Text Recognition on the Khmer Identification Cards and Its Application in Electronic Know Your Customer (e-KYC)

Abstract

Extracting digital information from the identification cards (ID) is one of the key elements of Electronic Know Your Customer (e-KYC), a process wherein customer’s identity is verified electronically. Rapid digitalization of the Cambodian economy along the Covid-19 pandemic has accelerated the e-KYC adoption by different sectors such as banking, finance, telecommunication and insurance. Extracting text fields in digital text format such as full name, ID number, and date of birth from an image of the Khmer identification (ID) cards is one of the key steps in the e-KYC pipeline. Identity data are printed both in Latin in machine-readable zone (MRZ) font and Khmer in various fonts, respectively. Thus, in-house text recognition models for Latin and Khmer are developed and trained on the synthetically generated data to extract identity data from the Khmer identification card images. Text recognition for Khmer is more challenging, owning to its complex features which are, otherwise, not present in Latin script. The trained models are evaluated on a private collection of the Khmer identification card images and comparisons are made against Google Tesseract OCR 4.0. The evaluation CER (character error rate) and SR (success rate) show that the trained models outperform Tesseract OCR across all fields on the Khmer ID cards. The proposed text recognition framework for performing e-KYC process on the Khmer ID cards is known as CamKYC -OCR.