Graphics Reference
In-Depth Information
Figure
.
.
Upper row: detail of the original image and the same detail from the noisy version. Bottom
row: local constant and local quadratic reconstructions. See online version for colors
trast. It was also shown that, up to a constant, the procedure retains the best quality
of estimation reached within the iteration process at any point. Related results for
the local polynomial propagation-separation approach can be found in Polzehl and
Spokoiny (
).
Intheformpresentedhere,theprocedureis,fordimension d
,entirelyisotropic.
It can be significantly improved by introducing anisotropy adaptively (i.e., depend-
ing on the information about θ obtained in the iterative process) in the definition of
the location penalty.
A reference implementation for the adaptive weights procedure described in
Sect.
.
.
isavailable asapackage(aws)fromthe R-ProjectforStatistical Computing
(RDevelopment Core Team,
)at http://www.r-project.org/. Imageprocessing is
implemented in R-package adimpro, see Polzehl and Tabelow (
).
