CAREER: Transient Signal Processing for Realistic Imagery

点击翻译按钮获取中文摘要

This research characterizes light transport from a signal-processing perspective with applications to image synthesis and material-appearance acquisition. A crucial aspect of visual signals is their frequency content, that is, whether light intensity varies smoothly or presents high-frequency content and discontinuities. While such characteristics are implicitly exploited by a number of techniques, there has been little work on the formal frequency analysis of global lighting. This research develops a theoretical framework that describes the effect of all light interaction in a scene in terms of frequency characteristics. This theory is also applied to develop more efficient image synthesis algorithms, and to solve inverse problems in material-property acquisition. The research provides strong signal-processing fundamentals to realistic graphics and enables faster and easier creation of high-quality computer graphics imagery.This research establishes a novel signal-processing framework for the mechanisms of light transport. The local frequency content of the radiance function is studied with an emphasis on how it is affected by phenomena such as shading, occlusion, and travel in free space. This work is comprehensive in that it treats both the angular and spatial dimensions of the radiance domain, which extends previous work that usually treated only one of these components. This theoretical work serves as conceptual inspiration and provides analytical formulas for applications. In particular, practical algorithms are developed for lighting simulation and real-time rendering. These algorithms exploit knowledge of the frequency characteristics of the radiance signal. The acceleration strategies rely on lower sampling rates when appropriate and perceptual masking due to high-frequency patterns. New techniques are also developed to acquire real-world material properties, including the acquisition and characterization of ground-truth data. This permits the validation of assumptions made in other projects, and also the exploitation of these characteristics for faster acquisition.