A Study of Spectral Rendering Algorithms

• Main Authors: Chern Jin-Ren, 陳金仁
• Other Authors: Wang Chung-Ming
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/15202717366641891499

博士 === 國立中興大學 === 資訊科學研究所 === 93 === Images involving important optical effects, such as light dispersion, interference, diffraction, and fluorescence, must be synthesized with spectral rendering. However, the handling of data and SPDs (Spectral Power Distribution) for spectral rendering is really more difficult than the handling of RGB data, since SPDs include a vast number of wavelengths. In addition, spectral rendering has been notorious for its high-complexity computing. In this dissertation, we present two methods to solve the problems depicted above. Spectral surface reflectance curves are the surface reflectance constructed according to specific illumination models for spectral rendering. Nevertheless, it is a tedious and an inefficient job to construct spectral surface reflectance curves without any tools. In this dissertation, we first design and implement an interactive spectral rendering visualization tool (ISRVT), by which users can interactively construct spectral surface reflectance curves by dragging a mouse on the graphical user interface. Due to the time-consuming characteristic of spectral rendering and the term of the Phong illumination model, quick rendering to meet the requirement for interactive operation is a key challenge in designing the ISRVT. We propose an acceleration algorithm which utilizes both the importance-driven dividing method to find better positions of tabulating intervals and a parabolic interpolation method to improve the accuracy of linear interpolation to overcome the problem of slow response. The dissertation also presents a new progressive refinement algorithm for full spectral rendering. This algorithm adopts wavelet transformation to efficiently represent full spectral data. We implemented the proposed technique for Monte Carlo direct lighting, and divided the rendering process into nine stages (i=1 to 9), each of which employs the first leading 2i coefficients to produce progressive results. The progressive results at stage i can be joined together in the next stage (stage i+1). Clearly, we only need to calculate another 2i coefficients at stage i+1, since the same number of coefficients have been available at stage i. This fact makes it possible to reuse results of previous stages and reduce the rendering time of succeeding stages. The quality of the rendered image is visually plausible, being indistinguishable from those rendered by the non-progressive method. Our algorithm demonstrates features of fast convergence and high image fidelity. It is graceful, efficient, progressive, and flexible for full spectral rendering. In this dissertation, we contribute two efforts for spectral rendering. Firstly, we designe and implemente a WYSIWYG (what you see is what you get) visualization tool which can assist users in constructing spectral reflectance, including both the diffuse and the Phong specular reflectance model. Results show that this tool performs well for interactive operations with accuracy maintained. Secondly, we present a progressive refinement algorithm using a wavelet transformation for progressive, full spectral rendering. This technique provides not only quick image preview with visually plausible qualities during early progressive stages, but also a high progress percentage within few stages. By way of two techniques proposed in this dissertation, two problems encountered in spectral rendering, the difficulty of handling of data and SPDs for spectral rendering and high-complexity computing, can be solved.