[gdal-dev] gdalwarp makes landcover files blocky when shrinking pixel size

[John Twilley]

I believe my landcover data is a classified raster.  It has a color
table and uses values like 11 for water and 12 for ice.

The majority algorithm I implemented that looked okay in the old code
used a number of the nearest neighbors.  Taking just one nearest
neighbor made the blocky images I described in my original message.
If gdalwarp had gdaladdo's mode algorithm, would that do a better job
than the nearest-neighbor algorithm?

When I tried using pct2rgb.py on the VRT, then running the gdalwarp
with a variety of resampling methods, then using rgb2pct.py on the
resulting output (referencing the VRT for the palette), the output was
filled with speckles.  For example, when a finger of sand poked out
into the water, the various methods created many tiny one-pixel lakes
and islands.  Not quite right.

Jack.
--
mathuin at gmail dot com



On Fri, Jan 27, 2012 at 11:27, Chaitanya kumar CH
<chaitanya.ch at gmail.com> wrote:
> John,
>
> You are right. gdaladdo is not for creating higher resolution images. I
> thought you needed the opposite.
> I assume your landcover data is a classified raster. Classified data is
> usually zoomed in with nearest neighbor interpolation. Unless you fiddle
> with the interpolation window, the majority algorithm is pretty much the
> nearest neighbor while zooming in.
>
> If you want, you can convert the image into an RGB using pct2rgb.py and work
> with the RGB image.
>
>
> On Fri, Jan 27, 2012 at 11:30 PM, John Twilley <mathuin at gmail.com> wrote:
>>
>> I've never seen gdaladdo -- neat command!  Alas, the overviews are the
>> wrong way 'round -- I don't need to zoom out, I need to zoom in.
>> gdaladdo doesn't handle fractional scale values well, so I can't make
>> something bigger out of something smaller like I can with gdalwarp.
>>
>> Jack.
>> --
>> mathuin at gmail dot com
>>
>>
>>
>> On Thu, Jan 26, 2012 at 20:01, Chaitanya kumar CH
>> <chaitanya.ch at gmail.com> wrote:
>> > John,
>> >
>> > Try gdaladdo [1] with the resampling algo set to 'mode'. If you use the
>> > -ro
>> > option, it will create an external overview. Check your output with
>> > different combinations of levels.
>> >
>> > [1]: http://www.gdal.org/gdaladdo.html
>> >
>> > On Fri, Jan 27, 2012 at 5:09 AM, John Twilley <mathuin at gmail.com> wrote:
>> >>
>> >> I am working with elevation and landcover data downloaded from the
>> >> USGS.  I use gdalwarp to convert the data to a much smaller pixel. The
>> >> elevation data works very nicely with cubic resampling, but the only
>> >> resampling that works at all for the landcover data is
>> >> nearest-neighbor and that's very blocky.  When I last worked with
>> >> landcover data, I used a majority algorithm which produced smoother
>> >> output -- but that algorithm is not implemented in gdalwarp.
>> >> I am looking over the source to gdalwarp to see how hard it is to add
>> >> a new algorithm.  Other than that, though, what options are available
>> >> to me?  Thanks in advance!
>> >> Jack.--
>> >> mathuin at gmail dot com
>> >> _______________________________________________
>> >> gdal-dev mailing list
>> >> gdal-dev at lists.osgeo.org
>> >> http://lists.osgeo.org/mailman/listinfo/gdal-dev
>> >
>> >
>> >
>> >
>> > --
>> > Best regards,
>> > Chaitanya kumar CH.
>> >
>> > +91-9494447584
>> > 17.2416N 80.1426E
>
>
>
>
> --
> Best regards,
> Chaitanya kumar CH.
>
> +91-9494447584
> 17.2416N 80.1426E