Robust image denoising using kernel-induced measures

We propose a class of novel nonlinear robust filters for image denoising by incorporating the kernel-induced measures into the classical linear mean filter. Particularly, we place more focus on the Gaussian kernel based filter (GK) due to its simplicity. The GK filter not only generalizes and makes the original linear mean filter highly resistant to outliers but also outperforms a typical and powerful mean-logCauchy filter recently developed by Hamza et al in the mixed noise removal in certain specific conditions in the normalized mean square error (NMSE) sense. The experimental results also illustrate that the kernel-based nonlinear filters are promising.