Very interesting. I am smelling Fourier optics. If you treat light propagation as Fourier transform, then with an assumption you have point-like incident light, and image (that photo) at a fixed distance, you can derive the intensity distribution needed on the glass and mill accordingly. I don't know about roughness of milling but this is very cool. I'll show this to students taking Fourier optics if I get to teach it next time.