Model Adaptation for Enhanced Liveness Face Detection: Experimental Findings on SiWMv2 Dataset

• Published in: International Conference on Computing Communication Control and Automation (Online) pp. 1 - 6
• Main Authors: Khairnar, Smita, Gite, Shilpa, Thepade, Sudeep D, Mahajan, Kashish
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.08.2024
• Subjects: Adaptation models; Computational modeling; Face detection; face detection architectures; Faces; fine-tuning; Focusing; Glass; liveness detection; pre-trained models; Residual neural networks; Robustness; Security; spoof and liveness; spoof attacks; Systematics
• ISSN: 2771-1358
• DOI: 10.1109/ICCUBEA61740.2024.10774695

The field of liveness face detection, an essential component of security and biometrics, has seen notable advancements through developments in computer vision and deep learning. It's crucial for accurately distinguishing between genuine and spoofed data samples. This study aims to elevate the performance of automated liveness face detection by focusing on model adaptation, utilizing the SiWMv2 dataset. This dataset is particularly valuable as it encompasses 14 kinds of spoofing attacks, that include print, replay attacks, obfuscation makeup, paper glasses, and various others, which significantly enhances the robustness of the detection systems against a wide array of spoofing techniques. By fine-tuning and adapting seven pre-trained architectures-VGG16, DenseNet201, InceptionV3, VGG19, ResNet50, MobileNetV2, and Xception-paper aims to push all boundaries of current automated liveness face detection capabilities. Through systematic experimentation and analysis, this study aims to contribute to the improvement and effectiveness of liveness face detection systems across various applications, ensuring they are more resilient and capable of handling diverse spoofing scenarios.