pnonlocalmedianfiltering

Performs a non local median filtering on image.

Synopsis

pnonlocalmedianfiltering sigma nb_iter connectivity [im_in|-] [im_out|-]

Description

Performs a non local median on image im_in. Let $f$ be im_in, this algorithm performs iteratively:
$f(u)^{t+1} = Med_{v \sim u}{v \tilde u}(\sqrt{w(u, v)}(f^t(v) - f^t(u))) + f^t(u)$
with $u$ a pixel location, $v$ a neighbor of $u$, $w(u, v)$ the weight between $u$ and $v$ (which can be a similarity measure, a distance, etc.).

We compute weights between a pixel value and its neighbors (here with an exponential similarity measure):
$w(u, v) = \exp^{-||f(v) - f(u)||^2 / \sigma^2}
With $w(u, v) = 1$ you get the median filter.

Parameters

sigma: sigma used for the similarity measure. The greater the sigma value, the lower the filtering.

nb_iter: number of iteration of the algorithm.

connectivity: neighborhood, can be 4 or 8.

Inputs

im_in: a 2D image.

Outputs

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

Result

Returns SUCCESS or FAILURE.

Example

Applies a non local median filter to tangram.pan, with 8-connectivity neighborhood, 10 iterations, and sigma = 15:
```
pnonlocalmedianfiltering 15 10 8 tangram.pan out.pan
```

C++ prototype

template Errc lplRegularization(Imx2d &imgIn, Imx2d &imgOut, float sigma, int connectivity, IRunThroughImage * imageRunner, IPerformReg * regPerformer);

Version française

Filtrage median non local.

Author: Matthieu Toutain