[mapserver-users] Shift of half a pixel

[Fritz van Deventer]

Dear Whoever is interested,

 

We found the problem.

 

It was not in projection, not in the bounding box model. It wasn't even
in the software opening the WMS. 

 

Resampling defaults to Nearest Neighbour. At the level we look at the
images (sometimes at very high zoomlevel) this was noticeable. We
switched to Average resampling:

 

PROCESSING "RESAMPLE=AVERAGE"

 

This may not be the prettiest resampling method, but it was the best
solution between long processing time and the right location of the
individual pixels.

 

Thanks !

 

Fritz

 

________________________________

Van: Fritz van Deventer 
Verzonden: donderdag 7 juni 2012 9:08
Aan: 'mapserver-users at lists.osgeo.org'
Onderwerp: RE: Shift of half a pixel

 

Dear Steve,

 

Thanks for your reply. And this is my first time posting, so forgive my
lack of information.

 

I'm running mapserver 6.0.3 with gdal 1.9.0 (compiled with libecwj) on
Ubuntu 12.04.

 

In both QGis 1.7.0 and Mapinfo 10.5 I tried to load both the original
image (both as tif and ecw), and the served image in the same viewer. As
I toggled between the images (after drawing some reference points) I
could see the image shifted slightly in relation to the reference point.
I did for several corners, and the shift seems to be somewhat
inconsistent, moving in several directions in different areas of the
image.

 

But assuming a bounding box model problem, how would I be able to find
such a thing out?

 

Fritz

 

________________________________

Van: Lime, Steve D (DNR) [mailto:Steve.Lime at state.mn.us] 
Verzonden: maandag 4 juni 2012 23:35
Aan: Fritz van Deventer; mapserver-users at lists.osgeo.org
Onderwerp: RE: Shift of half a pixel

 

What version are you talking about? How do you know there is a shift?
That is, what is your reference? Is it possible there's a difference in
the bounding box model? For WMS the bbox represents the outside
dimensions of the pixels in an image. 

 

Steve

 

Note: MapServer uses a pixel center bbox model but that difference (with
WMS's representation) is taken into account in the WMS code.

 

From: mapserver-users-bounces at lists.osgeo.org
[mailto:mapserver-users-bounces at lists.osgeo.org] On Behalf Of Fritz van
Deventer
Sent: Monday, June 04, 2012 9:27 AM
To: mapserver-users at lists.osgeo.org
Subject: [mapserver-users] Shift of half a pixel

 

Hi all,

 

We are trying to get VHR satellite imagery to run as a WMS in different
GIS clients. When we compare the original image (we tried different
formats, tif, ecw, virtual formats etc.) to the hosted image we see a
shift in the pixels. Looks like a minor warp or transform. 

 

We are unable to find what is causing this error. We tried using
different output drivers, and different output and input formats. It
looks like serving with GDAL as output format and Imagemode on FLOAT32
gives the best result, however in black and white. 

 

If this is a known issue, is anyone able to tell me the (in)accuracy of
the shift? Or is it not supposed to happen altogether?

 

Regards,

 

Fritz van Deventer - http://neo.nl 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/mapserver-users/attachments/20120607/61533bc6/attachment.htm>