Input Compression and Efficient Algorithms and Architectures for Stack Filters

The output of a stack filter is always one of the sample points in its input window. Which sample is chosen depends on the relative ranks and positions of the samples in the window, not on the actual magnitude of the samples. The sample points in a window of size w can therefore be mapped to the integers 1 through w before the stack filter is applied. Once the compressed data has been stack filtered, the compressed sample chosen is then mapped back to its original value. This new approach to stack filtering can be used to reduce the complexity of algorithms that have been proposed for stack filters. For example, in the threshold decomposition implementation of stack filters the number of binary filters needed is reduced from 2/sup b/ to w, where b is the number of bits used to represent the input data.