[GRASS-user] Landsat classification with CORINE CLC color codes?

[Hamish]

Dimos wrote:
> > > 44 CORINE CLC RGB  color codes are mentioned for each of the 44
> > > land use classes at:
  http://dataservice.eea.europa.eu/download.asp?id=14234&filetype=.csv
> > >
> > > Can we classify a Landsat RGB image based on these rgb color codes
> > > in GRASS GIS and how?

Markus: 
> > In my opinion it doesn't make much sense to use the RGB colors here
> > since they are arbitrary (well, ok, ideally "close" to natural colors).

I am trying to understand; there are 256^3 RGB possibilities (@ 8bits per
channel) and CORINE have selected 44 of those possibilities to give labels
to? Your chances of hitting those exactly are rather low. I suppose these
are meant to be 3D spectral peaks somehow with the land-use category doing
like a 3D form of a nearest-neighbor thiessen polygon.


It seems a rather lossy and sensitive approach, ie it would be better to
base it on all 7 bands & include some method to remove time of day/luminance
issues. e.g. specify color normalization step first [eg i.landsat.rgb],
or convert LANDSAT r,g,b pseudo-visual bands to HIS and classify on that
(I'm guessing that hue would be less sensitive to sun angle).


Nikos:
> In my humble opinion, there is no way in this case to completely avoid
> some manual digitisation of training samples, or edit the samples you
> will extract from CORINE.
....
> There is, on the web, a nice step-by-step which uses COREIN + i.smap on
> Landsat [1].
....
> [1] http://www.custom-scenery.org/Building-Scener.331.0.html

interesting.


Hamish