OCR issue with Color format

henryma · December 2020

Is there a reason why color caused issue with OCR? Even though the font is readable, OCR have issue with reading it. If I binarize the image then it works fine; however there are time when image have highlighting over the text which I can read from the image, but binarize will cause issue. Is there a reason why color shading and the correct font color to make it readable to a human, but OCR cant read it?

Example of original image value that OCR have issue reading

Image: https://us.v-cdn.net/6030453/uploads/editor/mw/tjkz8nc35zcx.png

But if I binarize it then it will read it

Image: https://us.v-cdn.net/6030453/uploads/editor/5y/jk8mzax3ctun.png

problem with binarize is if there is highlighting it will be an issue.

Image: https://us.v-cdn.net/6030453/uploads/editor/wf/x61r9pfm0e6f.png

Image: https://us.v-cdn.net/6030453/uploads/editor/w7/bu2rjfoqhwg1.png

GrooperGuru · February 2021

Thought I should shed some light on this.

Any time you pass a color or grayscale image to the OCR engine and your OCR's IP Profile does not perform binarization, the OCR engine will perform "Simple" thresholding to the image behind the scenes. Simple thresholding has a very high likelihood of destroying the highlighted text as you've observed in your second screenshot. The only way to prevent this is to perform Binarization in an IP Profile. It is very likely that Simple or Auto binarization will have issues. But if you try Adaptive or Dynamic Thresholding, the black text within the yellow zone should come forward in a highly legible way.

OCR issue with Color format

Answers