Enhancing ATM Security Via Deep Transfer Learning-Based Face Recognition

Abstract

Computer vision, a critical branch of artificial intelligence (AI), has transformative applications in information security, biometric authentication, banking, and beyond. Despite the widespread adoption of Automated Teller Machines (ATMs) for 24/7 financial transactions, security vulnerabilities persist, necessitating robust authentication mechanisms. This study proposes a deep transfer learning-based face recognition system to enhance ATM security by accurately verifying legitimate users. Leveraging on facial biometrics—a highly distinctive and tamper-resistant identifier, we benchmark state-of-the-art Convolutional Neural Networks (CNNs), including, VGG16, ResNet50, InceptionV3, and MobileNet, against various train-test splits (70/30, 80/20, and 90/10). Empirical results show that InceptionV3 outperforms the best with 98.7% training accuracy and 97.8% validation accuracy, outcompeting comparative models. Our methodology involves qualitative and quantitative analysis, where RGB facial images (128×128×3 pixels) and binary class labels. The research offers an efficient, scalable solution for the prevention of ATM fraud, in accordance with the latest trends in AI-driven financial security.