Separable Subsurface Scattering: Expanded Technical Report

In this paper we propose two real-time models for simulating subsurface scattering for a large variety of translucent materials, which need under 0.4 milliseconds per frame to execute (for typical portrait shots). This makes them a practical option for real-time production scenarios. Current state-of-the-art, real-time approaches simulate subsurface light transport by approximating the radially symmetric non-separable diffusion kernel with a sum of separable Gaussians, which requires multiple (up to twelve) 1D convolutions. In this work we relax the requirement of radial symmetry to approximate a 2D diffuse reflectance profile by a single separable kernel. We first show that low-rank approximations based on matrix factorization outperform previous approaches, but they still need several passes to get good results. To solve this, we present two different separable models: a) kernel preintegration, which yields a high-quality diffusion simulation; and b) an artist friendly model, which offers an attractive trade-off between physical accuracy and artistic control. Both reduce rendering subsurface scattering to just two 1D convolutions, while delivering results comparable to techniques with higher cost. Using importance-sampling and our stochastic sampling strategies, that rotate part of the kernel, only seven samples per pixel are required, even for high resolution imaging or close ups where our separable kernels might yield artifacts. We additionally show that separable profiles built with only two 1D Gaussians are able to perceptually match Monte 2 Jimenez et al. / Separable Subsurface Scattering: Expanded Technical Report 0 5 10 15 20 10−4 10−2 100 Singular values M ag ni tu de Channels