Efficient Estimation of Spatially Varying Subsurface Scattering Parameters

We present an image-based technique to efficiently acquire spatially varying subsurface reflectance properties of a human face. The estimated properties can be used directly to render faces with spatially varying scattering, or can be used to estimate a robust average across the face. We demonstrate our technique with renderings of peoples’ faces under novel, spatially-varying illumination and provide comparisons with current techniques. Our captured data consists of images of the face from a single viewpoint under two small sets of projected images. The first set, a sequence of phase-shifted periodic stripe patterns, provides a per-pixel profile of how light scatters from adjacent locations. The second set of structured light patterns is used to obtain face geometry. We subtract the minimum of each profile to remove the contribution of interreflected light from the rest of the face, and then match the observed reflectance profiles to scattering properties predicted by a scattering model using a lookup table. From these properties we can generate images of the subsurface reflectance of the face under any incident illumination, including local lighting. The rendered images exhibit realistic subsurface transport, including light bleeding across shadow edges. Our method works more than an order of magnitude faster than current techniques for capturing subsurface scattering information, and makes it possible for the first time to capture these properties over an entire face.