[gdal-dev] VRT KernelFilteredSource and NoDataValue

[Even Rouault]

Selon Belaid MOA <belaid_moa at hotmail.com>:

>
> As always, thank you very ... very much Even. Just to make sure I understood
> your reply correctly,
> I do the following to set the source_0 and the nodatavalue (passed as
> argument)
> after creating the XML for the VRTKernelFilteredSource and set it in the
> papszMetadata array.
> The poBand is the RasterBand I get from the VRT dataset.
>
>       poBand->SetNoDataValue(noDataValue);
>       poBand->SetMetadata( papszMetadata, "vrt_sources" );
>
> Is this right place to set the NoDataValue of the VRTRasterBand?
>
> For the normalization effect (normalized=1), does it mean that:
> 1- The no data value pixels are ignored AND
> 2- The coefficients of valid pixels only are adjusted so that their sum  is
> equal to 1?

Yes, correct. Well, that's my understanding of the code : no warranty expressed
or implied ;-) I guess you just have now to try

> If this is the case, then it is exactly the behavior I want!
>
> With best regards.
> ~Belaid.
>
> > Date: Sun, 30 Aug 2009 10:12:04 +0200
> > From: even.rouault at mines-paris.org
> > To: belaid_moa at hotmail.com
> > CC: gdal-dev at lists.osgeo.org
> > Subject: Re: [gdal-dev] VRT KernelFilteredSource and NoDataValue
> >
> > Selon Belaid MOA <belaid_moa at hotmail.com>:
> >
> > Belaid,
> >
> > First, the VRTKernelFilteredSource will only recognize the nodata value if
> you
> > set it on the VRTRasterBand (it will not fetch it directly from the
> > <SourceFilename>)
> >
> > When a nodata value is set and found in one of the values in a NxN square,
> the
> > value will be ignored. The effect of this depends whether normalized="1" is
> set
> > on <Kernel> or not. If normalized=1, the sum of the coefficients is not
> > incremented for that nodata value, so the normalization will operate only
> on
> > valid pixel values. If normalized=0, the nodata value is just ignored
> >
> > As far as what happens in the borders is concerned, the code says that :
> "Fill
> > in missing areas.  Note that we replicate the edge valid values out.  We
> don't
> > using "mirroring" which might be more suitable for some times of filters.
> We
> > also don't mark these pixels as "nodata" though perhaps we should."
> >
> > Best regards,
> >
> > Even
> >
> > >
> > > Hi everyone,
> > >     I am using VRT KernelFilteredSource to implement an average filter as
> > > done in the VRT tutorial. However, I am still unclear on how it handles
> the
> > > NoDataValue pixels. Are they too averaged or does it leave them as they
> are?
> > > If they are masked,  does do an adjustment for the kernel coefficients to
> > > compensate for their absence? Another related question, is how does it
> handle
> > > the pixels on the borders and the kernel coefficients for those pixels?
> > >
> > >    Thank you very very much in advance for any help on these issues.
> > >
> > >    With best regards.
> > > ~Belaid.
> > >
> > > _________________________________________________________________
> > > Stay on top of things, check email from other accounts!
> > > http://go.microsoft.com/?linkid=9671355
> >
> >
>
> _________________________________________________________________
> Click less, chat more: Messenger on MSN.ca
> http://go.microsoft.com/?linkid=9677404