I'm frustrated with v.generalize trying to simplify polygons. Those polygons stem from ArcGIS's raster to vector function (imported into GRASS from Shapefile), so they are pretty zigzag like this:
Trying to simplify them with
v.generalize input=polygons -c method=douglas threshold=0.1 layer=-1 output=polygons_dp
Leads to a weird result like this:
Some isles and even partially borders remain untouched. Tried v.clean, v.build.polylines before simplifying, but same result.
I finally got a workaround for the problem, based on Micha's suggestion (thank you!) to import the raster to GRASS and vectorize it there.
1 Import Raster:
r.in.gdal input=raster.tif output=raster
2 Vectorize (option -s leads to slightly smoothed 45-degree edges):
r.to.vect -s input=raster output=vector_blue feature=area
3 Generalize with Douglas to get rid of excessive points:
v.generalize input=vector_blue -c method=douglas threshold=0.05 output=vector_green
4 Smooth with Chaiken's algorithm to get smoother curves:
v.generalize input=vector_green -c method=chaiken threshold=0.1 output=vector_orange
5 Remove small polygons (area in m² if you have a lat/lon region):
v.clean tool=rmarea thresh=1000000000 input=vector_orange output=vector_red
If you can go back to the original raster, and import that into GRASS, then the module r.to.vect has a "-s" flag for smoothing the output vector boundaries. No generalize needed. Check this StackExchange question for a way to convert raster to vector in Arc and get smoother boundaries.
Back to this topic: you really wanted to apply "smoothing" here, see http://grass.osgeo.org/grass70/manuals/v.generalize.html#smoothing and not what you have tried (douglas). Use instead e.g. "chaiken":
http://grass.osgeo.org/grass70/manuals/v.generalize.html#smoothing-example
BTW: we added the possibility to generate an "error" map which contains those lines which cannot be simplified due to topological conflicts. Like this you can now easily identify issues in order to optimize the control parameters for simplification, smoothing, or displacement.
If you need to preserve the topology you can also have a look to Mapshaper. This is a javascript command line tool, but also a web interface and can be accessed in R through the rmapshaper package.
Image from here: 5% simplification; left: Visvalingam simplification (default in mapshaper), right: Douglas-Peucker simplification. Works extremely fast and is as easy as this line of code:
mapshaper input.shp -simplify 10% -o out.shp
