Decision-based order statistic filters

The authors propose two adaptive order statistic filters which generate outputs based on the results of hypothesis tests. To reduce edge shifting of the median filter, a modified median filter (MMF) is proposed. First, a hypothesis test is applied to detect the edge location. Then, whenever an edge is detected, the MMF generates the output by modifying the median output based on a comparison of median value to the current input value. The MMF can retain more fine details than the median filter with the same window size, as well as reduce the output mean square error which is caused mainly by edge shifting. To suppress nonimpulsive noise as well as impulsive noise, an L-type structure is combined with the MMF. The proposed filter generates the output by using asymmetric α-trimming on the MMF output. In addition to noise suppression, it can enhance the gradient of blurred edges provided that the size of the window is greater than the transition width of the blurred edges. Experimental results for one- and two-dimensional signals are presented