This project aims to improve the quality of computer graphic images of outdoor, natural scenes. To date, renderings of outdoor terrain have had a cartoon-like quality that significantly distracts from a sense of realism. Partially, this is due to computational and source data constraints that limit the geometric complexity of terrain that can be rendered. The thesis of the research described here is that illumination and material properties play an equally important role in creating a sense of realism from these scenes. Moreover, there are important interactions between geometry, illumination, and material properties in a model of outdoor terrain that should be understood when real-time constraints must also be satisfied.
Significant progress has been made in the last decade in understanding how to generate realistic renderings of indoor scenes. The general approach is to analyze the physics of light transport in such environments and then to embody approximations to the physics in computational algorithms. Correct modeling of illumination and material properties is vital. It is now known that a sense of realism depends critically on accounting for shadows, secondary illumination, and non-uniform reflectance functions. Accurately approximating the effect of these properties involves great computational expense. As a result, methods for rendering realistic imagery almost always exploit assumptions about the nature of the geometric structure and illumination and materials properties likely to be encountered. Most of these assumptions derive from a presumption of indoor environments.
Outdoor scenes present very different computational characteristics. While the physics is the same, geometry, illumination, and reflectance properties are all distinctly different. Many of the techniques developed to support realistic rendering of indoor scenes will require substantial modifications for natural, outdoor environments. The most difficult computational problem to overcome is the need to be able to aggregate the effects of micro-structures into large enough units that they can be rendered effectively, while at the same time preserving key aspects of visual appearance. This problem exists across a wide range of scales, ranging from foliage, in which a collection of individual leaves generates a collective appearance that is quite different than that of the constituent members, to distant landmarks, where detail must be suppressed without removing those properties that make landmarks distinctive and thus useful.
The physics of light transport in outdoor scenes is sufficiently complex that full photorealism in generated imagery is not likely to be achieved in the foreseeable future. As a result, research in improving realism needs to be informed by a knowledge of the visual saliency of various aspects of the outdoor environment so that algorithm development and computational resources can be appropriately allocated. This requires a research team with a broad range of expertise. This project draws on researchers with substantial experience in photorealistic rendering, terrain modeling and simulation, and visual perception.
Our current efforts include:
This work is supported by NSF award 9731859.
| Projects | Home | Computer Science |