I measured my OCR pipeline on 36 old book scans. The phone-photo fix was one division.

I've been building a PDF-to-text tool that also has to cope with scanned pages. At some point "it...