Face recognition using Nuclear Norm-based Angle Two-Dimensional Principal Component Analysis

Several methods employing 2DPCA for dimensionality reduction have exhibited limitations in classification accuracy. Angle Two-Dimensional Principal Component Analysis (Angle 2DPCA) seeks to enhance the robustness of standard 2DPCA by investigating the connection between reconstruction error and variance utilizing the Frobenius norm. Despite these improvements, Angle 2DPCA still faces challenges. To address these issues and further bolster robustness, we introduce a novel model called Nuclear Norm-based Angle Two-Dimensional Principal Component Analysis (AN-2DPCA). This model redefines Angle-2DPCA by replacing the Frobenius norm with the nuclear norm. AN-2DPCA showcases superior robustness and interpretability compared to Angle 2DPCA. Additionally, we present a rapid, non-greedy iterative algorithm for solving the AN-2DPCA problem. Experimental outcomes from various open databases underscore the effectiveness of our proposed approach.