[gdal-dev] Re: [Fwd: Re: Warping an already georeferenced image with control points

Brent Fraser bfraser at geoanalytic.com
Thu Feb 12 14:09:22 EST 2009


  I think what you want gdalwarp to do is [Delauney] image triangulation.  Not an uncommon (or unreasonable) request.  I occasionally have a need for triangulation when mosaicking several satellite images.  Its visually pleasing to have the roads match up precisely where the images overlap...

Brent Fraser

Even Rouault wrote:
> Jan,
> (I'm CC'ing the list)
> I'm not sure what you mean with adapting the pixelsize, now that the output 
> has only GCPs and no more geotransform matrix.
> As far as including this option in baseline gdal_translate.cpp, I'm currently 
> not really enthousiastic, as there are lots of way to get the result achieved 
> (what is done with that patch could be done similarly with GDAL python API 
> for instance) and the choice of corners is something rather arbitrary. Why 
> should you trust the georeferenced coordinates of the corner points as 
> reliable GCPs ? Why not some regularly discretized points along the edges ? 
> Or a regular grid over the whole raster ? etc. etc. So, I'd prefer hearing 
> from other GDAL developers or users that this patch is a sensical addition 
> before commiting it.
> Even
> Le Thursday 12 February 2009 11:47:00, vous avez écrit :
>> Hi Even,
>> Thanks a lo. Just one small addition: the pixelsize has also to be
>> adapted, else you get an image with false dimension. If it's not too
>> much to add that, I'm going to try it out and will let you know the
>> results. If it works, I have indeed a practical solution for my problem.
>> Even so, I am not convinced that this is an uncommon problem. Consider a
>> very basic GIS functionality: edge matching.  You have digitized a map
>> from many sheets, and at the end the georeferenced images don't match
>> precisely at the edges. Edge-matching functions (ArcGIS has lots of
>> them) ensure that the edge of a feature on one map perfectly matches the
>> feature edge on the adjacent map. This is exactly the kind of thing you
>> can do with gdalwarp using control points on a georeferenced image. It
>> is also more or less the thing I want to do with my historical maps.
>> Would this be an argument to preserve the boundaries of a georeferenced
>> image, when adding control points?
>> Jan
>> Even Rouault wrote:
>>> Jan,
>>> I concur with Jukka's analysis. Your need is rather uncommon, and I also
>>> thinks it shouldn't be a default behaviour. However, adding that
>>> capability to gdal_translate is quite easy. I've attached a patch for
>>> gdal_translate.cpp that adds the capability of adding the 4 corners as 4
>>> additional GCPs with the "-add_corners_as_gcps" option. You might give it
>>> at try (it applies cleanly on GDAL 1.6.0 and trunk). I'm not sure yet if
>>> it is valuable enough to include it in baseline gdal_translate.cpp.
>>> Best regards
>>> Le Wednesday 11 February 2009 15:59:48 Jukka Rahkonen, vous avez écrit :
>>>> Jan Hartmann <j.l.h.hartmann <at> uva.nl> writes:
>>>>> Sorry to keep moaning about this, but I need an indication what's going
>>>>> on here. Mind, I don't need an immediate solution: for the time being I
>>>>> have a workaround. Just an idea whether this a real problem, a dumb
>>>>> question, something that can be handled in the foreseeable future (or
>>>>> not), perhaps with adequate funding. Everything is better than talking
>>>>> to a blind wall.
>>>>> Sorry again Frank,
>>>>> Jan
>>>> Hi,
>>>> I think that your need to keep the original extents but add more ground
>>>> control points inside the image frame is rather uncommon. Doesn't it
>>>> mean that you trust that the image corners are correctly warped to a new
>>>> projection, but there are local distortions in the middle of image which
>>>> should be corrected with a few extra ground control points? For my mind
>>>> it shoudn't be the default behaviour of gdal but it might be usable as
>>>> an user selectable option sometimes.
>>>> I know that missing gcp's at the image corners often leads to very odd
>>>> result with polynomial warping because the formula shoots over.  Even
>>>> unaccurately placed gcp's could help a lot in preserving the original
>>>> shape of the map. I guess that you are perhaps playing with scanned
>>>> historical maps?  I have a few old scanned parcel maps which are
>>>> covering just the area of the farm, and two of the map corners are just
>>>> white background. It is impossible to measure any real ground control
>>>> points from the corners because there is nothing on the map to compare
>>>> with, and warping with all gcp's on the middle area makes really funny
>>>> looking curves into the hand drawn rectangle framing the original map.
>>>> -Jukka-
>>>> _______________________________________________
>>>> gdal-dev mailing list
>>>> gdal-dev at lists.osgeo.org
>>>> http://lists.osgeo.org/mailman/listinfo/gdal-dev
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/gdal-dev

More information about the gdal-dev mailing list