[postgis-devel] [raster] Raster wrapper ; Raster View

[Bryce L Nordgren]

Thinking about raster alignment today, I thought it might be nice to have a
"Raster View" object or "Raster Wrapper" object which could allow you to
index a raster using a different raster's indices. Say you have two rasters
which are offset by exactly three pixels in the x direction. Normally, to
reference the pixel in which covers the same spot on the ground, in rasters
A and B, you'd need to write

A[i+3, j] and B[i, j]

In this case, i & j are indices into B, and we're calculating the index of
the corresponding pixel in A. This is exactly how mapalgebra currently
handles the task.

What if we could "wrap" A in such a way that the wrapper itself remembered
the translation?

C = RasterView(A, dest=B)
C[i, j] = A[i+3, j]

Potentially, the "offset calculator" in mapalgebra could be replaced with
this wrapper code. It could also be used wherever rasters need to be
aligned. The advantage of using this "index arithmetic" is that the rasters
do not need to be resampled and stored somewhere. You would simply use the
coordinates in the "result" raster to lookup data in all of the inputs.
Ideally, the raster wrapper would behave exactly like a "regular" raster,
except the indices are transformed before use.

Now here's the best part: scrounging through GDAL documentation, I found

http://www.gdal.org/gdal__alg_8h.html#7671696d085085a0bfba3c3df9ffcc0a

void* GDALCreateGenImgProjTransformer()
>
> Create image to image transformer.
>
> This function creates a transformation object that maps from pixel/line
> coordinates on one image to pixel/line coordinates on another image. The
> images may potentially be georeferenced in different coordinate systems, and
> may used GCPs to map between their pixel/line coordinates and georeferenced
> coordinates (as opposed to the default assumption that their geotransform
> should be used).
>
There are three different versions. The treatment seems quite general, and
will handle different orientations and even different projections.

Could someone with more knowledge of PostGIS comment on the difficulty of
creating a raster wrapper using the above described GDAL functions? Would we
need another type in the database, or could we have two C implementations of
the raster PostGIS type?

Bryce
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20110617/a50e8c96/attachment.html>