Actually, I have often done exactly what you are trying to do with surprisingly good results.
First I scan the original as an image, then I use Photoshop (or equivalent) to remove surplus characters, borders and blemishes, and increase the contrast, save the amended scan. Then just use any simple, unsophisticated OCR; the fewer the features, the less to go wrong. Both my equivalent of Photoshop and my OCR are old Windows 95 products which came with a digital camera in the last century but which still run on Windows 10 !.
After that, I open the OCR document in Word, "SELECT ALL" and change the format to "TEXT". There will inevitably be a few misreads which can be manually corrected and the document can then re-formatted to suit your fancy. If there are unnecessary double lines, then use "FIND AND REPLACE" to remove them and other repeated errant interpretations.
I have used this process hundreds of times, of course with a few failures, but it does depend on the quality of the original copy. I suspect that you may be using sophisticated software which is trying to be too clever by half. Keep it simple, strip out all the options and features and ideally use some old fashioned software !
