Vietnamese National ID Card Image Authentication for Integrity Assessment in Optical Character Recognition Systems
TL;DR
The current research investigates authenticity assessment for Vietnamese chip-embedded National ID Card images as a prerequisite step prior to OCR execution, substantiating the framework’s viability for operational deployment, addressing systemic vulnerabilities in automated identity validation pipelines.
Abstract
Optical Character Recognition (OCR) systems are widely deployed in electronic identity verification, yet their integrity remains susceptible to presentation attacks. The current research investigates authenticity assessment for Vietnamese chip-embedded National ID Card (CCCD) images as a prerequisite step prior to OCR execution. The objective is the detection and mitigation of common spoofing vectors, specifically printed photographs and screen captures. A pre-verification framework utilizing the ResNet50 architecture is introduced, analyzing morphological and texture variations to classify input media. Evaluations were conducted on a proprietary dataset containing 2,029 images, exploring both multi-class and binary configurations. Performance metrics indicate the binary formulation achieves a 98.6% accuracy rate, a 1.00 recall for spoof detection, and a 0% False Acceptance Rate. These empirical findings substantiate the framework’s viability for operational deployment, addressing systemic vulnerabilities in automated identity validation pipelines.