Your error diffusion dither is strange to me. Why do you not use two-dimensional patterns like Floyd-Steinberg or Jarvis-Judice-Ninke? Is your assumption that

Your error diffusion dither is strange to me. Why do you not use two-dimensional patterns like Floyd-Steinberg or Jarvis-Judice-Ninke? Is your assumption that you're diffusing error linearly in Morton or Hilbert curves? Because I'm pretty sure that's going to be outperformed in quality by any common 2D error diffusion pattern.