There's a lot of great work in Wikimedia commons, but I think there are 2 primary / very common issues:
* Format - many of these maps are .svg or illustrator-created images with no projection information, which makes data extraction very difficult. This is why we're trying to share with GIS-formatted data, so people can do with it what they'd like.
* Sourcing - very often, a map is depicted with a source of "own work". That, unfortunately, really isn't good enough & we're trying to source all objects in OHM. What's scary is that putting something in Wikimedia almost ensures wide propagation as "fact", even if it lacks a source. Here's an example... the base map is sourced, but where did this red line come from? It appears to be hand-drawn on a map... who drew it? How'd they get that outline? https://commons.wikimedia.org/wiki/File:Comancheria.jpg
Relatively accurate projection detection should be feasible for some, particularly vector, maps. Perhaps a library for this could be made. A simple approach would be convolution of nearby topographic features by assumed potential projection and then a least-difference comparison with observed positions. I am sure the CIA foreign map processing people have had this stuff for decades already.
