[gdal-dev] Adding gcps to a raster using gdal_translate can result in data loss.

[Even Rouault]

Jukka' suggestion is a good one. Using -tps solver or affine 
approximation -order 1 fixes the issue. The default behaviour with GCPs 
when there are 6 or more of them is to use the second order polynomial 
adjustment. The issue here is when computing output bounds. We actually 
compute two 2nd order polynomial adjustments: one from source image 
coordinates to georeferenced ones, and one from georeferenced ones to 
source image. The former one is used to compute the output bounds, but 
the second one is used to do warping. The issue with 2nd order 
polynomial is that those two transformers are generally not the inverse 
of each other... We should likely compute the source image coordinates 
-> georeferenced coordinates as an iterative inversion of the other one, 
so they are invertible. The first order polynomial / affine 
transformation hasn't that issue due to how affine transformation works. 
The TPS one isn't likely perfectly invertible but TPS transformation is  
guaranteed to be exact at GCPs, so that diminishes the effect. But 2nd 
order polynomial has overshoots particularities that can cause the issue 
observed.

Another workaround is to specify the desired output bounds with -te

Anyway, Hays, could you file an issue about that in 
https://github.com/OSGeo/gdal/issues, ideally copying over your web page 
(or pointing to it if it can remain stable), and pasting my above comment.

Even

Le 30/03/2021 à 18:51, Rahkonen Jukka (MML) a écrit :
>
> Hi,
>
> I did not have a close look into your ground control points but 
> perhaps they are not very well distributed and therefore the default 
> warping algorithm eats one corner of the image. Have a try with thin 
> plate splin (-tps) instead.  There may be some bug, it does not feel 
> right that some source data are lost with default algorithm.
>
> -Jukka Rahkonen-
>
> *Lähettäjä:*Hays Barrett <hays.barrett at gmail.com>
> *Lähetetty:* tiistai 30. maaliskuuta 2021 19.32
> *Vastaanottaja:* Rahkonen Jukka (MML) 
> <jukka.rahkonen at maanmittauslaitos.fi>
> *Kopio:* gdal-dev at lists.osgeo.org
> *Aihe:* Re: [gdal-dev] Adding gcps to a raster using gdal_translate 
> can result in data loss.
>
> I put together a page that better explains the issue and used one 
> of the actual images from our project.
>
> https://haysbarrett.com/gdal_translate/ 
> <https://haysbarrett.com/gdal_translate/>
>
> Thanks for looking into this for me.
>
> -Hays
>
> On Fri, Mar 26, 2021 at 2:59 PM jratike80 
> <jukka.rahkonen at maanmittauslaitos.fi 
> <mailto:jukka.rahkonen at maanmittauslaitos.fi>> wrote:
>
>     Hi,
>
>     Please give an exact example. Preferably with a raster that can be
>     created
>     from command line with gdal_create followed by other required
>     commands. Or
>     alternatively Python code that re-produces the issue.
>
>     -Jukka Rahkonen-
>
>
>
>     Hays Barrett wrote
>     > If the gcps stretch the image beyond the dimensions of the
>     original raster
>     > it appears to clip the data that sits outside of the original raster
>     > dimensions rather than increase the raster dimensions to fit the
>     stretched
>     > data .
>     > Is this the expected behavior?
>     > If so is there a way to increase the dimensions of a raster so
>     there is no
>     > data loss?
>     > Thanks,
>     > -Hays
>     >
>     > _______________________________________________
>     > gdal-dev mailing list
>
>     > gdal-dev at .osgeo <mailto:gdal-dev at .osgeo>
>
>     > https://lists.osgeo.org/mailman/listinfo/gdal-dev
>     <https://lists.osgeo.org/mailman/listinfo/gdal-dev>
>
>
>
>
>
>     --
>     Sent from:
>     http://osgeo-org.1560.x6.nabble.com/GDAL-Dev-f3742093.html
>     <http://osgeo-org.1560.x6.nabble.com/GDAL-Dev-f3742093.html>
>     _______________________________________________
>     gdal-dev mailing list
>     gdal-dev at lists.osgeo.org <mailto:gdal-dev at lists.osgeo.org>
>     https://lists.osgeo.org/mailman/listinfo/gdal-dev
>     <https://lists.osgeo.org/mailman/listinfo/gdal-dev>
>
>
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/gdal-dev

-- 
http://www.spatialys.com
My software is free, but my time generally not.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20210330/0c2ea5ed/attachment.html>