Graphics Reference
In-Depth Information
Figure 8.78 Comparison of images rendered using the recovered parameters (top row) to the corre-
sponding captured images (bottom row). The rendered images have a darker background.
(From [Zinke et al. 09] c
2009 ACM, Inc. Included here by permission.)
search algorithm can converge to a local minimum rather than the global mini-
mum of the best fit. The authors use an approach known as simulated annealing ,
which is a probabilistic model for global optimization.
Figure 8.78 compares captured photographs, from which the BRDF parame-
ters are fit, and the results of rendering a comparable geometric hair model using
the recovered parameters. Although the geometric model is not very precise (so
the shifting in specular highlights on different fibers does not match the pho-
tographs) the hair color is faithfully reproduced. Figure 8.79 contains a simi-
lar comparison, except that the view direction is different from that used in the
image capture. Figure 8.80 contains renderings under different lighting environ-
ments. Figure 8.81 shows some results of rendering the reconstructed geometric
model described in the 2009 paper by Jakob, Moon, and Marschner. The render-
ing was done using Monte Carlo path tracing, the most accurate rendering method
available.
The method of fitting the BRDF in the paper by Zinke et al. fits both the
single and multiple scattering components in a single pass. Another possible
approach, which might provide more insight into this split, might be to use the
high frequency lighting separation method described by Nayar et al. in the 2006
Figure 8.79 Comparison of images rendered using the recovered parameters (top row) to photographs
(bottom row). In this case, the photographs have different view directions than those used
for the parameter fitting. The rendered results still match quite well. ( c
Arno Zinke, GfaR
mbH.)
Search WWH ::




Custom Search