GPU-Based soft shadow rendering using non-linear pre-computed radiance transfer approximation

  • Authors:
  • Lili Wang;Jing Wang;Qinping Zhao

  • Affiliations:
  • School of Computer Science and Engineering, Bei Hang University, Beijing, P.R. China;School of Computer Science and Engineering, Bei Hang University, Beijing, P.R. China;School of Computer Science and Engineering, Bei Hang University, Beijing, P.R. China

  • Venue:
  • VSMM'06 Proceedings of the 12th international conference on Interactive Technologies and Sociotechnical Systems
  • Year:
  • 2006

Quantified Score

Hi-index 0.01

Visualization

Abstract

Soft shadow computation plays an important role in generating photo-realistic images. A new method of real-time all-frequency shadow rendering in 3D scenes under dynamic illumination is introduced in this paper. The non-linear wavelet approximation on the pre-computed radiance transfer matrix is used, which can help us to select the main directions sensitive to illumination of each sampling point. This approach efficiently avoids non-linear selection of the light vector in real-time rendering, and the approximation is used only once in the pre-computed phrase. We encode the resulting radiance transfer matrix into a sparse form, and store it as textures in 2D texture buffers. Meanwhile, the light vector is organized into a cube map on GPU to accelerate computation in the run-time process. This approach has been applied to render a variety of 3D scenes with soft shadows, and proved to be able to increase the relighting speed with no image quality reduction.