MRZ OCR: read, parse and check the machine readable zone of passports and ID cards

The machine readable zone (MRZ) is the strip of characters at the bottom of a passport's data page or on the back of an ID card. It repeats the key fields in a fixed layout with check digits, which makes it the most reliable part of the document to read by machine. Koncile reads it from a photo, a scan or a PDF, parses it, recomputes the check digits and compares it with the printed zone. A software MRZ reader, not a hardware scanner.

What an MRZ contains

The format is defined by the ICAO 9303 standard. A passport carries two lines of 44 characters (TD3). A national ID card carries three lines of 30 characters (TD1). Some older ID documents and visas use two lines of 36 (TD2, MRV). The character set is limited to capital letters, digits and the filler « < », printed in the OCR-B font at fixed positions.

P<FRADOE<<JOHN<MICHAEL<<<<<<<<<<<<<<<<<<<<<<
19AB123454FRA9303036M3103016<<<<<<<<<<<<<<02

line 2, position by position:
19AB12345  document number      4  check digit
FRA        nationality
930303     birth date (YYMMDD) 6  check digit
M          sex
310301     expiry date         6  check digit
<<<<<<<<<<<<<<  optional data   0  check digit (empty field)
2          composite check digit

How the check digits work

Each protected field is followed by a digit computed from its characters with the weights 7, 3, 1 repeated, modulo 10 (letters count A = 10 to Z = 35, « < » counts 0). The last digit of the line checks the number, birth date, expiry date and optional data together. A reader that recomputes these digits knows immediately whether it misread a character, or whether a value was modified without the digit being updated. A full worked example on the ICAO specimen passport is in the guide to the MRZ code; the protected positions on each format, and the reading errors an MRZ checksum cannot see, have their own guide.

What Koncile does with the MRZ

1. Locate the zone on the image, whatever the orientation, from a photo, a scan or a PDF page.
2. Read the characters as fixed positions, which is more reliable than free text OCR.
3. Parse the lines into fields according to the format (TD1, TD2, TD3, MRV).
4. Recompute every check digit and flag a mismatch.
5. Compare the parsed fields with the printed zone: surname, number, birth date, expiry. A disagreement lowers the confidence and raises the fraud score.
6. Return the lines as read plus the parsed fields, in the same JSON as the rest of the document.

The MRZ is only part of the reading. The issue date, the place of birth, the address on an ID card and the issuing authority are not in the MRZ: Koncile reads them in the printed zone. See the passport OCR API and ID card OCR pages for the full field list.

What an MRZ does not prove

A valid MRZ with correct check digits proves that the characters are consistent with each other. It does not prove that the document is genuine: a forger who knows the algorithm produces valid check digits, which is why MRZ generators exist and why Koncile does not publish one. That is also why Koncile looks beyond the MRZ: the file's editing traces on a PDF, photo retouching and AI generation on an image, and the agreement between the MRZ and the printed zone. The result is a fraud score from 0 to 100, a signal to verify, not a verdict on the person.

Software MRZ reader versus hardware scanner

Desktop MRZ swipe readers and full page passport scanners read the zone at a counter, with a dedicated device. A software reader like Koncile's reads the zone from any image your application receives: a phone photo uploaded in a form, a scan sent by email, a PDF from a partner. No device, no installation, the same API for the MRZ and for the rest of the document. For a capture on a phone inside your own app, an MRZ scanner SDK (Dynamsoft, Microblink, Regula, Scanbot) guides the user and reads on device; Koncile then processes the image on the server. The comparison of passport OCR APIs and SDKs covers both. Hotels face the same choice at check in, and the guide to the hotel passport scanner lists what each country asks them to record.

Open source MRZ readers

PassportEye (Python, on Tesseract), readmrz, the mrz parser packages for JavaScript and Java: free, self hosted, good for a prototype, limited to the MRZ, sensitive to image quality, without support or fraud control. Koncile is the step after the prototype: the printed fields, confidence levels, both sides of an ID card, a fraud score, and an account that reads the other documents of the file.

Pricing

The MRZ is read as part of the document: 1 credit per page for an extraction, 2 for a fraud analysis. 20 free credits at signup, then from €199 per month for 1,000 credits on koncile.ai.