Master Thesis - Computer Graphics and Visualization - TU Delft

Chapter 1
Introduction
Since the introduction of computer graphics, synthesizing computer images has found many
useful applications. Of these, the best known are probably those in the entertainment industries,
where image synthesis is used to create visually compelling computer games and to
add special effects to movies. Besides these and many others, important applications are
found in advertisement, medicine, architecture and product design. In many of these applications,
image synthesis is used to give a realistic impression of the illumination in virtual
3D scenes. For example, in architecture, CG is used to answer questions about the illumination
in a designed building [15]. Such information is used to evaluate the design before
realizing the building. In advertising and film production, computer graphics is used to
combine virtual 3D models with video recordings to visualize scenes that would otherwise
be impossible or prohibitively expensive to capture on film. Both these applications require
a high level of realism, giving an accurate prediction of the appearance of the scene and
making the synthesized images indistinguishable from real photos of similar scenes in the
real world.
Physically based rendering algorithms use mathematical models of physical light transport
for image synthesis and are capable of accurately rendering such realistic images. For
most practical applications, physically based rendering algorithms require a lot of computational
power to produce an accurate result. Therefore, these algorithms are mostly used
for off-line rendering, often on large computer clusters. This lack of immediate feedback
complicates the work for the designer of virtual environments.
To get some immediate feedback, the designer usually resorts to approximate solutions
[6, 32, 36, 55]. Although interactive, these approximations sacrifice accuracy for performance.
Complex illumination effects such as indirect light and caustics are roughly approximated
or completely absent in the approximation. Figure 1.1 shows the difference between
an approximation 1 and a physically based rendering. Although the approximation looks
plausible, it is far from physically accurate. The lack of accuracy reduces its usefulness to
the designer. Designers would benefit greatly from interactive or near-interactive physically
based rendering algorithms on a single workstation for fast and accurate feedback on their
1 This approximation algorithm simulates a single light indirection and supports only perfect specular and
perfect diffuse materials.
1