Brendan wrote:t00fri wrote:Normally, at such high resolution one has to run some (non-linear) stretching corrections here and there to optimize alignment. You apparently have not done that.
By non-linear scaling, do you mean scaling the original 512x512 BlueMarble tile differently along the horizontal and vertical lengths before overlaying the new images on it? Or is it more than that?

Redfish and I were talking about what non-linear stretching meant.
Brendan
Redfish, Brendan,
here come some more explanations:
Consider for simplicity just the horizontal matching of two textures by rescaling one of them.
Let the pixel locations p and width of the texture L be
0 <= p <= L
linear rescaling means matching via a
global rescaling factor s
p' = s*p, L' =s*L
Such that the pixels are mapped
linearly.
What often happens, however, is that after matching the features near one end of the texture, the matching gets worse e.g. in the middle...What is actually needed in such cases is a
non-linear rescaling,
p'=s(p)*p, L'=s(L)*L
Well known problematic examples are the Viking/USGS (colored) Mars texture and the MOLA elevation maps.
There are various possible strategies to determine the best scaling function s(p).
-- Fiddling around...
-- using ImageMagick scripting (e.g. Perl,...)
--...
Bye Fridger