Sparse sampling for image-based SVBRDF acquisition

We acquire the data-driven spatially-varying (SV)BRDF of a flat sample from only a small number of images (typically 20). We generalize the homogenous BRDF acquisition work of Nielsen et al., who derived an optimal minmal set of lighting/view directions, treating a 4 degree-of-freedom spherical gantry as a gonioreflectometer. In contrast, we benefit from using the full 2D camera image from the gantry to enable SVBRDF acquisition. Like Nielsen et al, our method is data-driven, based on the MERL database of isotropic BRDFs, and finds the optimal directions by minimizing the condition number of the acquisition matrix. We extend their approach to SVBRDFs by modifying the optimal incident/outgoing directions to avoid grazing angles that reduce resolution and make alignment of different views difficult. Another key practical issue is aligning multiple viewpoints, and correcting for near-field effects. We demonstrate our method on SVBRDF measurements of new flat materials, showing that full data-driven SVBRDF acquisition is now possible from a sparse set of only about 20 light-view pairs.