A Simple Adaptive Filter for the Restoration of Nonstationary Signals

A simple adaptive filter is proposed for the restoration of smoothly varying signals in a nonstationary noisy background. The filter switches between the median filter and the running mean filter according to local estimates of the tall behavior (impulsivity) of the noise distribution. A theoretical threshold is derived against which the local estimate is compared. The local median is selected if the thresh old is exceeded; otherwise, the local mean is used. Simulations using a family of generalized exponential distribution were performed to compare the median, the mean filter, the -/spl gamma/ -filter, the adaptive /spl gamma/ filter and the proposed filter.