Fast MRZ

This repository extracts the Machine Readable Zone (MRZ) from document images. The MRZ typically contains important information such as the document holder's name, nationality, document number, date of birth, etc.

️Features:

• Detects and extracts the MRZ region from document images
• Contour detection to accurately identify the MRZ area
• Custom trained models for ONNX and Tesseract
• Contains checksum logics for data validation
• Outputs the extracted MRZ region as text/json for further processing or analysis

🛠️Built With

⚙️Installation

1. Install Tesseract OCR engine. And set PATH variable with the executable.

2. Install fastmrz

   pip install fastmrz

   This can be done through conda too if you prefer.

   conda create -n fastmrz tesseract -c conda-forge
   conda activate fastmrz

3. Copy mrz.traineddata in tessdata folder of repo to the tessdata folder in installed tesseract location

💡Example

from fastmrz import FastMRZ
import json

fast_mrz = FastMRZ()
# Pass file path of installed Tesseract OCR, incase if not added to PATH variable
# fast_mrz = FastMRZ(tesseract_path=r'/opt/homebrew/Cellar/tesseract/5.3.4_1/bin/tesseract') # Default path in Mac
# fast_mrz = FastMRZ(tesseract_path=r'C:\\Program Files\\Tesseract-OCR\\tesseract.exe') # Default path in Windows
passport_mrz = fast_mrz.get_mrz("../data/passport_uk.jpg")
print("JSON:")
print(json.dumps(passport_mrz, indent=4))

print("\n")

passport_mrz = fast_mrz.get_mrz("../data/passport_uk.jpg", raw=True)
print("TEXT:")
print(passport_mrz)

OUTPUT:

JSON:
{
    "mrz_type": "TD3",
    "document_type": "P",
    "country_code": "GBR",
    "surname": "PUDARSAN",
    "given_name": "HENERT",
    "document_number": "707797979",
    "nationality": "GBR",
    "date_of_birth": "1995-05-20",
    "sex": "M",
    "date_of_expiry": "2017-04-22",
    "status": "SUCCESS"
}


TEXT:
P<GBRPUDARSAN<<HENERT<<<<<<<<<<<<<<<<<<<<<<<
7077979792GBR9505209M1704224<<<<<<<<<<<<<<00

📃MRZ Wiki

MRZ Types & Format

The standard for MRZ code is strictly regulated and has to comply with Doc 9303. Machine Readable Travel Documents published by the International Civil Aviation Organization.

There are currently several types of ICAO standard machine-readable zones, which vary in the number of lines and characters in each line:

• TD-1 (e.g. citizen’s identification card, EU ID card, US Green Card): consists of 3 lines, 30 characters each.
• TD-2 (e.g. Romania ID, old type of German ID), and MRV-B (machine-readable visas type B — e.g. Schengen visa): consists of 2 lines, 36 characters each.
• TD-3 (all international passports, also known as MRP), and MRV-A (machine-readable visas type A — issued by the USA, Japan, China, and others): consist of 2 lines, 44 characters each.

Now, based on the example of a national passport, let us take a closer look at the MRZ composition.

🗹ToDo

• Test for mrva and mrvb documents
• Add wiki page
• Send numpy array as input
• Bulk process
• Train Tesseract model with additional data

⚖️License

Distributed under the AGPL-3.0 License. See LICENSE for more information.

⭐Show your support

Give a ⭐️ if this project helped you!