recently I have added a standardized option for sampling interpolation
methods [1], based on lib/raster/sample.c and codes defined in
raster.h [2].
Some modules use 'bicubic' other 'cubic' value. What do you think that
is more preferable? It's bicubic interpolation [3] accomplished using
cubic convolution, right?
recently I have added a standardized option for sampling interpolation
methods [1], based on lib/raster/sample.c and codes defined in
raster.h [2].
Some modules use 'bicubic' other 'cubic' value. What do you think that
is more preferable?
Whichever is used, it should be consistent. I.e. either
nearest/linear/cubic or nearest/bilinear/bicubic.
The former might be preferable if we want to extend it to 3D
interpolation (i.e. use nearest/linear/cubic regardless of the
dimensionality, rather than using separate 2D/3D terminology).
It's bicubic interpolation [3] accomplished using cubic convolution,
right?
recently I have added a standardized option for sampling interpolation
methods [1], based on lib/raster/sample.c and codes defined in
raster.h [2].
Some modules use 'bicubic' other 'cubic' value. What do you think that
is more preferable?
Whichever is used, it should be consistent. I.e. either
nearest/linear/cubic or nearest/bilinear/bicubic.
The former might be preferable if we want to extend it to 3D
interpolation (i.e. use nearest/linear/cubic regardless of the
dimensionality, rather than using separate 2D/3D terminology).
I fully agree, otherwise we will end up using bilinear, trilinear,
bicubic and tricubic.
Best regards
Soeren
It's bicubic interpolation [3] accomplished using cubic convolution,
right?
recently I have added a standardized option for sampling interpolation
methods [1], based on lib/raster/sample.c and codes defined in
raster.h [2].
Some modules use 'bicubic' other 'cubic' value. What do you think that
is more preferable?
Whichever is used, it should be consistent. I.e. either
nearest/linear/cubic or nearest/bilinear/bicubic.
The former might be preferable if we want to extend it to 3D
interpolation (i.e. use nearest/linear/cubic regardless of the
dimensionality, rather than using separate 2D/3D terminology).
I fully agree, otherwise we will end up using bilinear, trilinear,
bicubic and tricubic.
Whichever is used, it should be consistent. I.e. either
nearest/linear/cubic or nearest/bilinear/bicubic.
The former might be preferable if we want to extend it to 3D
interpolation (i.e. use nearest/linear/cubic regardless of the
dimensionality, rather than using separate 2D/3D terminology).
OK, done in r56327, any review welcome of course. Probably would make
sense to incorporate into raster lib as INTERP_TYPE (see
lib/raster/sample.c) other methods, at least `lanczos`.
I am not sure if we can replace all occurrences in the
manual automatically (eg. by `sed`). Probably it will need some
extra care(?)
perhaps some combination of 'grep -rI' for hits, a quick look at
the file it found, then 'sed -i -e' on the file if it's ok? could
be pretty quick for something as rare as bilinear. When describing
the method it is perhaps good to leave some mention of "bilinear"
and "bicubic" in the help pages, to aid with web/module description
searches where "linear" and "cubic" would be too generic terms to
get to the right answer quickly.
see also the grass-addons/tools/svngrep script to help avoid
some false positives in doing the recursive greps.