<![CDATA[In the digital era, facial recognition systems have become increasingly vulnerable to spoofing attacks, as demonstrated by cases of identity theft using photos or smartphone screens. This study develops a real-time face liveness detection system using YOLOv8 to address these vulnerabilities. Under controlled laboratory conditions, the system achieved exceptional performance metrics: accuracy of 1.0, precision of 1.0, and recall of 1.0, with a mean Average Precision (mAP) of 0.96. However, this study reveals critical insights about the challenges of real-world deployment, including significant performance degradation under poor lighting conditions where genuine faces were misclassified as spoofed images. Compared to existing methods such as Attention-Based Two-Stream CNN (accuracy: 0.91) and Deep Spatial Gradient approaches (accuracy: 0.90-0.92), our system demonstrates superior performance in controlled environments but highlights the persistent challenge of environmental variability in practical applications. These findings emphasize the need for robust preprocessing techniques and diverse training datasets to bridge the gap between laboratory performance and real-world reliability. The study contributes to understanding the limitations of current face anti-spoofing technologies and provides a foundation for developing more robust systems suitable for practical deployment.]]>
Copyrights © 2025
