A computationally efficient implementation of the L-1 vector median filter

In spite of their good filtering characteristics, the use of vector median filters is often limited by their high computational complexity. This work describes a fast algorithm which permits to considerably reduce tile computational burden of the vector median filter based on 1-norm. The validity of the algorithm is demonstrated both by means of theoretical results and experimental tests. In particular. it is shown that the proposed technique permits to obtain the filtering accuracy of vector medians based on the aggregate sum of distances at a computational complexity which is typical of filters based on reduced ordering.