Modified decision based two-phase unsymmetrical trimmed/winsorized mean filter for removal of very high density salt and pepper noise from images and videos

In this paper, a novel two-phase modified decision based unsymmetrical trimmed mean filter, for removal of very high density salt and pepper noise (SPN) from images and videos is proposed. The first phase comprises the use of an unsymmetrical trimmed mean filter when the processing window is fully noisy and contains both outliers (0 and 255). The second phase is applied to eradicate the residual noisy pixels by replacing the processing pixel conditionally either with mean or with unsymmetrical trimmed mean. A second version of the algorithm is also devised by just replacing the unsymmetrical trimmed mean with unsymmetrical Winsorized mean in the second phase. The efficacy of different algorithms are evaluated, upto 99% density of SPN against standard Grey scale images, color images and video databases, in terms of Peak Signal to Noise Ratio (PSNR), Image Enhancement Factor (IEF) and Structural Similarity Index (SSIM). It has been observed that the proposed algorithms exhibit excellent noise suppression capabilities by giving high value of PSNR, IEF and SSIM. The edge preserving capability of the proposed algorithms is evaluated by Pratt’s Figure of Merit (PFOM), quantitatively and qualitatively it has been proved that edge preservation capability of the proposed algorithms is best among the state-of-art-algorithms.