Edge detectors based on linear and least median operators

The author introduces a class of edge detectors based on a modified FIR-median filter. The input signal is filtered with three linear-phase filters, and the output of the filter is the shortest half of the order statistic of the outputs of the linear filters. The statistical properties of the edge detector filter are analyzed for input signals imbedded in innovative noise and additive noise. It is shown that the proposed filter is very robust, whereas the standard median filter and the FIR median hybrid filter are very weak in rejecting outliers. The filter preserves corner points and ramps effectively when the input signal is imbedded in innovative or/and additive noise.<>