What font is used in the MRZ (OCR-B)?

The MRZ is printed in OCR-B, a monospaced font designed in the 1960s specifically to be read by both machines and people. It was developed by Adrian Frutiger, and its defining feature is that every character has a distinct, unambiguous shape.

The letter O and the number 0 look different enough that a scanner, or a human squinting at a worn passport, won't easily confuse them, and the same goes for other commonly mixed-up pairs like 1 and I.

This matters more than it might seem, because MRZ reading relies on the assumption that a scanner can tell characters apart reliably even under imperfect conditions: a slightly blurry photo, uneven lighting, ink that's faded after years in a wallet. A typical typeface optimized for human readability actually performs worse here, since fonts like that often use stylistic flourishes or tighter spacing that make automated recognition harder.

OCR-B strips that away in favor of clean, evenly spaced, geometrically simple characters.

The specification is fixed by ICAO Doc 9303, which mandates OCR-B for all MRZ printing so any compliant reader, anywhere in the world, can process a document regardless of which country issued it. This is part of why MRZ reading tends to be so much more reliable than reading, say, a handwritten address field or a stylized logo.

The font itself was engineered for the job.

For companies building document scanning products, this uniformity is a genuine advantage. Because every legitimate MRZ uses the same font at a predictable size and spacing, an OCR engine can be trained and tuned specifically for that one font rather than needing to generalize across hundreds of different typefaces the way visual zone OCR does.

That's a meaningful part of why MRZ extraction tends to achieve higher accuracy rates than OCR on the printed name or address fields elsewhere on a document.

ScanDoc's OCR engine takes advantage of this by applying MRZ-specific recognition tuned to OCR-B's character shapes, separate from the broader OCR models used on the visual zone, which have to handle dozens of different national fonts and layouts. The result is that MRZ fields typically come back with very high confidence, and any mismatch against the visual zone or a failed check digit is a strong, early signal that something about the document deserves a closer look.