You're right, that's a very big if.

Also, most of the research in the field has been done on black and white image, so it not directly usefull.

And come to think on it, if we're talking about image stego, what prevents the NSA from controlling the statistical distributions of output of cameras in such a way that stego will be noticed(even after some image transforms) ?