Scanaerial is a python-based plugin for JOSM, that scans aerial imagery to detect areas (so far it's limited to detect lakes)
http://wiki.openstreetmap.org/wiki/Scanaerial

Scanaerial is an amazing advantage for armchair mapping. At the moment the detection leaves much to be desired. That was a reason for me to make some thoughts about aerial image detection.

I started collecting some material and information, if you're interested in algorithms and segmentation in general, have a look at:
https://github.com/jonasstein/scanaerial/wiki/Possible-image-processing-to-capture-objects-%28segmentation%29-on-aerial-images (you're invited to add some more information, examples or sources if you want)

But because I'm busy in these days and my time is limited, I started this blog-entry to encourage some more people to participate in the Scanaerial-project.

Help Scanaerial to get more efficient. We are in need of theoretician that can develop a good algorithm or enhance an existing algorithm and coders that can implement algorithms or add ideas to the python-code (or in a further state create a full Java-based plugin). Even if you don't want to work actively on that project, you could help by improving documentation/wiki or give new impetus with feedback & criticism.

Thanks for your attention!