[gdal-dev] Warping an image using GDALAutoCreateWarpedVRT() and other questions

[Even Rouault]

Le Thursday 09 July 2009 20:38:13 Adeel Raza, vous avez écrit :
> Hi,
>
> I'm using the "GTiff" driver and using it to create an empty file.
> Afterwards I write data to the (one and only) rasterband using a float
> array as the buffer (the GDALDataType I use for the GDALDataset is
> GDT_Float32 - I believe that should be ok). Then I warp the image
> using  the GDALAutoCreateWarpedVRT() function. I have noticed that after
> the image is warped, the number of pixels are not the same as the
> original TIFF image. Is this normal? I know this sounds like a stupid
> question, but I need to be sure.

Yes it is normal if you change projection. The shape and resolutions of the 
output will not be the same as the input.

>
> Secondly I was also wondering how the GDALRasterBand::RasterIO()
> function feeds data into the raster band. I make a function call to the
> RasterIO function according to the following code:
>
>     GDALDriver *tifDriver;
>     GDALDataset *srcDataset;
>
>     tifDriver = (GDALDriver *)GDALGetDriverByName("GTiff");
>     assert(tifDriver != NULL);
>     srcDataset = tifDriver->Create("srcDataset.tif", nCols, nRows, 1,
> GDALDataType::GDT_Float32, NULL);
>     assert(srcDataset != NULL);
>
>     srcDataset->GetRasterBand(1)->RasterIO(GF_Write, 0, 0, nCols, nRows,
> realCellVals, nCols, nRows, GDALDataType::GDT_Float32, 0 , 0);
>
> Both nCols and nRows are unsigned shorts and realCellVals is defined as
> "float *realCellVals = new float[nCols*nRows];". I have filled
> realCellValls so that the row at the top of the raster image is the
> first row and the row at the bottom of the raster image is the last row
> in the array. Also columns indices increase from left to right. Will
> this implementation work properly with the RasterIO() function or do I
> need to arrange my array in another way? 

I guess you could try and check yourself ;-) If I've understood your 
description, this should work as you expect.

> The reason why this is critical 
> is so that I can create a north up TIFF image, warp it to the desired
> projection system, and get an affine geo transform of the warped north
> up image so that GT[2] and GT[4] are 0 (GT is the returned affine geo
> transform from the function - GDALDataset::GetGeoTransform()).

GDALAutoCreateWarpedVRT() will return a dataset with such characteristics. (to 
be exact, it won't necessary be exactly "north up", it depends on the 
destination projection, but I play a bit on words)

>
> Lastly I initially tried the same approach using a virtual database and
> creating it in memory using the GDALDriver::Create() function. I tried
> using the GDALRasterBand::RasterIO() function on the database but I got
> an error saying "writing to VRTWarpedRasterBands is not supported".

A VRT file can be created (the XML description), and data can be read from it. 
But writing pixels into it doesn't make any sense of course, so the error 
message is logical.

> Therefore I had to use the tiff format to warp the image. Is there
> anyway to bypass this so that I don't have to create a temporary tiff
> file to warp my image?

Yes, I see 2 possibilities : 
* use the MEM driver instead of the GTiff driver to create an in-memory source 
dataset (easiest way)
* use the /vsimem virtual filesystem to creaty an in-memory GeoTIFF file

>
> Any help would be greatly appreciated. Thanks.
>
> Best Regards,
> Adeel.
>
>
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/gdal-dev