Last night I ran Finereader 9 over a 400-page English translation from 1936 that I had scanned some time ago at 400 dpi. I then settled down for the onerous task of correcting scanner errors; only to find very few indeed. There were perhaps a dozen in the whole book! Probably if I had just exported it to Word and used the spell-checker, I would have found most of them.
I repeated the exercise on another text, with the same result.
FR9 is perceptibly better than FR8 at OCR. It has some annoyances in the user-interface. Worse it forces me to use my Plustek Opticbook 3600 at 300dpi or 600dpi, when FR8 allowed 400 dpi (the optimal resolution). But the fact is that there has been a considerable advance here.
When I look back ten years to the misery of “99% accurate” recognition (i.e. 6 errors a page), it is truly amazing. Recommended.