I have a schoolproject where the task is to locate all possible shapes (of a certain complex polygonic structure) in a layer of several heterogenous polygons.

Cadastral polygon

And now the task is to find all the ones that assimilate (more or less)

Highlighted examples

Detailed example

Starting from above the exact tools, I think we are dealing with a machine-learning concept here. I've looked at Grass-GIS GeoMorphon and it seems that it has the potential to start using as a tool. However I have never tried such an advanced task before. I have also considered OpenCV to maybe provide me with a solution, however the libraries for pattern recognition is based on raster photos and perhaps not suited for a vector-scaled dataset.

So far I've been filtering out all polygons <= 6 vertices after having done a (dangerous) simplification of the polygons. That reduced the number of polygons left to about 60 %. But there is still a long way to go and I have no idea of how to progress.

  • What exactly is the task? I think you left a crucial word out of your question. It says, "the task is to _______ all possible shapes." Please fill in the blank.
    – csk
    Commented Apr 3, 2018 at 18:27
  • 2
    Also, what do you mean by "assimilate"? This looks like parcel data. Are you trying to find properties where one property has only a narrow strip for road access?
    – csk
    Commented Apr 3, 2018 at 18:28
  • 3
    You could compare the ratio of circumference to area. Polygons with a long narrow strip extending off of the main body will have a higher ratio than polygons that are more compact.
    – csk
    Commented Apr 3, 2018 at 18:31
  • @csk interesting proposal. I'd look into that and see if there is a pattern.
    – MichaelR
    Commented Apr 3, 2018 at 19:59
  • See also grass.osgeo.org/grass74/manuals/v.to.db.html which offers vector line statistics.
    – markusN
    Commented Apr 13, 2018 at 9:25


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.