I’ve found two methods for breaking camouflage (from this paper)

1.  Multiple Camouflage Breaking by C0-Occurrence and Canny

2. Convexity-based Visual Camouflage Breaking

So, I’m starting to hope we really don’t need this…

CloudSight doesn’t even work given the images that the Canny paper (1) can deal with.  So, CloudSight must not be too cutting edge because I’d think that they would have that. 😉