Abstract
Synthesis of realistic images which predict the appearance
of the real world has many applications including
architecture and interior design, illumination engineering,
environmental assessment, special effects and film production,
along with many others. Due to costly global illumination computation,
which is required for the prediction of appearance,
physically-based
rendering still remains the domain of research laboratories,
and is rarely used in industrial practice.
The main goal of this work is to analyze problems
and provide solutions towards making predictive rendering
an efficient and practical tool.
First, existing global illumination techniques are discussed,
then efficient solutions which handle complex geometry,
multiple light sources, and arbitrary light scattering
characteristics are proposed.
Since real-time lighting computation
is not affordable for complex environments,
techniques of lighting storage and real-time
reconstruction using pre-calculated results are developed.
Special attention is paid to the solutions
which use perception-guided algorithms to improve their performance.
This makes it possible to focus the computation on readily visible
scene details, and to stop it when further improvement
of the image quality cannot be perceived by the human
observer. Also, by better use of
perception-motivated physically-based partial solutions,
meaningful images can be presented to the user
at the early stages of computation.
Since many algorithms make simplifying assumptions about
the underlying physical model in order to achieve gains
in rendering performance, a validation procedure
for testing lighting simulation accuracy and image
quality is proposed. To check the requirement of
appearance predictability imposed on the developed algorithms,
the rendered images are compared against the corresponding real-world
views.
