Adaptive Matrix Column Sampling and Completion for Rendering Participating Media

摘要

Several scalable many-light rendering methods have been proposedrnrecently for the efficient computation of global illumination. However,rngathering contributions of virtual lights in participating mediarnremains an inefficient and time-consuming task. In this paper, wernpresent a novel sparse sampling and reconstruction method to acceleraternthe gathering step of the many-light rendering for participatingrnmedia. Our technique explores the observation that the scatteredrnlightings are usually locally coherent and of low rank evenrnin heterogeneous media. In particular, we first introduce a matrixrnformation with light segments as columns and eye ray segments asrnrows, and formulate the gathering step into a matrix sampling andrnreconstruction problem. We then propose an adaptive matrix columnrnsampling and completion algorithm to efficiently reconstructrnthe matrix by only sampling a small number of elements. Experimentalrnresults show that our approach greatly improves the performance,rnand obtains up to one order of magnitude speedup comparedrnwith other state-of-the-art methods of many-light rendering for participatingrnmedia.