Removal of Mixed Poisson and Impulse Noise in 1-D Signals by Non-Adaptive and Adaptive Nonlinear Filters

摘要

A case of filtering 1-D signals corrupted by the Poisson fluctuative and impulse noise is considered. Statistical characteristics of output signals are analyzed for several conventional nonlinear non-adaptive filters. It is shown that output bias can be present and it depends upon the filter type, scanning window size and information content of the processed signal. Residual noise variance is also derived and studied. A soft-switching locally adaptive filter that takes into account the observed statistics is proposed and tested.