18 at 07:33

1 Answers1

0

In the end, I implemented an OCR using Keras and CTC loss function. Results are great!

You can find details of the OCR on the Keras github and explanations about the method here.

guhur
  • 2,500
  • 1
  • 23
  • 33