Blue noise sampling using an N-body simulation-based method

We present a physically based blue noise sampling approach which can be evaluated efficiently by using the N-body simulation method. A set of sample points is modeled as electrically charged particles on an imaginary 2D plane where they self-organize by movement to minimize the electrostatic force that they each experience. The resulting particles' positions at equilibrium exhibit an equidistant neighborhood characteristic that fulfills the essential requirement of a quality blue noise point set. We propose to use the Velocity Verlet algorithm commonly used in molecular dynamics simulation as our integration method, and we apply custom adaptation to improve the convergence rate for our purpose. Our method uses the magnitude of electrical charge of particles as an intuitive control parameter of the spectral behavior of the generated blue noise point sets. We are able to obtain high-quality blue noise point sets comparable to the state-of-the-art results, and we have also implemented a simple GPU application to evaluate our method on the image stippling application.