Abstract
Density estimation employed in multi-pass global illumination algorithms gives
cause to a trade-off problem between bias and noise. The problem is seen most
evident as blurring of strong illumination features. This thesis addresses the
problem, presenting four methods that reduce both noise and bias in estimates.
Good results are obtained by the use of anisotropic filtering. Two methods handles
the most common cases; filtering illumination reflected from object surfaces.
One methods extends filtering to the temporal domain and one performs filtering
on illumination from participating media. The applicability of the algorithms is
demonstrated through a series of tests.
cause to a trade-off problem between bias and noise. The problem is seen most
evident as blurring of strong illumination features. This thesis addresses the
problem, presenting four methods that reduce both noise and bias in estimates.
Good results are obtained by the use of anisotropic filtering. Two methods handles
the most common cases; filtering illumination reflected from object surfaces.
One methods extends filtering to the temporal domain and one performs filtering
on illumination from participating media. The applicability of the algorithms is
demonstrated through a series of tests.
| Original language | English |
|---|
| Place of Publication | København |
|---|---|
| Number of pages | 138 |
| Publication status | Published - 2009 |