An Enhanced Face Spoof Detection using ResNet50 Based Cosine Learning Rate

In recent times, face biometric systems have recognized widely and got attention in computer vision. Further, the biometric face recognition systems are susceptible for face spoofing attacks, where an attacker uses a fake face image to copy a legitimate user. An effective face spoof detection method can protect user identity, prevent unauthorized access and maintain trust in biometric authentication. The existing face spoof detection methods offer low efficiency in detecting small objects and cause overfitting. Therefore, this paper introduces ResNet50- Cosine Learning Rate (CLR) for face spoof detection which accurately distinguishes between real and fake face images by preventing unauthorized access in biometric systems. The proposed ResNet50-CLR takes input images from Institute of Automation, Chinese Academy of Sciences (CASIA) dataset. These images were preprocessed by background subtraction where features are extracted by Local Binary Patterns (LBP); then, the extracted features are classified by Proposed ResNet50-CLR which achieved better results in terms of Equal Error Rate (EER) value as 0.12 (Btu/h per watt) and Half Total Error Rate (HTER) as 0.45%, False Acceptance Rate (FAR) as 0.65% and False Rejection Rate (FRR) as 0.66% when compared with existing Stacked models.