Robust subspace learning via binary weights for face recognition

In extreme noise environments, existing subspace learning algorithms applied to face recognition tasks exhibit limitations in feature extraction capabilities, robustness, and anomaly detection. To address these issues, this paper proposes a robust two-dimensional principal component analysis (2DPCA) method based on binary weights, aimed at effectively identifying and eliminating outliers, thereby enhancing the robustness and efficacy of face recognition systems in practical applications. By incorporating a binary weight strategy and optimal means, our model maintains a high capability for low-dimensional representation across varying noise levels. We conducted extensive experiments by artificially adding 40% and 60% occlusion noise into four publicly available facial datasets: Aberdeen, YALE, CaltechFaces, and GT. The experimental results demonstrate that the proposed algorithm achieves satisfactory outcomes even when handling severely corrupted data, particularly under conditions of 60% noise, where its reconstruction performance significantly surpasses that of the comparison algorithms, underscoring its superior robustness and adaptability.