Graphics Reference
In-Depth Information
10
Radiance Transfer
Much of the history of computer graphics has been devoted to photorealistic ren-
dering, the goal of which is to simulate the real-world interaction and transfer of
light. As evidenced by the work described in the earlier chapters, much progress
has been made toward the completion of this goal. The rendering equation intro-
duced in Chapter 2 is a fundamental expression governing global illumination, a
key component of photorealistic rendering; however, the lack of volumetric ef-
fects, including participating media and subsurface scattering, is an intrinsic lim-
itation. Chapters 3 and 4 described methods for rendering these effects. Various
optimizations have improved accuracy and efficiency, but accurate interactive GI
computation remains difficult.
The emergence of programmable graphics hardware brought the possibility
of GI rendering entirely in graphics hardware. The existence of programmable
GPUs has been a driving force in research into improvements in image-based
approaches. As explained in Part II, image-based lighting (rendering from envi-
ronment maps) in its basic form reduces rendering to texture lookup and simple
linear operations. This is well suited to hardware rendering, as GPUs are op-
timized for just these operations. However, a drawback to this method is the
inability to handle local shadowing and indirect lighting. By computing these ef-
fects in a preprocess simulation, then combining the results on the GPU at render
time, a final rendering can be generated very quickly. This is the basic idea of
the
precomputed radiance transfer
(PRT) method originally introduced in 2002
by Peter-Pike Sloan, Jan Kautz, and John Snyder [Sloan et al. 02]. In the paper
entitled “Precomputed Radiance Transfer for Real-Time Rendering in Dynamic,
Low-Frequency Lighting Environments,” these authors described a technique that
can process GI almost entirely on a GPU.
The representation of precomputed radiance transfer simulation is a major
issue in PRT because the simulation results must be passed to the GPU in an ap-
339
