Memory Conserving Rendering Method for Hair/fur Systems in Computer Graphics

Abstract

In a very CPU/memory intensive field of photo-realistic computer graphics, various techniques are employed in the attempt to conserve resources. One group of such optimization methods is dedicated to optimizing a representation of hair systems, grass systems or any group of objects that can be looked at as a generalized hair system. A classical method of computing hair systems is to represent each hair as a spline in memory and then compute intersections with each of them. This method gives good results, but usually consumes large amounts of memory. Another problem is - visually doubling the density of hair quadruples memory consumption. Even when gigabytes of memory are available, a realistic hair scene, may overwhelm memory size, which may lead to an application crash or at least, to an I/O bottleneck and to increasing time of rendering. Another method is to compute a hair system procedurally inside of a specified volume. This produces a small memory foot-print, but makes animation difficult because individual hairs within the volume are not controllable. In this work we propose a hybrid approach, where a single hair particle represents a cylindrical volume, in which multiple hair fibers will be computed on-the-fly. This approach will produce a constant memory footprint for that cylindrical volume, regardless of how many individual hairs are computed and it will allow individual hairs to retain the behavior of that volume. As a result, a proposed approach provides a significant reduction in memory footprint while increasing the number of hairs being computed.