2D-examples |
Edge Preserving Smoothing
|
Nonlinear Gaussian Filter Chains
|
|
Prof. Dr. V. Aurich
Institut für Informatik
Heinrich-Heine-Universität
Düsseldorf
|
Scale Space Example
|
An optimal nonlinear Gaussian filter chain has only two free parameters SigmaX and SigmaZ which determine the coarseness and the contrast of the signal fluctuations which are to be eliminated.
Hence they span a two-dimensional scale space. This is illustrated by the following table of filter chain outputs for various parameter choices. Click to the images to view them in their original size of 512x512.
|
|
Original |
|
|
|
SigmaX=1 |
SigmaX=2 |
SigmaX=3 |
SigmaZ=16 |
![[b116]](baboon/s_b116fffF.gif) |
![[b216]](baboon/s_b216fffF.gif) |
![[b316]](baboon/s_b316fffF.gif) |
SigmaZ=32 |
![[b132]](baboon/s_b132fffF.gif) |
![[b232]](baboon/s_b232fffF.gif) |
![[b332]](baboon/s_b332fffF.gif) |
SigmaZ=64 |
![[b164]](baboon/s_b164fffF.gif) |
![[b264]](baboon/s_b264fffF.gif) |
![[b364]](baboon/s_b364fffF.gif) |
|
Example for the Smoothing with a Non-Horizontal Gaussian Filter Chain
|
 |
 |
 |
|
|
true data |
noisy data |
filter result |
|
This is the example shown in the article on smoothers for discontinuous signals of Winkler, Liebscher, Aurich, p.29.
|