Forum Discussion

RoxanaTFA's avatar
RoxanaTFA
Copper Contributor
Oct 20, 2020

Vision API: preprocessing

Hello,

 

I am looking to do Vision OCR on some bulk scanned docs/photos of docs and the quality (of the image and subsequently of the text output) is not always good. What kind of preprocessing/tools would you recommend to improve the model's results?

Thanks,

Roxana

AMA
  • jakeatmsft's avatar
    jakeatmsft
    Icon for Microsoft rankMicrosoft
    Oct 20, 2020
    Hi Roxana, in my experience OpenCV and SciPy are my go to libraries for preprocessing of images. It's important to provide a high-quality image in order for the OCR to detect text. Try to use at least 300dpi resolution images in order to get acceptable results.

Resources