[GRASS-user] Pansharpening (how, and is it necessary?)

[Michael Perdue]

I can't answer for i.fusion.brovey specifically, but generally pan- 
sharpening in not recommended before classification as the process of  
image fusion generally does not preserve the Spectral integrity of the  
original data. There are fusion algorithms out there that claim to  
preserve the spectral integrity of the input data (Smoothing Filter- 
based Image Modulation being one that I've considered trying to  
implement in Grass) but when I played with SFIM, I found that it was  
far from perfect (visual artifacts around strong edges and the Blue  
channel wasn't well correlated with the input blue channel...  
Atmospheric effects?). Seems there is no free lunch. FWIW, SFIM and a  
Local correlation fusion are implemented in OSSIM.
I'm unfamiliar with Quickbirds channels, but the panchromatic channel  
usually overlaps with the lower resolution multispectral channels.  
i.e., on landsat 7, the panchromatic band encompasses the Spectral  
range from 500nm to 910nm (green, red & NIR). This results in the  
panchromatic band being highly correlated with the green, red and NIR  
bands. In fact, you can do a linear combination of the three which  
will closely match the Panchromatic band. The end result is that the  
panchromatic band really doesn't add anything new to the  
classification and are not normally added to the classification.
Something else to consider... you could try generating texture images  
and incorporating them into you group.

Cheers,

Mike

On 29-Jul-09, at 2:56 AM, Georg Kaspar wrote:

> I wonder wether pansharpening of each channel is necessary to  
> classify a
> QuickBird-Scene using i.maxlik and i.smap or wether I can just add the
> pan channel to the group.
> _If_ pansharpening is necessary: What is the best way to do it?
> i.fusion.brovey and (i.rgb.his -> exchange intensity -> i.his.rgb)  
> only
> work with three channels, so I'd have to do this twice with different
> combinations of channels!?
> What about principal components? Is it possible to transform them  
> back,
> so I could exchange the first component?
> Thanks!
> Georg
>
> _______________________________________________
> grass-user mailing list
> grass-user at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/grass-user