Blue-noise dithered sampling

Abstract

The visual fidelity of a Monte Carlo rendered image depends not only on the magnitude of the pixel estimation error but also on its distribution over the image. To this end, state-of-the-art methods use high-quality stratified sampling patterns, which are randomly scrambled or shifted to decorrelate the individual pixel estimates. While random pixel decorrelation yields an eye-pleasing whitenoise image error distribution, it is far from perceptually optimal. We show that visual fidelity can be substantially improved by instead correlating the samples among pixels in a way that minimizes the low-frequency content in the output noise. Inspired by digital halftoning, our blue-noise dithered sampling method can produce significantly more faithful images, especially at low sampling rates.