A recursive nonlinear filter for removal of impulse noise

We present a new algorithm for suppression of impulse noise which is based on a special combination of impulse detection and nonlinear filtering. The nonlinear filter possesses a recursive structure and provides an estimate of the original sample and the impulse height. The impulse detector is realized as a radial basis function network and yields a fuzzy decision about the impulsivity of a sample. Due to the special combination of both systems, we need only a small number of parameters which can be computed recursively using a procedure derived from a joint optimality criterion. Thereby, we achieve a dramatic gain in performance over other existing methods for both weakly and highly corrupted signals. Further, the algorithm preserves signal details and edges and is applicable to arbitrary impulse distributions