Visual Comput (2006) 22: 541–549 DOI 10.1007/s00371-006-0029-z ORIGINAL ARTICLE Richard Sharp Raghu Machiraju Accelerating subsurface scattering using Cholesky factorization Published online: 10 June 2006  Springer-Verlag 2006 R. Sharp (✉) · R. Machiraju 395 Dreese Laboratories 2015 Neil Avenue Columbus, OH 43210 {sharpr,raghu}@cse.ohio-state.edu Abstract In this paper we present a simplified subsurface scattering model that exploits a diffusion mechanism to provide a simpler solution to the transport equation. Our model is based on numerical analysis techniques that are amenable to Cholesky factorization. We treat the factorization as a precomputed scattering quantity which can be used to significantly speed up mul- tiple scattering calculations as the global light source changes. On low resolution meshes, we have been able to achieve real-time solutions of the subsurface scattering while still maintaining good visual quality of the solution. Keywords Subsurface Scattering · Picture/Image Generation · Three- Dimensional Graphics and Realism 1 Introduction Recently practical computational subsurface scattering models [12] have been introduced in computer graphics literature. This has been in part a drive to increase the quality of realistic image synthesis, although there has has been interest from the biomedical community to simu- late and predict the behavior of light scattering in human tissue. Previous research in this area has yielded various models of light transport, each suited for modeling scat- tering and substrate transport in different types of materi- als [1, 2, 6, 8, 12, 20, 28]. Realistic looking images have been obtained through these models. Unfortunately, many of them have limita- tions such as restrictions to simple geometries [1, 2, 20], homogeneous materials [2, 12, 20] or are limited to one di- mensional transport theory or BRDFs [1, 2, 6, 8, 18, 24]. Our transport model, which incorporates material modeling and addresses the issues raised above, has been previously presented in [26]. In this version, we present enhancements which drastically reduce the computation time for the underlying scattering calculations. These include a method to factor the scattering matrix and an im- plementation of spatial subdivision techniques to update source values due to changes in lighting in near real-time. The model presented in this paper assumes that the mate- rial properties can be modeled by a diffusion process (the scattering coefficient is much larger than the absorption coefficient) and, to ensure that our precomputation tech- niques can be used, we assume that the geometry and ma- terial scattering properties are fixed during the simulation. In the computer graphics literature diffusion has been used to model subsurface scattering in media where diffu- sion dominates [12, 28]. Although we also use diffusion, we employ a simpler scattering framework which can can handle inhomogeneous materials. In essence we em- ploy a restricted model of transport where flux propagates along a 3D grid of cells. Although this may seem restric- tive, it should be noted that in the limiting case our cell transport diffusion model will model the complete diffu- sion process and hence subsurface scattering in its entirety. 1.1 Paper overview The rest of our paper is as follows. First in Sect. 2 we dis- cuss previous works including different subsurface light