| CGGM.mean {epsi} | R Documentation |
Computation of two dimensional redescending M-kernel estimators.
CGGM.mean(data, h, g=NULL, silent=FALSE) CGGM.lts(data, h, g=NULL, trim=0, silent=FALSE) CGGM.autoscale(data, h, silent=FALSE)
data |
numerical matrix of observed data. |
h |
positive number. Bandwidth for the kernel. |
g |
optional positive number. Scale parameter. If g is NULL
(default), the scale parameter is determined automatically by the
function CGGM.autoscale. |
trim |
optional number within [0,1). Part of observations trimmed
by CGGM.lts |
silent |
optional boolean. If true, CGGM.autoscale
produces no output. |
CGGM.mean implements a corner-preserving smoothing method
introduced by Chu et al. (1998) which is based on a redescending
M-kernel estimator. As kernel and score function the density of the
standard normal distribution is used. A robust version of this
estimator is introduced by Hillebrand (2002) and implemented in
CGGM.lts.
CGGM.autoscale calculates the median of the interquartile
ranges within the 'windows' used in CGGM.mean and
CGGM.lts. This can be used as scale parameter.
Return value is a numerical matrix containing the smoothed data.
Tim Garlipp (garlipp@mathematik.uni-oldenburg.de), Martin Hillebrand (Martin.Hillebrand@gmx.de)
Chu, C.K., Glad, I.K., Godtliebsen, F., Marron, J.S. (1998) Edge-Preserving Smoothers for Image Processing, J. Amer. Statis. Assoc. 93, 526-541.
Hillebrand, M. (2002) On Robust Corner-Preserving Smoothing in Image Processing, Carl-von-Ossietzky-Universität Oldenburg, Dissertation (http://docserver.bis.uni-oldenburg.de/publikationen/dissertation/2003/hilonr03/hilonr03.html).
y <- matrix(rep(0,60*60),nrow=60) y[21:40,21:40]<-1 y <- y + matrix(rnorm(60*60,0,0.1),nrow=60) image(y,col=gray(seq(0,1,1/255))) ymean <- CGGM.mean(y,0.04) image(ymean,col=gray(seq(0,1,1/255)))