Not one of my ideas, but our designers say you can always use something 'near' black for painting black color in your graphics (something like (R,g,b)=(0,0,1)). This allows them to then evaluate the flattened image to see where the 'transparency' (0,0,0) is versus the real black (0,0,1).

I believe you can also put to use an image map (a black & white image that identifies the 'hot' areas in the graphic. Increases the amount of image data you need to keep, but allows you more flexibility (i.e. you can detect which portion of the monk has been hit for 'Soldier of Fortune' style hit detection).
-- termix