I've only used Adobe Acrobat 9 pro and Readiris Pro 11 at my work place so far, so I haven't used the Japanese version of the Abbyy Finreader (we use only western OCR-programs because we share the licenses with the local department of Chinese studies). The source material has to be at least 300dpi to get usable results. As far as I know there's no OCR software for asian languages that use kanji where you can change the character matrix. I also haven't scanned texts with unusual fonts so far... Revision of the OCR-results is the really tedious part. Setting manual text boxes in Readiris makes for good results, but it takes time and some finetuning to do so.
Adobe acrobat 9 pro: - character matrix cannot be changed > no learning - fast - almost no setting changes possible!!! - change of text boxes (=manual setting of text boxes) not possible Test: source: 400dpi, black text on white background, horizontal text, very good quality > ca.95+% recognition rate source: 300dpi, black text on white background, vertical text, medium quality > ca.90+% recognition rate
readiris pro 11: - very sensible (good quality of the source text is necerssary) - about as fast as adobe - manual correction of text boxes possible - more complicated than adobe Test: source: 400dpi, black text on white background, horizontal text, very good quality > ca.98-99% recognition rate source: 300dpi, black text on white background, vertical text, medium quality > ca.95% recognition rate
If you want to, I could run more tests with various books (TenraWAR corebook for example).