A class of robust edge-preserving filters for image restoration

Presents the theory and simulation results of robust estimation procedures for restoring real images. These procedures are capable of smoothing images corrupted with a wide range of noise types while preserving the edges of the original images. The simulation results have shown that a robust edge-preserving (REP) filter is superior in the sense of both the mean square error (MSE) and mean absolute error (MAE) to the standard linear and median filters. In addition, the outputs of the REP filter have a sharper look from the perspective of human vision as compared to the outputs of both the linear and median filters. >