AN OPTIMIZED FEATURE EXTRACTION FRAMEWORK FOR HYBRID LEARNING-BASED FACE ANTI-SPOOFING

Abstract

Face spoofing attacks has questioned the soundness of face recognition systems. Several studies on face antispoofing (FAS) have suggested various methods to detect these attacks . The traditional approaches lack generalization and efficiency in real-world application. Software based approaches have resulted in significant improvement in application of FAS. The mix of several approaches lead to hybrid models for being more discriminative and resilient. A novel and optimized feature extraction framework is proposed for FAS using Local Binary Patterns (LBP). The approach integrates the Whale Optimization Algorithm (WOA) within a hybrid component learning-based architecture for improved accuracy.LBP is employed to extract robust and discriminative texture features that effectively capture micro-texture variations indicative of spoofing attacks. To enhance the performance, WOA is utilized. It optimized the feature selection through identifying the non-redundant and most relevant features, thereby minimizing computational complexity and significant improvement in classification accuracy. The hybrid techniques combine multiple learning strategies to strengthen the system’s ability aim to generalize across various spoofing methods, print, replay and three-dimensional (3D)mask attacks. Some results on benchmark datasets have achieved with the proposed framework.Significant improvements are achieved over traditional methods in the areas of detection precision, and robustness, and accuracy. The results confirm an effective and reliable face anti-spoofing solution for biometric authentication systems in security applications.