pdenoisePDE

Performs anisotropic smoothing on image.

Synopsis

pdenoisePDE nb_iter amplitude sharpness anisotropy alpha sigma [im_in|-] [im_out|-]

Description

pdenoisePDE regularizes color or multispectral images. This operator acts as an image regularizer, by doing anisotropic smoothing of the input multi-valued image. Such regularization technique is very efficient to remove local image artifact such as noise or compression artifact. It is also anisotropic and thus the smoothing preserves the important image structures such as edges, corners or discontinuities.

The processing time can be very high depending on the given parameter values.

Parameters

nb_iter defines the number of iterations. The greater is the number of iterations, the stronger is the smoothing. A typical value is 2.
amplitude defines the smoothing amplitude for one iteration. The greater is the amplitude, the quicker is the processing. In general, values are from 5..200.
sharpness defines the edge contrast of the edges to be preserved. The greater is the value, the more there are contours. A value of 0 indicates that each pixel is smoothed with the same strength eventually in different directions. Typical values are from 0..2.
anisotropy defines anisotropic degree. An anisotropic smoothing (anisotropy=1) is oriented along the edge direction. An isotropic smoothing (anisotropy=0) is not oriented. According to the noise type, it can be useful to restrict the anisotropic to avoid texture effects apparition. Allowed values are 0 or 1.
alpha defines the noise scale. It represents the estimated variance of the noise. A typical value is 0.1.
sigma defines the image geometry scale. Before each iteration, the local image geometry is evaluated. The sigma parameter corresponds to a pre-smoothing of this geometry. The greater is sigma, the less details are preserved but the more the smoothing is coherent. In general, a value less than 1 is convenient. A typical value is 0 8.

Inputs

im_in: an image.

Outputs

im_out: an image of the same type as the input image.

Result

Returns SUCCESS or FAILURE.

Examples

Denoises tangram.pan with only 2 iterations:

   pdenoisePDE 2 100 2 0.7 0.1 0.8 tangram.pan a.pan

C++ prototype

Errc PDenoisePDE( const Imx3d &ims, Imx3d &imd, Long nb_iter, Float amplitude, Float sharpness, Float anisotropy, Float alpha, Float sigma );

Version française

Régularisation d'images multivaluées par lissage anisotrope basé EDP.

Inportant notice

The source code of this Pandore operator is governed by a specific Free-Software License (the CeCiLL License), also applying to the CImg Library. Please read it carefully, if you want to use this module in your own project (file CImg.h).

IN PARTICULAR, YOU ARE NOT ALLOWED TO USE THIS PANDORE MODULE IN A CLOSED-SOURCE PROPRIETARY PROJECT WITHOUT ASKING AN AUTHORIZATION TO THE CIMG LIBRARY AUTHOR ( http://www.greyc.ensicaen.fr/~dtschump/ )

Reference

D. Tschumperlé, "Fast Anisotropic Smoothing of Multi-Valued Images using Curvature-Preserving PDE's", Cahier du GREYC No 05/01, Avril 2005.

Author: D. Tschumperlé