Paper Contents
Abstract
Face biometric systems are now widely employed across governments, commercial sectors, and social media platforms, but this widespread adoption has made them susceptible to various attacks, including face spoofing, Distributed Denial of Service (DDoS), and phishing. These threats pose significant security challenges, underscoring the urgent need for more robust protective measures to safeguard these systems. The widespread adoption also comes with serious security concerns and underlines the necessity for advanced protective measures. In this regard, the paper proposes an enhanced face spoofing detection framework, which may effectively cover these vulnerabilities with improved accuracy. The proposed framework builds on deep learning advancements and integrates pre-trained convolutional autoencoders for feature extraction and dimensionality reduction, followed by a softmax classifier for classification. It is also further strengthened by advanced architectures such as ResNet, which uses deep residual learning, with Convolutional Block Attention Modules to enhance feature focus. Besides, the hybrid models combining CNNs with LSTM networks are used to extract both spatial and temporal features, which are very essential in detecting spoofing either in images or videos. Comprehensive experiments on benchmark datasets like Idiap Replay Attack, CASIA-FASD, and 3DMAD demonstrate that an extended framework can at least be as good as, and often better than, state-of-the-art methods, ensuring better generalization across various spoofing scenarios. This fundamentally increases the security and reliability of biometric systems through a view to offer a robust solution against sophisticated spoofing attacks.
Copyright
Copyright © 2024 Ratti Pallivi. This is an open access article distributed under the Creative Commons Attribution License.
