Multivariate median filters and their extensions

Median filters are based on running median estimates of location, and have been useful for edge-preserving and detail-retaining smoothing of noisy signals. The outlierresistance of the univariate median estimate of location and its extensions is also of interest in multivariate filtering applications. In this paper we consider one new class of multivariate median filters, based on the radial medians. The radial medians and the minimum-distance median are then investigated for their use in defining multivariate L-filters, for better smoothing in the presence of Gaussian noise