When recognizing low-quality scans, the results will contain errors. This will be true for every OCR program, and there is no turn-key solution (if there were, it would be enabled by default). If you need high-quality text extraction from low-quality documents, then you would need to have a human reviewer correct errors using a tool like Scribe OCR.