Collaborative Research: Theory and Algorithms for High Quality Real-Time Rendering and Lighting/Material Design in Computer Graphics

Computer graphics is commonly used for interactive visualization and rendering in video games, electronic commerce or scientific visualization. These applications often demand real-time results, including multiple bounces of light (global illumination), material changes and spatially-varying local lighting. Computer graphics is also increasingly used to prototype or design illumination and material properties, for industries as diverse as animation, entertainment, automobile design, and architecture. A lighting designer on a movie set wants to pre-visualize the scene lit by the final illumination and with objects having their final material properties — be they paint, velvet or glass. An architect wants to visualize the reflectance properties of building materials in their natural setting. In many applications, much greater realism and faithfulness can be obtained if the lighting or material designer could interactively specify these properties. In this research, we are developing the theoretical foundations and next generation practical algorithms for high quality real-time rendering and lighting/material design. Our research involves both significant theoretical and practical components. We are developing a new theoretical framework for analyzing the dimensionality of local object regions and locally-low dimensional approximation. We also analyze the theory and practical algorithms for designing materials, while rendering the final scene with full global illumination. Because appearance is not linear in the BRDF, our formulation involves a multi-linear tensor representation to handle multiple bounces. For interactive rendering of scenes with local lighting, and for general lighting design, we are developing an approach that relights a scene given a full incident light field. Finally, one of the banes of precomputation-based approaches is the long time for precomputation, especially once global illumination effects are taken into account. We are developing the theory and algorithms for a new photon-mapping approach that substantially accelerates the process.