Biometric Presentation Attack Detection

Purpose: Biometric systems play a crucial role in authentication and identification processes but are vulnerable to various attacks that compromise their security and reliability. Detecting such attacks is critical to ensuring the integrity of these systems and maintaining user trust. This study focuses on detecting face presentation attacks using a cost-effective thermal sensor array. The primary goal is to combine an RGB camera, a thermal sensor array, and deep convolutional neural networks (CNNs) to differentiate between genuine face presentations and facial presentation attacks. The aim is to develop a novel biometric attack detection technique using the thermal sensor array, which is more affordable compared to other existing technologies. Materials and Methods: The process involves the collection of 46,000 thermal images under various conditions and the application of CNN models for analysis. The thermal images are gathered under diverse lighting conditions, distances, and environments, and are then analyzed using deep learning models, specifically AlexNet and ResNet. The thermal sensor array is chosen for its cost-effectiveness. Findings: The research findings demonstrate the effectiveness of the proposed approach in detecting attacks on biometric systems. Performance metrics such as an accuracy of 0.9671, an F1 score of 0.9893, a precision score of 0.9872, and a recall of 0.9914 highlight the robustness of the model in distinguishing between genuine and attacked presentations. Implications to Theory, Practice and Policy:: This study contributes to the field of biometric attack detection by introducing a cost-effective approach using a thermal sensor array. It offers insights into detecting various types of attacks and highlights advancements made in the area of biometric system security. The findings have significant implications for enhancing the security and reliability of biometric systems in diverse applications.