MRZ Scan Screen Wiki

• Overview · Concept
• Deep Linking (?mrz=&scan=) · Feature
• Device-Local Storage · Feature
• Check-Digit Verification · Algorithm
• OCR Pipeline (Tesseract.js) · Technical
• Band Ladder & Repair · Technical
• Licence Barcode Mode (PDF417) · Feature
• URL Parameter Reference · Reference
• Exported Helper API · Reference

The MRZ (Machine-Readable Zone) is the two- or three-line block at the bottom of passports, ID cards and visas, defined by ICAO Doc 9303. The scan screen at /mrz/scan reads that block on device with Tesseract.js, then hands the text to the same parseMrz the rest of the toolkit uses.

• On-device OCR — the MRZ photo is never uploaded; pixels stay in the browser tab.
• Deep linking — pass an MRZ in as ?mrz=<base64> to pre-fill the editor.
• Device-local storage — the last verified read is remembered for 30 days.
• Check-digit verification — every read is graded verified or review.
• Licence mode — a separate card decodes a US/CA licence PDF417 barcode and maps it onto MRZ fields.

Supported formats: TD3 (passport, 2×44), TD1 (ID card, 3×30), TD2 (large ID card, 2×36), MRV-A (visa, 2×44) and MRV-B (visa, 2×36).

The scan screen accepts two URL query parameters:

• decodeMrmzParam(value) — base64url-decode and split on |.
• encodeMrmzParam(lines) — join with | and base64url-encode.
• getScanPayload(url) — extract and validate mrz and scan params.

The scan screen remembers the last successful MRZ read using the browser's localStorage API. Nothing leaves the device.

• restoreLastRead() — reads and validates the stored payload, returning null if missing, stale, or corrupt.
• rememberLastRead(mrz, format) — persists a successful read with the current timestamp.

Every MRZ field has an adjacent check digit computed with the 7-3-1 algorithm. Each character is mapped to a numeric value, multiplied by weights 7, 3, 1, and the sum's last digit is the check digit.

OCR runs entirely in the browser using Tesseract.js with the eng traineddata model (5 MB, tracked in git at the project root).

1. Camera captures a frame and isolates the MRZ region.
2. Image is preprocessed (grayscale, threshold) for contrast.
3. Tesseract.js worker runs inference with the eng model.
4. Raw text is split into lines and passed to parseMrz.
5. Check digits are validated; mismatches are flagged.
6. Result is committed to localStorage and/or the editor.

When the OCR editor has a parsed document (from OCR or from a deep link), the screen shows a synchronized view: the photo (or its OCR patch) alongside the editor, with fields that differ between the two highlighted.

• Fields extracted from OCR that don't match editor input are yellow-highlighted.
• Check-digit mismatches are red-highlighted inline.
• Clicking a highlighted field jumps the editor cursor to that position.
• The overlay can be toggled via a button in the top-right of the photo panel.