Algorithms for max and min filters with improved worst-case performance

This brief presents three algorithms for implementing a running max/min filter of arbitrary order K, in which the average computation time per sample is asymptotically independent of K when the input data samples are statistically independent and identically distributed. The algorithms differ in their worst-case performance when acting on correlated input signals: for one of the algorithms, the computational complexity is of order K, while for the other two it is of order log(K). This brief gives the theoretical and experimental performance for a number of real and synthetic input signals.