PointOnSurface strange behavior

[sbe]

Hi all,

Using Postgis's ST_PointOnSurface, i've got some weird results in recent version of Postgis (> 2.0).

See that image :

I would like to know how the PointOnSurface is calculated, and why there are so much differences between polygon A,B and C.

With Postgis 1.5 it was working fine (i.e. the PointOnSurface for my polygons were at almost the same place than the Centroids)

Thank you.

Sylvain b.

[mloskot]

WKT of your geometries would be helpful

[sbe]

Here are the WKT:

A:POLYGON((182111.031940953 141935.935903267,181944.315698016 141813.663222482,181993.426552077 141735.161360171,182025.215323227 141755.058841504,182042.126877935 141723.338145732,182089.035943744 141755.453352846,182096.699289843 141744.892410235,182142.107084301 141773.967727472,182140.268233846 141825.482203511,182123.233970367 141827.369156254,182111.031940953 141935.935903267))

B:POLYGON((182512.231897141 141935.935903267,182344.315698016 141813.663222482,182393.426552077 141735.161360171,182425.215323227 141755.058841504,182442.126877935 141723.338145732,182489.035943744 141755.453352846,182496.699289843 141744.892410235,182542.107084301 141773.967727472,182540.268233846 141825.482203511,182522.034014178 141825.369229273,182512.231897141 141935.935903267))

C:POLYGON((182635.760119718 141846.477712277,182826.153168283 141974.473039044,182834.952846998 141857.67730337,182862.151853936 141851.277537031,182860.551912351 141779.280165725,182824.553226698 141748.881275618,182814.953577191 141758.480925126,182766.155358861 141721.682268681,182742.156235092 141744.881421657,182692.558045971 141716.882443927,182635.760119718 141846.477712277))

Projection: srid= 31370

Thanks.

[mloskot]

FWIW, ST_PointOnSurface returns an arbitrary point. The only guarantee is that such point is located inside interior of a geometry. Subsequent calls of ST_PointOnSurface for the same geometry might possibly return different points (all inside the interior, of course).

[mloskot]

FYI, added the three similar polygons to the tests ​https://git.osgeo.org/gogs/geos/geos/commit/64e823b26bc6a771571f39af245b0e0f4e256df8

[mdavis]

The reason for the discrepancy in computed interior points is due to how the Interior Point in Area algorithm works. It is designed to quickly find a point lying in the polygon interior, and is optimized for performance, not quality of point location. The two polygons just happen to be different enough that the algorithm chooses a different "band" of the polygon to place the interior point.

If a better quality location is desired, then there are other (less performant) algorithms to do this (e.g. computing the LargestInscribedCircle AKA Pole of Inaccessibility). I'm hoping to bring this to GEOS soon. Or perhaps compute both the centroid and interior point and pick the best one (= furthest from polygon boundary AND in the interior)

@mloskot: the InteriorPointArea algorithm is deterministic, so it does actually return the same point each time it is called.

Finally, there's a big performance improvement available for the InteriorPointArea algorithm, and that should land soon in GEOS.