Shortly, I want to make the pre-processing procedures before OCR with the suggestion comes from ABBYY \'s technology. There are two parts in article:
I would encourage you to use this code: http://liris.cnrs.fr/christian.wolf/software/binarize/ In particular wolf's binarization, it works really well in practice and it needs very little change to c++ code if you want to use it with opencv. Basically you have to pass the pointer to your image data to this function.