APPROXIMATE PERFORMANCE ANALYSIS OF EDGE-PRESERVING FILTERS

A variety of robust filtering procedures have been applied in edge preserving signal enhancement. The edge performance of these filters has been gauged through the evaluation of their output mean and the variance, in the vicinity of an edge. When exact analytical techniques for evaluation of these measures prove to be intractable, one is forced to rely on statistical simulations for performance analysis. Here we apply approximate, asymptotic methods, specifically for analyzing the edge response of a filter. This technique is tractable for the case of filters based on estimators that are expressible as functionals of the empirical distribution of the sample. As an illustration of this approach, we use this method to analyze the modified trimmed mean and a few examples of the class of L-filters. The resulting performance predictions are compared with simulation results. Fairly accurate predictions are obtained for realistic sample sizes.