NSF Young Investigator: Studies in Image Synthesis

真实感图像合成是建筑、照明和工业设计、卫星和机器人遥感、科学可视化、仿真系统、娱乐和广告等领域的重要问题。它还需要大量的计算。因此，精心选择的近似对于提高计算效率至关重要。不幸的是，很少有这样的近似承认良好的误差度量，这使得控制图像保真度充其量是困难的。在许多应用中，这种准确性是至关重要的。这项研究为一般的光传输系统开发了一种可控精度的渲染方法，既提供了可靠的精度界限，又允许算法将精力集中在误差减少最大的地方。除了照片级真实感图像合成，研究还开始于非照片级真实感渲染，这是一个具有巨大商业潜力的领域，到目前为止还相对较少受到关注。开发了支持从三维连续色调图像生成有效插图的机制。具体的研究目标包括创建传统插图技术的分类；为指定和计算插图开发新的建模范例、成像模型和用户交互方法；建立从具有更高细节级别的模型简化几何模型的方法；以及开发模拟传统艺术家工具的技术，如铅笔、粉笔、气刷和水彩。

Photorealistic image synthesis is an important problem for such diverse fields as architecture, lighting and industrial design, remote sensing from satellites and robots, scientific visualization, simulation systems, and entertainment and advertising. It also requires enormous computation. Carefully chosen approximations are therefore essential for making the computation efficient. Unfortunately, few such approximations admit good error metrics, which makes controlling image fidelity difficult at best. In many applications this accuracy is vital. This research develops a controlled-precision rendering approach for general systems of light transport both to provide reliable bounds on accuracy, and also to allow an algorithm to concentrate its effort where the error reduction is largest. In addition to photorealistic image synthesis, research begins in non-photorealistic rendering, an area with great commercial potential that has received relatively little attention to date. Mechanisms are developed to support the generation of effective illustrations from three-dimensional continuous-tone images. Specific research goals include creating a taxonomy of traditional illustration techniques; developing new modeling paradigms, imaging models, and user-interaction methods for specifying and computing illustrations; establishing methods for simplifying geometric models from models with a higher level of detail; and developing techniques for simulating traditional artist tools, such as pencil, chalk, air brush, and watercolor.