Effective Multi-resolution Rendering and Texture Compression for Captured Volumetric Trees

Abstract

Trees can be realistically rendered in synthetic environments by creating volumetric representations from photographs. However, volumetric tree representations created with previous methods are expensive to render due to the high number of primitives, and have very high texture memory requirements. We address both shortcomings by presenting an efficient multi-resolution rendering method and an effective texture compression solution. Our method uses an octree with appropriate textures at intermediate hierarchy levels and applies an effective pruning strategy. For texture compression, we adapt a vector quantization approach in a perceptually accurate color space, and modify the codebook generation of the Generalized Lloyd Algorithm to further improve texture quality. In combination with several hardware acceleration techniques, our approach achieves a reduction in texture memory requirements by one order of magnitude; in addition, it is now possible to render tens or even hundreds of captured trees at interactive rates.