LinSSS: linear decomposition of heterogeneous subsurface scattering for real-time screen-space rendering

Screen-space subsurface scattering is currently the most common approach to represent translucent materials in real-time rendering. However, most of the current approaches approximate the diffuse reflectance profile of translucent materials as a symmetric function, whereas the profile has an asymmetric shape in nature. To address this problem, we propose LinSSS, a numerical representation of heterogeneous subsurface scattering for real-time screen-space rendering. Although our representation is built upon a previous method, it makes two contributions. First, LinSSS formulates the diffuse reflectance profile as a linear combination of radially symmetric Gaussian functions. Nevertheless, it can also represent the spatial variation and the radial asymmetry of the profile. Second, since LinSSS is formulated using only the Gaussian functions, the convolution of the diffuse reflectance profile can be efficiently calculated in screen space. To further improve the efficiency, we deform the rendering equation obtained using LinSSS by factoring common convolution terms and approximate the convolution processes using a MIP map. Consequently, our method works as fast as the state-of-the-art method, while our method successfully represents the heterogeneity of scattering.