Rank Order Weighted Mean Switching Filter for Salt and Pepper Noise Removal with Edge Preservation

Salt and pepper impulse noise remains one of the most disruptive forms of image corruption from acquisition, transmission, and low-quality sensor pipelines, with classical filters continuing to trade off noise suppression against edge and texture fidelity. This paper presents the Rank Order Weighted Mean Switching Filter (ROWM-SMF), a two-stage restoration scheme that first detects candidate impulse pixels using a rank-order median-based detector, then restores only the identified noisy pixels through a locally adaptive weighted mean combining a similarity weight with a rank-order salt-and-pepper penalty term, leaving uncorrupted pixels untouched. The filter is evaluated on the 24-image Kodak Lossless True Colour Image Suite corrupted with salt-and-pepper noise at six total noise densities (10%–60%) derived from nine salt-and-pepper density combinations, yielding 214 grayscale test cases. Performance is assessed using PSNR, SSIM, and FSIM, and compared against a standard 5×5 Median Filter and the recent OPERA-Net method. Averaged over all test cases, ROWM-SMF achieved a mean PSNR of 23.68 dB, SSIM of 0.790, and FSIM of 0.656, versus 21.81 dB, 0.617, and 0.420 for the Median Filter, gains of 1.87 dB, 0.173, and 0.236, respectively. Comparisons at noise densities of 10%, 20%, and 30% further show that ROWM-SMF consistently outperformed OPERA-Net in both PSNR and SSIM. These results confirm ROWM-SMF's effectiveness over both classical and recent methods. Despite added computational cost in the current implementation, the framework offers an effective, lightweight, training-free solution for high-fidelity salt-and-pepper image denoising.