i trying binarize credit card digits using opencv ocr reconize them.
so after segmentation stage , cropping,i did these steps:
1.convert tje image grayscale image.
2.histogram equalization.
3.gaussianblur (3,3) kernel gaussianblur(digit,bluredimg,cv::size(3,3),9);
4.threshold(sauvola method).
here of results think noisy , insufficient ocr:
links original images:
http://s22.postimg.org/rqnlgz1od/image.jpg
if have tips enhance binarization quality please share me (code great).
i wouldn't apply "binarize" step @ image. i'd first try remove noise elminating pixels differ neighbours ("salt & pepper filter" because removes black & white spots)
Comments
Post a Comment