ADVANCING FACE SPOOFING DETECTION WITH LBP, PCA, AND SVM: A ROBUST AI SECURITY APPROACH

This work introduces a robust method for distinguishing between genuine and fake faces, addressing the crucial issue of biometric spoofing in AI-driven security systems. The proposed approach integrates Local Binary Pattern (LBP) for feature extraction, Principal Component Analysis (PCA) for dimensionality reduction, and Support Vector Machine (SVM) for classification. Evaluations demonstrate the method’s superior performance in face spoofing detection, achieving an overall detection accuracy of 96.7% in cross-validation, surpassing traditional methods such as Random Forest (94.5%). LBP extracts distinctive textural features, which are normalized for uniformity across samples. PCA reduces the dimensionality of the data by eliminating redundant information, maintaining only the most relevant features for analysis. The SVM classifier identifies patterns to differentiate genuine faces from spoofed ones, achieving high accuracy across diverse attack types. For instance, the proposed method achieves 98.1% accuracy for detecting printed photo attacks and 80.9% accuracy for challenging deepfake attacks on the created dataset, outperforming Random Forest by 1.2% and 1.1%, respectively. This comprehensive evaluation highlights the method’s robustness, computational efficiency, and adaptability to various spoofing scenarios. With consistent performance improvements across datasets, this technique addresses critical AI security challenges and provides a scalable solution for advanced face spoofing detection systems.