Usability of Digital Cameras for Verifying Physically Based ...

Abstract
Within computer graphics, the field of physically based rendering is
concerned with those methods of image synthesis which yield results that
do not only look real, but are also radiometrically correct renditions of
nature, i.e. which are accurate predictions of what a real scene would look
like under given lighting conditions.
In order to guarantee the correctness of the results, three stages of such a
rendering system have to be verified with particular care: the light reflection
models, the light transport simulation, and the perceptually based calcula-
tions used at display time.
In this thesis, the focus lies on the second step in this chain. Various
approaches for verifying the implementation of a physically based rendering
system have been proposed so far. However, the problem of proving that the
results are correct is not fully solved yet, and no standardized methodology
is available.
Using a photograph for verifying the results of a rendering system seems
obvious but an image produced by a common digital camera cannot be used
for this purpose directly. The sensor of a digital camera usually sees colors
differently than a human observer. Several techniques have been developed
to compensate for this problem. Our goal was to find and compare as many
meaningful ways of using a digital camera for verifying a physically based
rendering system as possible, in order to provide a practicable method for
any development environment. Some of the analyzed methods were taken
from the field of color management. Another method, that is based on a
novel approach, was developed throughout this thesis. We did an exhaustive
comparison of the usability and practicability of all the methods, focusing
on required equipment and cost. We found that in general more elaborate
methods give better results than low-end methods.
As some of the methods are based on XYZ color space, we considered
using this space as internal color space of our rendering system, rather than
doing full spectral rendering. However, we found a severe problem in using
XYZ space to determine the result of interactions of light and matter, as
XYZ space is not closed to component-wise multiplication of XYZ triplets.
Thus, based on this analysis we recommend full spectral rendering.
This thesis also contains a comprehensive overview of related work in
the field of verification of physically based rendering systems.
i