Cornell Box [LSG94]
Cornell University Program of Computer Graphics
Cornell Seal

Bounds and error estimates for radiosity.

Dani Lischinski, Brian Smits, and Donald P. Greenberg.

In Andrew Glassner, editor, SIGGRAPH 94 Conference Proceedings, Computer Graphics Proceedings, Annual Conference Series, pages 67--74. ACM SIGGRAPH, ACM Press, July 1994.

We present a method for determining a posteriori bounds and es­ timates for local and total errors in radiosity solutions. The ability to obtain bounds and estimates for the total error is crucial for reliably judging the acceptability of a solution. Realistic estimates of the local error improve the efficiency of adaptive radiosity algorithms, such as hierarchical radiosity, by indicating where adaptive refinement is necessary. First, we describe a hierarchical radiosity algorithm that com putes conservative lower and upper bounds on the exact radiosity function, as well as on the approximate solution. These bounds account for the propagation of errors due to interreflections, and provide a conservative upper bound on the error. We also describe a non­conservative version of the same algorithm that is capable of computing tighter bounds, from which more realistic error estimates can be obtained. Finally, we derive an expression for the effect of a particular interaction on the total error. This yields a new error­driven refinement strategy for hierarchical radiosity, which is shown to be superior to brightness­weighted refinement.

This paper is available as a compressed Postscript file ( 44K).

Last updated 08/23/02 PCG www Home