Image processing for OCR with leptonica (inverse color text)

问题

I am trying to process the following image with leptonica to extract text with tesseract.

Original Image:

Tesseract on the original image yields this:

i s l
D2J1FiiE-l191x1iitmwii9 uhiaiislz-2 Q ~37
Bottom linez
With a little time!
you can learn social media technology
using free online resources-
And if you donity
youlll be at a significant disadvantage
to
other HOn-pFOiiTS-

Not great, especially the top background. So using leptionica I use a background removal algorithm (blur, difference, threshold, invert) to get the following image:

But tesseract doesn't do a good job with it:

@@r-mair lkrm@W lh@w ilr@ mJs@ iklh@ ii@c2lhm1@ll
mm Mime
VWU1 a Mitt-Jle time-
@1m ll@@Wn Om @@@lh1
using free onhne resources-
Andifyoudoni
9110 ate a $0 D
to other non-profrts
I

The main problem, it seems, is that now all of the text is outlined instead of solid. How can I adjust my algorithm or what can I add to made the text solid?

回答1:

It seems that this paper proposes a binarization method which solves your problem:

T Kasar, J Kumar and A G Ramakrishnan. Font and Background Color Independent Text Binarization. (2007)

来源：https://stackoverflow.com/questions/11678542/image-processing-for-ocr-with-leptonica-inverse-color-text