Doc photos taken by cameras

[zqngetsu96]

Hello,

First of all, gotta say that this work is insanely good. I've been trying to apply for pictures taken by cameras instead but as this work is mainly aimed at scanned documents, I've been having fairly noisy results(images below). I've wondered if you have any tips to handle documents that have fairly varying light distribution(not as uniform as the light from a scanner) would it possible to get a clean white background on the image?

Thank you in advance

original

with noteshrink

[zvezdochiot]

• https://github.com/Liang-yc/IntegralThreshold
• https://github.com/pedrofrodenas/image-thresholding-OCR

[zqngetsu96]

@zvezdochiot Oh man thanks that's pretty cool actually it will help a lot in the BW scanning. But for now, I need the color scanning as well. Any leads on thresholding colored images (basically to get a full white background)?

[zvezdochiot]

@zqngetsu96

1. Use PIL to average the original and bw images.
2. Use T in IntegralThreshold not as a threshold, but as a normalizer: N=(128/(T+1)).

[zqngetsu96]

Thank you, am gonna try this out :')