Noise reduction for path traced imaging of participating media

The paper proposes a new noise reduction technique counteracting noise in path traced images: a filter kernel, referred to as gradient kernel, is defined exploiting the gradient direction of particle density of participating media. The gradient kernel is combined with the bilateral filter, so to enhance it: the rationale is to perform Monte Carlo noise suppression while preserving details in path traced images, exploiting the information inherent in the 3D scenes. The proposed method is applied to path traced images, the rendering results of multiple and non-isotropic light scattering in non-homogeneous participating media. The experimental results show that the novel approach behaves remarkably well both quantitatively and qualitatively.