ISBN 9783843931687

978-3-8439-3168-7, Reihe Informatik

Kai Selgrad
Efficient Visibility Queries and Metalinguistic Abstractions for Graphics Algorithms

288 Seiten, Dissertation Universität Erlangen-Nürnberg (2016), Softcover, A5

Zusammenfassung / Abstract

The use of computer generated images from virtual scenes has become ubiquitous. We find convincingly rendered images in everyday applications ranging from movies and games to advertisement, but also in specialized applications such as architectural visualizations and physical simulations. The determination of visibility between sets of points in such virtual scenes is one of the cornerstones of efficient image synthesis. Resolving visibility can be phrased as answering the questions what is visible from a given point in a given direction as well as whether two points are mutually visible or not. These seemingly simple questions have generated a tremendous body of research over half a century and are still a relevant area of research as the ubiquitousness of their application demands ever more convincing results at higher resolutions while providing more advanced effects.

The first part of this book focuses on reducing the amount of time required to resolve visibility in certain specific applications, namely computing shadows as well as indirect lighting from physically based light sources and computing defocus blur by simulating lenses. It also describes how large data sets can be compressed in such a way that visibility can be resolved with faster, but more memory-restrictive hardware.

The second part of this book describes methods that help to achieve well performing implementations of such algorithms by employing meta-programming techniques, allowing a more thorough exploration of the space of possible solutions. We also present a simple and efficient way of implementing domain-specific abstractions, such as programming languages that are tailored to a single application-domain. This way we show how graphics research can benefit from higher-level programming paradigms while avoiding heavy-weight abstractions that cannot be maintained by individuals without adversely affecting the productivity they initially hope to gain.