Research On Pre-computation And Sampling Based Real-time Realistic Image Rendering Technology

Real-time realistic image rendering is one of the most important and difficult problems in Computer Graphics, previous methods cannot provide highly realistic image in real time. This thesis focuses on pre-computation and sampling based real-time realistic image rendering. The key idea is to use the information in pre-computation results or pictures captured, for presenting effects of highly realistic shading and supporting fast rendering. In this thesis, we propose several pre-computing and sampling strategies for walkthrough system rendering, natural surface rendering with real-world material and 3D texture, and rendering with global illumination effects respectively. We also introduce efficient rendering methods, which work with these pre-computing and sampling strategies, and perform real-time realistic image rendering successfully.In research for walkthrough system rendering, we focus on image-based rendering (IBR) technique. The main problem of IBR is the organization method for large data set, and the cooperation strategy with geometry model. We propose a new 3D plenoptic function method - Boundary Light Field, which works with rough geometry model of scene objects. This method arranges sampling data set on the rough geometry model, then takes adaptive resampling to decrease the data amounts. Furthermore it could correct large distortion in depth, and make use of arbitrary movement bound for viewer. This method can be widely applied in walkthrough systems.For rendering of real-world surface material and 3D texture surface, the Bi-directional Reflection Distribution Function (BRDF) and Bi-directional Texture Function (BTF) are described. Sampling data for BRDF is analyzed with Spherical Harmonics, and then the BRDF surface could be rendered fast under environment illumination. For the 6D BTF, previous methods could hardly process it efficiently because of the large amount of sampling data. In this thesis, we propose a new Spherical.Harmonics based 3D texture model - Illumination-dependent Texture (IDT).This texture model concentrates on presenting the 3D surface detail effects due to light movement, and ignores the effects caused by variation of viewing direction. This method reduces the data amount effectively while keeping the major 3D texture appearance. We also introduce synthesis technique for IDT to transport the real-world texture samples on virtual objects. Finally, the texture-covered objects could be rendered under environment illumination in real time.For real-time global illumination rendering, the pre-computing techniques for radiance transfer function are described in this paper. These methods pre-compute the information needed for global illumination rendering, and access the pre-computed results in real-time rendering. The advantage is to transfer the major computation load of traditional rendering methods into preprocess. The main components of pre-computed radiance transfer technique are to explore suitable pre-computation and sampling strategy, to reduce the amount of sampling data, and to support fast rendering. Previous methods are limited to render single object or static scene only. In this thesis, we propose a new technique - Spherical Radiance Transport Maps, for rendering dynamic scene under environment illumination. This method could render soft shadow and inter-reflection for multiple dynamic objects efficiently.