gdaladdo does not ignore NODATA values

[jszatmary@…]

Contrary to documentation gdaladdo does not ignore NODATA values when averaging. This was confirmed by warmerdam@pobox.com (via e-mail discussions.) Additionally, as a note, no special case handling appears to exist in the overview code for RGBA TIFF files (a special-case can be seen for RGB.)

[Even Rouault]

The attached patch handles the case where the raster band has a NODATA value while computing average. I've only fixed the 'simple type' case, as I'm not sure of if it's revelant for the 'complex type' case. It fixes the problem for int16 tiff images converted from DTED with nodata samples. As the ticket I've submitted to mapserver about WCS interpolation (​http://trac.osgeo.org/mapserver/ticket/2181), it could be interesting to have a 'STRICT_NODATA' mode where the NODATA value is not simply ignored but also propagated. I'll maybe propose a patch for this.

[Even Rouault]

See also ticket "[GDAL] #1721: [PATCH] GTiffDataset::IBuildOverviews doest not propagate NODATA to overview rasterband" : http://trac.osgeo.org/gdal/ticket/1721

[Even Rouault]

An improved version of the patch that supports STRICT_NODATA mode. The '-strict_nodata' flag is added to gdaladdo. Man page updated accordingly.

[Even Rouault]

Cumulative patch that adds minimum and maximum subsampling, only for simple data type (I guess minimum and maximum don't make sense for complex data type...)

[Even Rouault]

I've commited in trunk in r12497 something very close to gdal-1.4.2-fix-overview-nodata-with-strict-nodata.patch.

[warmerdam]

I've removed the -propogate_nodata option which I feel is a bit to specialized to belong in GDAL (r12899), and added a test for nodata and average (r12897).

Closing as fixed.

Thanks Even!