Running Median Algorithm and Implementation for Integer Streaming Applications

A novel algorithm is proposed to compute the median of a running window of m integers in O(lg lg m) time. For a new window, the new median value is computed as a simple decision based on the previous median, and the values removed and inserted into the window. This facilitates implementations based on data structures that support fast ordinal predecessor/successor operations. The results show accelerations of up to factors of six for integer data streaming in typical embedded processors.