I bet 90% of the problem space is legacy PDFs. My company has thousands of these. Some are crappy scans. Some have Adobe's OCR embedded, but most have none at all.