MRZ Scanning

MRZ is the Machine Readable Zone — the two <<<-filled lines at the bottom of a passport (and some ID cards). It follows the ICAO 9303 standard and includes check digits, so a read can be self-validated.

How it works

camera frame → native OCR (Apple Vision / ML Kit) → text lines → parseMrz() (mrz + checksum)

• The native scan(frame) returns recognized text lines (sorted top-to-bottom, left-to-right within a row). It runs real OCR on every call — throttle calls in your worklet (e.g. every 5th frame). By default it recognizes only the central band of the frame; pass { roi: 'full' } for the whole frame.
• parseMrz(lines) finds the MRZ lines, parses them with the mrz library, and validates the check digits.

Parsing API

import { parseMrz, extractMrzLines, type MrzResult } from '@jieonist/vision-camera-ocr-scanner';

const result = parseMrz(ocrTextLines); // MrzResult | null

MrzResult

interface MrzResult {
  valid: boolean;          // true when the MRZ check digits validate
  format: string;          // 'TD1' | 'TD2' | 'TD3'
  documentNumber: string | null;
  firstName: string | null;
  lastName: string | null;
  nationality: string | null;
  issuingState: string | null;
  birthDate: string | null;      // raw YYMMDD
  expirationDate: string | null; // raw YYMMDD
  sex: string | null;
  lines: string[];               // the MRZ lines that were parsed
}

parseMrz scans consecutive candidate-line windows and returns the best parse (preferring checksum-valid + named), so an extra OCR line doesn't select the wrong pair. It returns null when no MRZ is found; a result may have valid: false when a check digit fails (e.g. a noisy read or a sample document).

Never log the result object

MrzResult.lines contains the raw MRZ lines — the passport number, birth date, and name in plain text. Logging the result whole (e.g. console.log(result)) leaks them into device logs and crash/analytics pipelines. Log only the fields you need (e.g. result.valid, result.format), and drop the result from state as soon as your flow is done.

Notes & limitations

• Dates are raw YYMMDD — interpret the century in your app (expiry is always future/20xx; a birth date can't be in the future).
• The iOS OCR currently assumes a portrait back-camera orientation.
• Accuracy is "good enough for autofill" — always let users confirm/edit; for certified KYC/liveness use a commercial SDK.