July 10, 2025

The Evolution of the MRZ Reading: from Scanners to AI

In the modern world, the need for fast and reliable reading of identity document information is ever present and ever growing. Are you going to the airport, opening a bank account or checking in at a hotel? None of it would be possible without the MRZ (machine-readable zone) of an identity document, which allows the data to be transferred quickly and efficiently.

The MRZ consists of two or three lines of text which contain important personal information, including the document holder’s nationality, sex, date of birth and name; as well as information about the document itself, such as its number and expiration date. Additionally, it has check digits, used to verify if each individual field has been read correctly.

The format of the MRZ is standardized and regulated by the ICAO (International Civil Aviation Organization), with all MRZs using the same font and only Latin letters, numbers and the sign “<”.

The MRZ can be found all over the world in passports, national IDs, visas, driver’s licenses and residence permits.

But can the MRZ only be used for simple and automatic data extraction? Absolutely not. While this is the first thing that comes to mind, and certainly one of its main functions, today it serves two more key roles: checking the authenticity of the document and helping to decrypt the information of the RFID chip.

Firstly, to check the document’s authenticity, we can verify if the MRZ complies with the standard font and format, as well as compare its data with the one found in the VIZ (visual inspection zone). We do this, because since they both contain the same exact information, an authentic document must find a perfect match between the two zones.

Its other role is decrypting the information present in the RFID chip. This chip is used to store personal and biometric information and is similarly found in identity documents, but also in bank cards and product tags, among others. Yet, due to security concerns, the information on these chips is encrypted and only becomes accessible to those who also have access to the corresponding MRZ. This adds a layer of safety to the transfer of personal biometric information.

An added benefit of the MRZ is that, unlike the VIZ which can only be read by humans, and the RFID chip, which can only be read by machines, the MRZ can be read by both.

As time has passed and technology progressed, the methods for reading the MRZ have changed considerably.

From the 1980s – when it first started appearing in identity documents – until the early 2010s, it was read exclusively through scanners. Due to the clear, standardized readings of scanners, the methods used for this were simple (by today’s standards) and had become extremely efficient by the late 1990s, with detection and recognition rates of close to 100%.

However, with smartphones becoming widespread in the early 2010s, MRZ reading faced new challenges: smartphone pictures and videos have blur, noise, projective distortions and capture the holograms of the document. This meant that new methods had to be developed to deal with the added complexities.

Different approaches were developed to deal with the two fundamental tasks of MRZ reading: detection and recognition. First, the MRZ must be correctly detected within the picture or video, and second, the content of the characters must be accurately recognized.

Nowadays, the most common methods for MRZ detection are deep learning approaches, such as HED-MRZ and YOLO-MRZ. These two methods show accuracies of close to but under 90%, which still leaves great room for improvement.

On the other hand, the most common methods for MRZ recognition are OCR frameworks. OCR frameworks are built to do optical character recognition, that is, taking the text in an image and converting it so that it can be read by computers.

OCR technology is extremely valuable for this task. However, not all types of OCR work as effectively with the MRZ, because some don’t adapt to its specific language model and have issues in calculating check digits. In any case, not all of them present the same problems, and the best ones have already achieved accuracies of a little over 90%, which shows significant progress over the first attempts at MRZ recognition in mobile photos and videos.

One of the problems facing this technology is in testing the methods in large databases with realistic document pictures and videos. The private nature of identity document information leads to a relatively small number of datasets that can be used for testing, and not all are equally suited for this task.

All in all, the implementation of the MRZ in identity documents has revolutionized both data extraction and verification. It made fast and reliable transfer of essential information possible, as well as playing an important role in verifying document authenticity and securely retrieving data from the RFID chip.

However, the path to a perfect MRZ reading is far from over, as many generic OCR solutions still fail to adapt to the unique task at hand, using models which are not optimized to the MRZ’s specific structure and struggling to achieve sufficient levels of accuracy.

The best results require specialized solutions, which are properly optimized for this very task, like the ones provided by OCR Studio. Designed specifically for fast and reliable processing of the MRZ, even in poor shooting conditions, OCR Studio stands out as an indispensable choice for organizations and institutions in need of top-of-the-line MRZ reading.