Facial Image Verification for Government Kiosk System

摘要

Some government organizations have been started mega-projects from which Thai citizens can access their welfare data and receive services from the government via kiosk machines. The kiosk machines are typically installed at the government offices and public places. To access any data or services, the user authentication is done by the Thai citizen smart ID card. However, the weakness of the current system is person verification; it is possible that someone uses the other’s ID card. In this paper, we propose a facial image verification system prototype that can integrate to the existing well-fare kiosks. The face verification is done by comparing the face image retrieved from the smart ID card and the face image taken from a camera on the kiosk. The prototype system aims to cross-check the ownership of the smart card holder and to protect personal data privacy. We implemented Face Verification APIs based on the ResNet and MobileNet. From the experimental results, when tested with Cross-Age LFW face dataset, the accuracies are 99.996% and 99.988% on the ResNet50 and the MobileNet, respectively. However, the accuracies reduce to 96.296% and 95.556% when tested on the low-quality face images that embedded in the national ID cards. The system is designed as a client-server model using Microservices Architecture (MSA) style that complements “loosely coupled” multiple services which can be leveraged to share important functionality throughout other government agencies.