International Research Journal of Education and Technology

Presentation assaults, or face spoofing, are a growing threat to biometric security systems. These attacks use fake facial inputs like 3D masks, videos, or photos to deceive facial recognition algorithms. As facial recognition becomes central to banking, access control, and mobile security, robust spoofing detection is vital. This research proposes using ResNet-18, a lightweight and effective convolutional neural network (CNN), for advanced face spoofing detection. ResNet-18’s residual learning structure enables precise extraction of deep facial features, distinguishing genuine users from attackers even in sophisticated spoofing attempts. Its lightweight design ensures real-time deployment in security applications without compromising efficiency. Validated on multiple datasets, the ResNet-18-based model consistently outperforms traditional techniques, offering higher accuracy and robustness across various attack scenarios. This approach underscores the potential of deep learning to enhance the reliability of facial recognition systems, ensuring secure, real-time detection in practical environments.

Key Words: Face Spoofing Detection, ResNet-18, Biometric Security, Deep Learning.