Abstract

Facial biometric authentication systems are crucial for the security of applications such as e-banking and mobile devices, and for safeguarding physical access. However, sophisticated attacks with printed images, video recordings, and 3D facsimiles can easily defeat these systems, thus reducing their value and acceptance. Existing anti-spoofing approaches that concentrate on assessing textures and manually generated features are often inadequate for detecting subtle anomalies such as micro-texture distortions or depth changes, motivating the development of advanced systems. This work is based on Convolutional Neural Networks (CNN). The dataset initially consisted of 2,000 images, but after adding data augmentation methods (rescale, rotation range, width/height shift range, shear range, zoom range), the generalization capability was strengthened, allowing texture anomalies, lighting changes, and depth discrepancies to be detected more easily. Model performance was evaluated using accuracy, precision, recall, F1 score, and equal error rate. CNN showed better performance with 91.5% accuracy. These results laid the ground for the development of strong real-time biometric security systems, which are key to reliability in critical applications.