Credit Card Scanning

Point the camera at a payment card to extract the card number, brand, expiry, and (best-effort) cardholder name. The card number is validated with the Luhn checksum, so a clean read is self-validating.

How it works

camera frame → native OCR (Apple Vision / ML Kit) → text lines → parseCard() (Luhn + BIN)

Like MRZ, the native side does OCR only and returns text lines; parseCard runs on the JS thread. It has no extra native dependency — it reuses the same scan(frame).

Parsing API

import { parseCard, detectBrand, type CardResult } from '@jieonist/vision-camera-ocr-scanner';

const result = parseCard(ocrTextLines); // CardResult | null

CardResult

interface CardResult {
  valid: boolean;              // true when the number passes the Luhn checksum
  number: string | null;       // digits only, e.g. '4111111111111111'
  numberFormatted: string | null; // grouped, e.g. '4111 1111 1111 1111' (Amex 4-6-5)
  brand: string | null;        // 'visa' | 'mastercard' | 'amex' | 'discover' | 'jcb' | 'diners' | 'unknown'
  expiryMonth: string | null;  // 'MM'
  expiryYear: string | null;   // 'YY'
  holderName: string | null;   // best-effort, may be null
  lines: string[];             // the OCR lines that were scanned
}

The card number is the anchor — parseCard returns null when no plausible number is found. It prefers a brand-known + Luhn-valid run, so a coincidentally-valid substring doesn't win over the real number. When several expiry dates appear (e.g. Amex "member since" + "valid thru"), the latest date is chosen.

detectBrand(number) is exported separately if you only need the brand from a known number.

Never log the result object

CardResult.lines contains the raw OCR lines — including the full card number (PAN). Logging the result whole (e.g. console.log(result)) leaks it into device logs and crash/analytics pipelines. Log only the fields you need (e.g. result.valid, result.brand), and drop the result from state as soon as your flow is done.

Notes & limitations

• Cardholder name is best-effort. Card fonts and embossing read poorly; treat it as a hint and let users edit.
• Embossed (raised) numbers scan worse than flat-printed ones — good lighting and a flat angle help a lot.
• valid reflects only the Luhn checksum — it does not mean the card is real, active, or authorized. Never use this for payment authorization; it's for autofill / data entry.
• iOS OCR currently assumes a portrait back-camera orientation.