A self-organizing nonlinear filter based on fuzzy clustering
Robert Sucher · 2002
In this paper we present a new adaptive algorithm for removal of impulse noise which is based on a combination of impulse detection and nonlinear filtering. The input samples are classified using fuzzy clustering, and the cluster center becomes the output of the nonlinear filter. Based on this estimate, the impulse detector computes the innovation which is used to adapt the weights of the nonlinear filter. This unsupervised learning method is related to blind equalizers and self-organizing neural networks. Thereby, we reduce the necessary a-priori information as well as the total computational complexity. Further, simulation results show that the performance of the new algorithm is equivalent to that of a previously reported method with data ordering. However, since no ordering process is required, the method can be easily extended to multivariate image data.