Holly Rushmeier

Comparing Real and Synthetic Images: Some Ideas About Metrics

This paper explores numerical techniques for comparing real and synthetic luminance images. We introduce components of a perceptually based metric using ideas from the image compression literature. We apply a series of metrics to a set of real and synthetic images, and discuss their performance. Finally, we conclude with suggestions for future work in formulating image metrics and incorporating them into new image synthesis methods.

Energy preserving non-linear filters

Monte Carlo techniques for image synthesis are simple and powerful, but they are prone to noise from inadequate sampling. This paper describes a class of non-linear filters that remove sampling noise in synthetic images without removing salient features. This is achieved by spreading real input sample values into the output image via variable-width filter kernels, rather than gathering samples into each output pixel via a constant-width kernel. The technique is nonlinear because kernel widths are based on sample magnitudes, and this local redistribution of values cannot generally be mapped to a linear function. Nevertheless, the technique preserves energy because the kernels are normalized, and all input samples have the same average influence on the output. To demonstrate its effectiveness, the new filtering method is applied to two rendering techniques. The first is a Monte Carlo path tracing technique with the conflicting goals of keeping pixel variance below a specified limit and finishing in a finite amount of time; this application shows how the filter may be used to “clean up” areas where it is not practical to sample adequately. The second is a hybrid deterministic and Monte Carlo ray-tracing program; this application shows how the filter can be effective even when the pixel variance is not known.

Rendering Participating Media: Problems and Solutions from Application Areas

Physically accurate rendering of radiatively participating media is an extremely demanding computational task. In this paper, current and potential applications requiring such renderings are reviewed. Some ideas for a practical rendering system, based on insights from application areas, are presented.

A system for measuring surface facet orientation from atomic force microscope data

The authors describe a graphical system developed for researchers in materials science for extracting information from data obtained by atomic force microscopy. In particular, they consider the problem of computing surface orientations from data obtained from ceramic materials. The visualization problems they consider in designing this system include finding useful mechanisms for the researcher to interact with the data, presenting results in forms familiar to the scientist, and enhancing traditional display techniques.

Specialized material models

While many materials can be defined as simple combinations of the general material models, other materials require specific models of small-scale geometric structure and multiple spatially varying layers. This chapter provides a guide to these specific models, which have been published in diverse, more specialized, conference documents and journals, as well as in major conference—ACM, IEEE, and Eurographics—proceedings. Some of these models have been curiosity motivated—researchers seeking to replicate intriguing appearance, whereas others are application motivated—researchers and practitioners seeking to replicate aspects of appearance that are critical in a particular simulation or design setting. Most of the models use a subset of effects coupled with domain-specific small-scale geometric models. Valuable specialized material models appear in a wide range of publications, from electronic proceedings of small computer graphics workshops to journals in fields outside of computer science. Listing of specialized materials is not an exhaustive survey of what exists in computer graphics. Further, many of the models discussed here do not boil down to simple equations for bidirectional reflectance distribution functions or bidirectional scattering surface reflectance distribution functions. Instead, many models involve complicated geometric structures, libraries of measured parameters, and nontrivial procedures. Our goal here is to provide an overview, with pointers to the relevant literature that provides greater detail.

Specifying and encoding appearance descriptions

This chapter reviews the various approaches to specification and discusses how material properties can be represented in the computer for efficient processing and rendering. The most common approaches to the specification of complex appearance characteristics relies on interactive tools and visual inspection by users, while others use the power of composition and programmability to define shaders and assemble complex appearance behavior from simple building blocks. The question of selecting the adequate parameters of a given appearance model, in order to achieve a desired look, remains a major difficulty for CG artists and designers. The simplest mechanism for specifying a set of appearance parameters for digital processing is, at least from a computers point of view, by means of a computer file. At the opposite end of the spectrum from interactive, direct-manipulation approaches, manually editing a catalog of parameter values is an effective, if tedious, way to control appearance. The Radiance lighting simulation system, which became a reference tool in professional lighting engineering in the 1990s, used this technique for handling material and light descriptions. Radiance material files use a fixed syntax modeled after the BRDF and lighting models used internally by the simulation program, and all parameter values are explicitly indicated. However, this system is not entirely monolithic since it also uses a simple form of programmability: Procedures (external programs) can be invoked to generate data such as patterns. The radiance system, therefore, offers a computer-oriented view of appearance specification, with a set of possible operations (programs) invoked using manually specified parameters.

General material models

This chapter presents the basic models for how materials scatter light. One approach for representation of material models is to store values of the reflectance for densely sampled values of position, wavelength, and direction. A table of such densely sampled models would require an enormous amount of storage for even simple objects. In addition to computational inefficiency, it is not a useful representation for a user to specify materials. A number of different criteria can be used to guide the construction of a good representation. Criteria include capturing the visible effects, obeying the physical constraints of conservation of energy, and reciprocity, compactness, complying with the electromagnetic wave theory of light, ease of evaluation, and meaningful parameters for user input. Different approaches choose to emphasize some subset of these criteria and ignore others. The most basic material models are bidirectional reflectance distribution functions (BRDFs) that describe the spectral and directional characteristics of a material at a spatial location. Although BRDFs often have complicated derivations and form, they generally reduce to evaluating an expression. It traces the development of the various named reflectance models such as Phong, Blinn, Cook–Torrance, and Oren–Nayar that appear in most graphics systems. It also puts in context more complex models that account for interference, diffraction, and volumetric scattering.

Observation and classification

This chapter emphasizes, in a technical framework, the way appearance is affected by how light rays reach the eye from a source to develop a computer model. Modeling materials begins with observation. One needs to observe what makes each material in the real world look different from other materials. The observation skills needed are similar to those developed by artists. Object appearance depends on shape, illumination, and perception as well as on the material itself. For any material there are spectral, directional, and spatial variations. There are many types of variations in each of these categories, and for many materials just one or two types of variations determine appearance. By classifying materials by the types of variations that are important, one greatly simplifies the mathematical and numerical models needed to reproduce material appearance. Material modeling does not start in front of a computer or by working from our often-faulty visual memories. It starts by looking carefully at the material one want to simulate, and possibly make some measurements. After this step, one can turn to selecting the appropriate models and determining parameters to drive a numerical simulation. Finally, applying the visual analysis process Wolfe describes for achieving visual literacy when observing physical materials is necessary to understand how to develop a model.

10 – Rendering appearance

Publisher Summary This chapter presents how material models are integrated into complete systems for generating images. Different representations of models are used for different rendering approaches. Different models can be appropriate for offline and real-time systems. The process of creating images with realistic appearance necessarily involves taking into account—either explicitly or implicitly—the interaction of light with the objects in a scene. Furthermore, rendering a digital image requires computing individual pixel colors, representing the appearance of an object as seen from the viewer. Spectral radiance is the relevant physical quantity characterizing light. Therefore, each pixel should be assigned a color based on the incoming spectral radiance at a viewing location. The finite resolution of the image, as well as the great variability of object colors even for neighboring locations, make it impossible to define a pixel color as simply the color of the visible object at that pixel. Two main classes of rendering techniques, based on the mechanism used to select samples are, object-space sampling and image-space sampling. In both cases, the essential operation consists in determining how a given point in space will appear from the viewpoint: either this determination is carried out for all elements of a scene, or it is only performed for elements that are visible in the image. It also discusses how lighting information is processed locally at (or below) the surface of an object and how local appearance characteristics are used for renderings. It concludes with describing precomputation strategies, allowing more interaction with the rendered elements.

4 – Mathematical terms

Publisher Summary This chapter defines the basic terms, concepts, and notations used in describing light and materials. One needs physically based models with parameters with physical meaning so that measuring materials, adjusting, and reusing them in scenes produce effects consistent with what one observes in the real world. To make physically based numerical definitions, there is a need to define various expressions for how incident light energy is distributed by a material with respect to position, direction, and wavelength. For many years, computer graphics systems used ad hoc descriptions of shades and lights making it difficult to compare and combine methods across different vocabularies, units, and scales. Adopting the standard terms used by the illumination and radiative transfer communities advanced successful realistic modeling. The key quantity in defining light transfer in a particular direction is radiance. The key quantity for expressing the directional effect of materials on the incident radiance is the bidirectional reflectance distribution function (BRDF). There are many materials that require an extended definition of scattering beyond the BRDF, but they are easily understood once radiance and BRDF are defined. While understanding the definitions of radiance and reflectance can be challenging, it is the key to being able to understand material models.