[gdal-dev] Best method to extract regions from a raster into a memory buffer

Aurélien Albert aurelien.albert at outlook.com
Thu Oct 11 07:56:33 PDT 2018


Yes, you're right, I may need some extra pixels on the boundaries (depend on the resampling method).


Does not GDAL manage this by itself ? I mean, if you ask for a warp, it doesn't read all necessary pixel to do the resampling ?


Since GDAL 2.0, there is some API changes, and I'm looking for the most efficient way to do my operation : extract a region specified with geographic coordinates (in the same SRS as the data source) to a fixed size pixel buffer. Resampling should occur because the pixel count of the extracted region (source) and pixel count of the buffer (destination) will not fit.



________________________________
De : gdal-dev <gdal-dev-bounces at lists.osgeo.org> de la part de Joaquim Manuel Freire Luís <jluis at ualg.pt>
Envoyé : jeudi 11 octobre 2018 16:47:27
À : gdal-dev at lists.osgeo.org
Objet : Re: [gdal-dev] Best method to extract regions from a raster into a memory buffer


Hi,



I think you need an extra point. If you will be doing resampling than you need extra row/columns for the boundary conditions (one for linear, two for spline/bicubic). Regarding 2) the options are:  round to closest node, or to the inner or to the outer node.



Joaquim





From: gdal-dev <gdal-dev-bounces at lists.osgeo.org> On Behalf Of Aurélien Albert
Sent: Thursday, October 11, 2018 11:28 AM
To: gdal-dev at lists.osgeo.org
Subject: [gdal-dev] Best method to extract regions from a raster into a memory buffer



Hi,



I'm doing C++ applications and I'm familiar with GDAL 1.7 C/C++ API. For a new application, I'm switching to GDAL 2.2.

In this application, I have the need to extract a region from a raster :



Let's say I have 4000px x 4000px elevation raster, covering the whole earth, in EPSG 4326.

Now I want to extract the rectangle region between two points (lat_1/long_1) and (lat_2/long_2) in a 128px x 128px tile.

I don't want to extract this region in a file, but in memory, to create a data buffer and then work on this buffer with my application.



Which is the best/more performant way to do this ?



Knowing that :

 1/ the region extraction may need some resampling, so I need to be able to configure the resampling method

 2/ the extracted region boundaries do no necessary fit on exact pixels of the original dataset

 3/ the SRS will not be modified, but I may need a vertical datum shift (EGM96 to WGS84 for example)

 4/ the extraction will be repeated a lot (a lot !) to extract several "tiles" with different bounds : the application can work on really big rasters, so I prefer to avoid complex objects initialization for each extraction



Should I :



- create a wrapped dataset (VRTDataSet) and make RasterIO on it ? This seems to solve 1/ and 3/ and 4/, but what about 2/ ?



- use a WarpOperation and then execute WarpRegionToBuffer method ? This seems to solve 1/ and 2/ and 3/, but what about 4/ ?



Any help will be appreciated.





Thanks


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20181011/cd899a31/attachment-0001.html>


More information about the gdal-dev mailing list