[Imports] Criteria for matching objects for conflation
Josh Doe
josh at joshdoe.com
Wed Apr 25 21:24:16 UTC 2012
I'd like to get feedback on criteria typically used for conflating
(merging) objects, either within the OSM database or between the OSM
database and an "authoritative" database.
The JOSM conflation plugin I've been slowly working at for a while now
is starting to come together, and can be actually used, though in a
limited number of cases, such as merging house number nodes with
building outlines.
I'd now like to create a dialog for choosing the criteria used to
determine the score of a potential match between objects. The score
ranges from 0 (definitely not a match) to 1 (perfect match). The
current score is simply the Euclidean distance between objects.
However, as I use the Java Conflation Suite and the Java Topology
Suite, I have a large assortment of criteria at my disposal, such as
Hausdorff distance, polygon overlap, angle histogram, string
similarity, etc. All of these matchers produce a score, but they can
then be combine in various ways to give a composite score. Examples
include doing a weighted sum of scores, thresholding scores, etc.
I'm uncertain of how to allow the user to specify the criteria to be
used, as the combinations can be limitless. I was thinking of sticking
to a simple weighted sum, allowing users to add any matchers they
choose, assigning a weight and threshold to each one. This dialog will
already be quite complicated, so I'd like to see if there's a way I
could simplify things.
http://wiki.openstreetmap.org/wiki/JOSM/Plugins/Conflation
https://github.com/joshdoe/josm-conflation-plugin
Thanks for any feedback,
-Josh
More information about the Imports
mailing list