[GRASSLIST:2001] Re: Image classification

[Glynn Clements]

Miha STAUT wrote:

> I have got 485 8-bit TIFF ortophotos that I would like to merge in a unique 
> image or maybe 3 to 5 images. The procedure I used was r.in.gdal --> r.patch 
> - NO PROBLEM WITH THAT. The problem is that the images are not of a very 
> good quality and have very different values for the same land class, i.e. 
> some are darker and some lighter + they are shot at different times of the 
> day (hilly region). That means the shadows extend differently.
> 
> To get a better view of the area I tried to use r.colors col=grey.eq, but 
> the edges of the original images are still very visible. Then I thought 
> about i.gray.scale. As I understood i.gray.scale changes the values in the 
> image. In that way I would manage to get together the images with the values 
> much more alike.
> 
> But there is no command like that any more. ???

r.colors alone won't help, because it only changes the colour table,
which isn't used by r.patch.

However, you could use r.colors followed by:

	r.mapcalc 'output = #input'

to replace each cell's value with its intensity level from the colour
table.

> Any better idea? I am not really and expert in image analysis. The aim is to 
> transform the images in a way the same values would indicate the same land 
> classes. After that get some spectral signatures and try to classify the 
> image in a more or less correct way.

If relief shading is an issue, you could try using shade.rel.sh on a
DEM to compute the expected illumination, then computing the ratios. 
That assumes that you can figure out how to transform either the DEM
or the computed intensity map to match the orthophoto.

-- 
Glynn Clements <glynn.clements at virgin.net>