Introduction to PDF417

PDF417 Haiku

OCR (optical character recognition) based document scanning technologies are known to be error prone. Documents get damaged, text gets read incorrectly and important data gets lost in the process. In an article written by Aleksandr Solonskyi on Towards Data Science, his team compared multiple OCR technologies for their ability to read drivers licenses and passports. Microsoft Computer Vision, OCR Space, Abbyy, Google Vision, Cloud Mersive, Tesseract 3 and Tesseract 4 were all given the same training data sets and Google Vision ended up with the highest accuracy rate at ~80%.

Nanonets also provides an OCR based driver’s license verification service that advertises a 95% accuracy rate. Even if a 99% accuracy could be achieved, 1 out of every 100 documents scanned would result in inaccurate data. In some scenarios this loss of accuracy might be acceptable. In healthcare and government, the consequences of data loss or inaccuracy could be very costly. Correcting this data would require someone to visually confirm each scanned document and manually correct for errors.

This problem can be solved by relying on a technology with built in error detection and correction. In 1991 a solution to this problem was developed called PDF417. The implementation of PDF417’s ensures that even damaged barcodes with missing sections could be scanned without data loss or corruption. PDF417 quickly became an industry standard for healthcare, government and other industries where document scanning is prevalent, and accuracy is paramount.

Where PDF417 is used:

  • FedEx and United States Postal Service Postage
  • Airline Industry’s Bar Coded Boarding Pass
  • Department of Homeland Security’s RealID
  • Immigrations Documents in Israel
  • Many Healthcare Electronic Medical Record Systems

USPS Shipping Example

In this series we will take a look at the PDF417 standard to show how its design allows readers to decode them, detect and also correct for errors.

, , , ,

Related Articles