Stable nonlinear filters with spatial prediction

The paper deals with nonlinear two-dimensional filters for image restoration. The filter employs classic predictor structure with nonlinear predictor and prediction error processed by a static (memoryless) nonlinear element or by a nonlinear two-dimensional filter. The filters discussed are both recursive and nonrecursive. Even in the case of recursive structures stability is guaranteed. The filters are suitable for impulsive noise removal from color images. Preservation of textures and even one-pixel thin lines are advantages of the filters proposed. The experimental data prove that these filters outperforms classic nonlinear median-based filters like vector median, recursive median and weighted median.