c:\Programme\ImageMagick-5.5.7-Q8>td -h
td : Implementation of the Tschumperle-Deriche's PDE for 2D Vector-valued image
restoration,
as described in the CVPR'2003 paper : 'Vector-valued Image Regul
arization with PDE's :
A Common Framework for Different Applications'
This code is distributed only for experiments and comparisons pu
rposes. This is not legal to use it or redistribute it
without the permission of the INRIA Sophia-Antipolis Lab.
Please contact the authors :
David.Tschumperle@sophia.inria.fr ,
or
Rachid.Deriche@sophia.inria.fr
for additional informations.
You are also pleased to cite our CVPR'2003 article if you are us
ing results obtained with this program,
in your own papers or slides or whatever...
Compiled at (Oct 7 2003)
-i = NULL : Input image. Native supported file formats are .pg
m and .ppm files,
but you can also read common image formats by in
stalling the 'ImageMagick' package
(see
http://www.imagemagick.org)
-m = NULL : Mask image (for inpainting). Pixels that will be i
npainted in the input image
must be non-null in the mask image. Same support
ed image formats as with option '-i'
-o = NULL : Output file. This is the filename corresponding to
the saved regularized image.
You can also save an image sequence correspondin
g to the PDE evolution (see below).
-dt = 20 : Adaptive time step. This is the maximum number of
intensity variation allowed during an iteration.
When using the filtering-based scheme, this opti
on corresponds to a fixed time step and
It must be of course reduced (close to 1 for cla
ssical color images
-alpha = 0.2 : Image gradient smoothing. A smoothing is applied o
n the image gradients in order to detect
a more coherent local geometry
-sigma = 1.5 : Structure tensor smoothing. The blur applied on th
e structure tensor field. Should be high for inpainting
-beta = 1e-006 : Beta. A value close to zero
-a1 = 0.7 : Diffusion coef along minimal variations. A low val
ue will diffuse more than a high value
-a2 = 1 : Diffusion coef along maximal variations. A low val
ue will diffuse more than a high value
-ng = 0 : Pre-process : Add gauss noise to image before star
ting regularization
-nu = 0 : Pre-process : Add uniform noise, before starting r
egularization
-ns = 0 : Pre-process : Add salt&pepper noise, before starti
ng regularization
-stats = false : Display image statistics at each iteration
-filter = false : Activate/Desactivate the filtering-based scheme, b
ased on local filtering with oriented gaussian kernels
-iter = 1000 : Number of PDE iterations that will be computed
-save = 0 : Iteration saving step. If >0, this parameters allo
ws to save the iterated images each 'nb' iterations
-visu = 1 : Visualization step. Set the image visualization up
dating step.
-h = true : Display this help
