Recognition of Machine-Readable Zone in Identity Documents: A Review

Encoding personal identity document information into machine-readable formats is one of the most important approaches to automatic data processing. The Machine-Readable Zone (MRZ) of a document contains text data in the form of 2 or 3 long text lines of a special structure regulated by ICAO, which are easily read by computing devices using OCR methods. The importance of the MRZ lies not only in the automatic data entry, but also in the ability to verify the authenticity of a document and access biometric data in the RFID chip. Originally developed for scanners, it has gradually come to be read from photos and videos taken in uncontrolled conditions from mobile device cameras. The main problem with this transition is that MRZ was not designed for recognition in photos and videos with projective distortions, blurs and other effects, which makes the task much more difficult. Modern document processing systems recognize MRZ in complex 3D scenes with a quality of around 90%, which is extremely low for a machine-readable form of data representation with a strict format. This paper provides an overview of the current state of the problem of MRZ recognition in identity documents. Recognition approaches from the 1990s to modern ones for scanned images and photos are considered. The review reveals understudied aspects of MRZ reading in contemporary research, such as the lack of specialized recognition and language models. It is also noted that many works are based on universal recognition models, the complexity of which is excessive for the MRZ. At the end of the review, a brief description of the state of the problem is given and directions for further research in this area are outlined.