A differential enhancement to the multiple window configuration (MWC) image restoration filter

Abstract

In this paper an extension is proposed for the MWC switch-based detection filter. The proposed algorithm uses the partial derivatives of the pixel channels for the x and y direction in order to calculate a new estimate for the pixels that were classified as corrupted and replaced by a median filter output. The algorithm is performed as a final stage of the filter process and as it is only carried out on the detected noise pixels, it has a relatively low computational cost. In addition, it requires no parameters or weighting coefficients so there is no additional training time incurred. The new extension performs exceptionally well, providing an improvement for PSNR measures with both impulse and Gaussian noise types. It is currently performed over each colour channel in the RGB space.