[gdal-dev] Warping quality with small geolocation arrays

[Knut-Frode Dagestad]

Frank,

I created a ticket with the same description as below:
http://trac.osgeo.org/gdal/ticket/4907

My test cases are from self-made Raw-VRTs, so I have no immediate 
reproducable test image/script to attach. However, if useful, I can 
construct an example to attach in a couple of days.

Thank you for looking into this.

Best regards from Knut-Frode



Den 26.11.2012 22:14, skrev Frank Warmerdam:
> On 12-11-26 12:11 PM, Knut-Frode Dagestad wrote:
>> Dear list,
>>
>> Warping datasets with geolocation arrays works fine if the arrays have
>> the same
>> size as the other raster bands. However, if the geolocation arrays are
>> much
>> smaller, the quality is very bad.
>>
>> Say I have a 2000x2000 px dataset with 100x100 px sized geolocation
>> arrays.
>> Then a warped image have only 200x200 blocks of identical values, even
>> if the
>> size of the warped image is much larger. If I change the interpolation
>> algorithm (neareast, cubic...) the values change slightly, but the
>> real output
>> resolution remains low. This is the same whether using Python API or
>> gdalwarp.
>> As a workaround I convert the geolocation arrays to GCPs. This gives
>> decent
>> warping quality, but is a bit awkward.
>>
>> Is this the expected behavior of the present implementation of
>> geolocation
>> arrays in GDAL?
>
> Knut-Frode,
>
> The intention is that the geolocation information will be linearly
> interpolated from the geolocation array.  However, I gather you are not
> seeing that effect.
>
> Looking at the "backmap" it seems that this linear interpolation is not
> being done, even though it is (after a fashion) in the forward case
> transformation case.  I'd appreciate it if you could submit a bug on
> this issue, and I'll look into correcting it.  It is a pretty serious
> error.
>
> Best regards,