Hum, I read a bit http://en.wikipedia.org/wiki/K-means_clustering and not
being familiar with it, it is not obvious how that could lead to the binary
result I expect ;-) I was hoping for a ready-made solution for my problem. I'm
kind of demanding tonight ;-)
Well, too late for tonight.
I guessed you would want a ready-to-use solution but what I can offer is
only my Mirone (and actually not mine originally) solution but I know
that's not what you need. But still, if you want to make a Windows
detour an play with the "Image -> k-means classification" you can see
it action and it's quite instructive.
Apparently the strategy I had in mind is a bit different from
Dimitriy's. My idea, driven by how it works in Mirone, is that you can
select in advance the number of clusters, and get them. So you can
analyze later their degree of uniformity. A map is expected to have few
well distinct classes so if you ask for, let's say 5 or 6, each one of
the will be rather coherent, with a low cluster variance (a measure of
the spreading about the cluster center). An image, on the other hand is
expected to have a high variance if one only allow a few clusters (the 5
or 6 above).
Maybe this code can be of use for you. I never tested it but it's a C
code with Matlab wrap functions to interface with it (a MEX), and the
guy is good programmer.
http://www.mathworks.com/matlabcentral/fileexchange/33541-fast-k-means-clustering
Joaquim
_______________________________________________
gdal-dev mailing list
[email protected]
http://lists.osgeo.org/mailman/listinfo/gdal-dev
