Andries, B. (2017). Efficient wavelet-based 3D graphics rendering. [PhD Thesis, Vrije Universiteit Brussel].

@phdthesis{59cfbb195b1742b4a17809a22a54840d,
title = "Efficient wavelet-based 3D graphics rendering",
abstract = "The visual quality of interactive 3D rendering has taken huge leaps during the last few decades. Starting from simple wireframes, we currently witness immensely detailed 3D graphics scenes and interactive 3D worlds, requiring huge amounts of data in the rendering process. These datasets comprise geometry, 2D textures, effects, animations and many more. In this context, providing efficient texture coding methods featuring various forms of scalability is of paramount importance. More specifically, quality and resolution scalability are important functionalities to achieve high frame-rates in modern interactive 3D engines. This work explores novel methods for scalable compression and streaming of 2D texture data. A first research line focuses on the use of the wavelet transform as a tool in texture compression. 2D texture data represents one of the main data sources in 3D graphics, requiring large amounts of memory and bandwidth. Texture compression is of critical importance in this context to cope with these bottlenecks. To improve upon the available supported texture compression systems, several transform-based solutions have been proposed in the literature. These solutions, however, are not suitable for real-time texture sampling or provide insufficient image quality at medium to low rates. This work proposes a new scalable texture codec based on the 2D wavelet transform suitable for real-time rendering and filtering, using a new subband coding technique. The codec offers superior compression performance compared to the state-of-the-art, saving between 10 % and 30 % on rate. Additionally, our codec offers resolution scalability coupled with a wide variety of quality versus rate trade-offs as well as complexity scalability supported by the use of different wavelet filters. A second research line utilizes our proposed wavelet-based texture compression system as a means to optimize texture streaming or loading times in interactive 3D applications. When employing 2D texture data, instead of preloading all the potentially required information, texture data can be streamed on demand at run-time. This is a very common scenario in web-based applications, games and large virtual environments. This work proposes a new texture streaming system, utilizing scene analysis, camera prediction and wavelet-based texture compression to provide maximal visual quality within bandwidth and real-time constraints. Wavelet-based texture streaming provides neat advantages compared to conventional techniques when rendering 3D scenery requiring different resolution levels (e.g. terrains at oblique camera angles). In this case the streaming of certain high-pass subbands can be prioritized, utilizing the available bandwidth much more effectively. The proposed solution can be easily plugged into existing texture mapping solutions, as it features drop-in replacement shaders and re-uses existing render facilities. Our work shows that the wavelet transform is a flexible tool in interactive 3D rendering, showing lots of potential thanks to its inherent resolution scalability, its critically sampled 2D image representation, and the wide variety of available filters.",
keywords = "3D Graphics",
author = "Bob Andries",
year = "2017",
language = "English",
school = "Vrije Universiteit Brussel",
}