Enhancing Face Liveness Detection: Novel Deep CNN Architectures for Anti-Spoofing

In today's technologically advanced environment, security is of paramount importance, and various methods are employed to ensure robust protection and implement strong security measures. Biometric security can be achieved using various means such as strong keys, efficient key-value pair challenges, etc. Biometric authentication systems use physiological or behavioral mechanisms to offer robustness in identity verification. In terms of physical aspects of authentication security, faces, fingerprints, palms, etc. are used as primary modalities. Face is the most widely used modality due to its ease of availability, but it is easily susceptible to spoof attacks. Attacks on face systems are broadly categorized into 2D and 3D attacks. Leveraging deep learning methods with CNN architectures has shown efficacy in detecting spoofing attempts. However, since pre-trained models have complex architecture and involve large training times, there is a need for simple and efficient architectures with lower computational cost for face liveness detection. This paper introduces two novel deep CNN architectures that have fewer parameters, require less computational resources for execution, and achieve improved results. A comparative analysis with existing methods was conducted using the NUAA Imposter Database and the 3D MAD dataset on RGB and YCBCR color spaces. One proposed architecture achieved an impressive accuracy of 99.87% in detecting 3D face spoofing attacks, with a Half Total Error Rate (HTER) of just 0.19%, outperforming existing methods. The proposed CNN architectures exhibit promising outcomes in enhancing the generalization of attack detection systems with lower computational cost.