[GRASS-user] i.pca noise reduction

[Markus Metz]

On Sat, Oct 13, 2012 at 6:18 PM, Markus Neteler <neteler at osgeo.org> wrote:
> On Sat, Oct 13, 2012 at 3:46 PM, Markus Metz
> <markus.metz.giswork at gmail.com> wrote:
>> i.pca in GRASS 7 can now do noise reduction by reconstructing the
>> input bands from a user.defined subset of the principal components.
>
> This sounds like an interesting extension but the usage is not clear to
> me:
>
> # NC data set:
>
> g.region rast=lsat7_2002_10 -p
> i.pca input=lsat7_2002_10,lsat7_2002_20,lsat7_2002_30,lsat7_2002_40,lsat7_2002_50,lsat7_2002_70
> -f percent=90 out=lsat7_2002_reduced
>
> g.mlist rast | grep reduced
> raster map(s) available in mapset <neteler>:
attempt to explain:
lsat7_2002_reduced.1 = filtered lsat7_2002_10
lsat7_2002_reduced.2 = filtered lsat7_2002_20
lsat7_2002_reduced.3 = filtered lsat7_2002_30
lsat7_2002_reduced.4 = filtered lsat7_2002_40
lsat7_2002_reduced.5 = filtered lsat7_2002_50
lsat7_2002_reduced.6 = filtered lsat7_2002_70

with perc=90, 2 PCA scores will be used for reconstruction, that is
each cell of the first output band will be input[1] * PCA[1][1] +
input[1] * PCA[2][1] and each cell of the last (6th) output band will
be input[6] * PCA[1][6] + input[1] * PCA[2][6]. Any scaling and
centering is considered.
>
> Say, this is not entirely clear to me:
>   -f   Output will be filtered input bands
>         Applies inverse PCA after PCA

Added to the manual.
>
> (Tha manual has a replicated parameters section which needs to
> be fixed, too. I can do that as soon as I understand the new
> functionality).

It seems that replicated parameters are kind of a feature in imagery
modules... To be fixed.

Markus M