[mapserver-users] Bilinear Resampling Bug?

[Patrick Young]

Hello!

Before I put in a bug in GitHub, I was hoping someone could confirm if what
I am seeing is expected behavior.

I've made a small test case (https://github.com/youngpm/alphabug) where I
have two adjacent rasters that abut but do not overlap.  They are the same
solid color, but contain a alpha band that goes from transparent to opaque
along the border where they touch.

When I use bilinear resampling and zoom in, I see a pixel of overlap (see
https://github.com/youngpm/alphabug/blob/main/example/bilinear.png), which
manifests as a darker line when the layers are composited in
semi-transparent areas.  This doesn't occur with nearest neighbor
resampling (see
https://github.com/youngpm/alphabug/blob/main/example/nearest.png).

I've placed these sample rasters and a mapfile with shp2map command
examples that illustrate what I am seeing here via individual raster layers
and when using a tile index:

https://github.com/youngpm/alphabug

For another experiment, in QGIS, I loaded one of the geotiffs directly and
again via WMS.  With bilinear resampling, one sees that the WMS layer has
an additional 1/2 pixel arpon (the pixel size being the native pixel size)
all the way around compared with the directly loaded version, but they
agree exactly when nearest neighbor is used; bilinear appears to be
interpolating past the image boundary by half a pixel.

I'm pretty new to using mapserver so it may very well be user error.  My
other thought is that during compositing, there is some mixup between pixel
centers and pixel edges.

Thanks for any suggestions! Happy to file an issue if anyone believes this
is a bug and not expected behavior.

Patrick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/mapserver-users/attachments/20210914/c214461c/attachment.html>