Discovering New Monte Carlo Noise Filters with Genetic Programming

Abstract

This paper presents a novel method for the discovery of new analytical filters suitable for filtering of noise in Monte Carlo rendering. Our method utilizes genetic programming to evolve the set of analytical filtering expressions with the goal to minimize image error in training scenes. We show that genetic programming is capable of learning new filtering expressions with quality comparable to state of the art noise filters in Monte Carlo rendering. Additionally, the analytical nature of the resulting expressions enables the run-times one order of magnitude faster than compared state of the art methods. Finally, we present a new analytical filter discovered by our method which is suitable for filtering of Monte Carlo noise in diffuse scenes.