Noise Reduction on G-Buffers for Monte Carlo Filtering

Bochang Moon12, Jose A. Iglesias-Guitian1, Steven McDonagh13, Kenny Mitchell14

Disney Research1, Gwangju Institute of Science and Technology2, Imperial College London3, Edinburgh Napier University4

Computer Graphics Forum (Presented at Eurographics Symposium on Rendering 2017)

Noise Reduction on G-Buffers for Monte Carlo Filtering
Our results for the Cars scene where each car has a different motion. The input image (a), and its close-up (b), are generated using 8 samples per pixel (spp) allocated by an adaptive sampler using weighted local regression (WLR) [MCY14]. Recent filtering methods utilize geometric buffers (G-buffers) such as texture (c), normal (e), and depth, which may contain severe noise in regions with strong motion blur effects. As a result, the state-of-the-art method (g) produces over- and underblurred results in those regions. Our method applies an anisotropic pre-filtering to the noisy feature buffers and generates the pre-filtered G-buffers (d) and (f). The recent filter (h) that utilizes our results instead of the noisy G-buffers, shows a reduced error, i.e., the relative mean squared error (rMSE) [RKZ11], and better preserved edges thanks to our high-quality pre-filtering.

Abstract

We propose a novel pre-filtering method that reduces the noise introduced by depth-of-field and motion blur effects in geometric buffers (G-buffers) such as texture, normal and depth images. Our pre-filtering uses world positions and their variances to effectively remove high-frequency noise while carefully preserving high-frequency edges in the G-buffers. We design a new anisotropic filter based on a per-pixel covariance matrix of world position samples. A general error estimator, Stein’s unbiased risk estimator, is then applied to estimate the optimal trade-off between the bias and variance of pre-filtered results. We have demonstrated that our pre-filtering improves the results of existing filtering methods numerically and visually for challenging scenes where depth-of-field and motion blurring introduce a significant amount of noise in the G-buffers.

Contents