Fig. 1. Main concept of noise identification in the first step of RPCA-SMF

Fig. 2. Second step noise processing related ideas of RPCA-SMF

Fig. 3. Robustness experiment of artificial dataset

Fig. 4. Partial face images， from left to right are some normal images and their noisy images from FEI， FERET， Yale B， ORL， PIE， and COIL20 datasets respectively

Fig. 5. Accuracy comparison of different algorithms on six benchmark datasets with 25% noise added to each dataset

Fig. 6. Reconstruction error comparison of different algorithms on six benchmark datasets with 25% noise added to each dataset

Fig. 7. Comparison of reconstructed images from different algorithms on FERET， PIE， and AR Datasets

Fig. 8. Anomaly detection results of RPCA-SMF （From top to bottom： 1/4 black and white noise， 1/2 noise block， pure noise block）

Fig. 9. Different algorithms for image compression and restoration effects

Fig. 10. Visualization weights of AR dataset

Fig. 11. Ablation experiment results

Fig. 12. Parameter sensitivity analysis of RPCA-SMF on datasets

Fig. 13. Convergence curves on four datasets of RPCA-SMF