A separable median filter for image noise smoothing

This paper investigates some properties of the separable filter resulting from successive applications of a one-dimensional median filter on the rows and columns of an image. Although the output of this separable filter is not identical to the corresponding nonseparable two-dimensional median filter with a square window, its performance in image noise smoothing is close. In particular, its effectiveness in smoothing noise and its behavior with edges are characterized and compared with those of the two-dimensional median filter. It is shown that the separable filter has a much simpler implementation in real-time hardware (at video rates, for example).