Performance Analysis of Weighted Low-Rank Approximation Models for Robust Face Recognition

Face recognition has become a research hotspot in the fields of computer vision, pattern recognition, and machine learning. The accuracy of recognizing faces with varying expressions and illumination as well as occlusions and noise, on the other hand, presents a unique challenge in face recognition systems. Facial images must be preprocessed for improved recognition accuracy. A major issue with the existing approaches is that they have limited capacity that cannot handle large-scale occlusion and noise situations adequately. In this paper, we present Low -rank matrix approximation (LRMA) models like Robust principal component analysis (RPCA), Weighted Nuclear Norm Minimization (WNNM), and Weighted Schatten p-norm (WSNM) for Robust face recognition. A confusion matrix is used for calculating the accuracy of face recognition. Experiments are conducted and the performance of these LRMA models is compared using the Yale database with facial occlusions, poor illumination, expressions, and noise. The results show, both intuitively and numerically, that WSNM outperforms RPCA in removing facial occlusions, resulting in restored low-rank images with greater PSNR and SSIM.