In my experience, you need to pre-process the text before Tesseract gives reasonable results. I'd wager 90%+ accuracy after pre-processing a book with ScanTail

In my experience, you need to pre-process the text before Tesseract gives reasonable results. I'd wager 90%+ accuracy after pre-processing a book with ScanTailor. Accuracy falls off a cliff if the text is handwritten rather than using a printing press, but it's still useful to get a side-by-side editable OCR <-> image.