The simplest would be a 2D ratio of the area covered by the purported image to a circle circumscribed about a central point determined by the midpoint of the long axis of the image in the cloud. This ratio would be compared to a like ratio determined for the image of each (in)famous persona. The closest match determining the Madonna-esness or Jefferson-esness of the cloud image. (you actually are stating the problem as it relates to the proposed automatic airport identity software. The problem is definitely solvable)
That might be an interesting application. Maybe some researchers will try it in the near future.
It would be interesting to see this software applied to this sort of research. A decade ago, protein folding was a hot research topic, and they were trying to work out approaches to it. I haven't followed the research at all.