Abstract

Median filters (MFs) (or the more general rankorder filters) have proved very powerful in image processing. However, rank-order filters are generally considered to be more time-consuming than linear, e.g., averaging, filters. Several fast algorithms were devised for on-line or off-line filtering. Some of them are based on histogram calculation, whereas others are based on bit-by-bit calculation of the median. The vector median filter (VMF), introduced earlier by the authors, is a two-parameter family of median-type filters that outputs at each position a set of elements, i.e., a median vector. The VMF has the same fundamental properties of the MF; i.e., it preserves edges while filtering out narrow impulses. The main advantage of the VMF is that it is computationally more efficient than the MF with a performance comparable to the MF. Furthermore, since the VMF is a two-parameter filter, it offers a wider range of filtering possibilities. Here, based on a histogram calculation, a fast VMF algorithm is described. Computer run times are presented for both fast MFs and fast VMFs.

© 1987 Optical Society of America