Advanced Global Illumination

Philip Dutre, Philippe Bekaert, Kavita Bala

• Publisher: A K Peters/CRC Press
• Year: 2018
• ISBN:
  9781315359878
• Edition: 2
• Topic: Computer Science
• Subtopic: Computer Graphics
• Index: Computer Science

This book provides a fundamental understanding of global illumination algorithms. It discusses a broad class of algorithms for realistic image synthesis and introduces a theoretical basis for the algorithms presented. Topics include: physics of light transport, Monte Carlo methods, general strategies for solving the rendering equation, stochastic path-tracing algorithms such as ray tracing and light tracing, stochastic radiosity including photon density estimation and hierarchical Monte Carlo radiosity, hybrid algorithms, metropolis light transport, irradiance caching, photon mapping and instant radiosity, beyond the rendering equation, image display and human perception. If you want to design and implement a global illumination rendering system or need to use and modify an existing system for your specific purpose, this book will give you the tools and the understanding to do so.

1

Introduction

1.1 What Is Realistic Image Synthesis?

Realistic image synthesis is a domain in the field of computer graphics that generates new images from other data. Typically, one starts from a complete description of a three-dimensional scene, specifying the size and location of objects, the material properties of all surfaces of solid objects in the scene, and the position and emission characteristics of light sources. From this data, a new picture is generated, as seen from a virtual camera placed in the scene. The aim is to make these pictures as photorealistic as possible, such that the difference with a real photograph (if the virtual scene would be constructed in reality) is not noticeable. This requires the underlying physical processes regarding materials and the behavior of light to be precisely modeled and simulated. Only by knowing exactly what one is trying to simulate does it become possible to know where simplifications can be introduced in the simulation and how this will affect the resulting pictures.

Generating photorealistic pictures is a very ambitious goal, and it has been one of the major driving forces in computer graphics over the last decades. Visual realism has always been a strong motivation for research in this field, and it is a selling point for many graphics-related, commercially available products. It is expected that this trend will continue in the coming years and that photorealism will remain one of the core fields in rendering.

Photorealistic rendering is not the only rendering paradigm that is used in computer graphics, nor is it the best solution for all rendering applications. Especially in the last couple of years, non-photorealistic rendering has become a field in itself, providing viable alternatives for the photorealistic rendering style. Non-photorealistic rendering (or NPR, as it is commonly called) uses a wide variety of drawing styles that are suited for a more artistic, technical, or educational approach. Drawing styles covered by NPR include pen-and-ink drawings, cartoon-style drawings, technical illustrations, watercolor painting, and various artistic styles such as impressionism, pointillism, etc. The possibilities are virtually limitless, and the algorithms are driven by trying to recreate a certain style rather than by trying to simulate a physical process found in nature. While there is clearly room for NPR, a variety of applications are interested in the physical simulation of reality.

1.1.1 The Importance of Realistic Image Synthesis

Photorealistic rendering is a rendering style that has many applications in various fields. Early applications were limited by the amount of time it took to compute a single image (usually measured in hours), but recently, interactive techniques have broadened the scope of photorealistic image synthesis considerably.

Film and Visual Effects

Visual effects in the film industry have always been a driving force for the development of new computer graphics techniques. Depending on the rendering style used, feature animations can benefit from global illumination rendering, although this might be limited to a few scenes where more complex lighting configurations are used. Movies with live footage can benefit too, especially when virtual elements are added. In this case, a consistent lighting between the real and virtual elements in the same shot needs to be achieved, in order to avoid implausible lighting effects. Global illumination is necessary to compute the light interaction between those different elements.

Architecture

Architectural design is often quoted as one of the most beneficial applications of photorealistic rendering. It is possible to make visualizations, whether they be still images or interactive walk-throughs, of buildings yet to be constructed. Not only can indoor illumination due to interior lighting be simulated, but outdoor lighting can be considered as well, e.g., the building can be illuminated using various atmospheric conditions at different times of the year, or even various times during the day.

Ergonomic Design of Buildings and Offices

Although not strictly a computer graphics application, the ergonomic design of office rooms or factory halls is very much related to global illumination. Given the design of a building, it is possible to compute the various illumination levels in different parts of the building (e.g., desks, workstations, etc.), and the necessary adjustments can be made to reach the minimum legal or comfortable requirements by changing the color of paint on the walls, the repositioning of windows, or even the replacement of walls.

Computer Games

Most computer games revolve around fast and interactive action, coupled with a suspension of disbelief in a virtual world. As such, photorealistic rendered imagery probably is a strong cue to draw players into the environments in which their virtual characters are acting. Since interactivity is more important in a gaming context than realistic images, the use of global illumination in games is still somewhat limited but will undoubtedly become more important in the near future.

Lighting Engineering

The design of lights and light fixtures can also benefit from global illumination algorithms. Specific designs can be simulated in virtual environments, such that the effect of the emission patterns of light bulbs can be studied. This requires an accurate measurement and modeling of the characteristics of the emission of the light sources, which is a whole field of study by itself.

Predictive Simulations

Predictive simulations encompass much more than just simulating the look of buildings as described above. Other areas of design are important as well: car design, appliances, consumer electronics, furniture, etc. This all involves designing an object and then simulating how it will look in a real or virtual environment.

Flight and Car Simulators

Simulators used for training, such as flight and car simulators, benefit from having an as accurate as possible visual simulation, e.g., aspects of street lighting are important in car simulators, accurate atmospheric visual simulation is important when designing a flight simulator, etc. Other types of training simulators also use or might use realistic imagery in the future; armed combat, ship navigation, and sports are a few examples.

Advertising

Producing accurate imagery of yet-to-be-produced products is probably a good tool for the advertising world. Not only does the customer have the ability to see what the product looks like when generated using photorealistic rendering, but he would benefit if he could place the product in a known environment, e.g., furniture could be placed, with consistent illumination, in a picture of your own living room.

1.1.2 History of Photorealistic Rendering

This section provides a brief history of photorealistic rendering algorithms and the quest for visual realism. Some more extensive background and history on specific algorithms can also be found in the relevant chapters.

Photorealism in the Precomputer Age

The history of photorealistic rendering, or the quest for visual realism, can be traced throughout the history of art. Although we are mainly interested here in the computer-driven photorealistic rendering programs, it might be useful to look at how the understanding of realistic rendering evolved in the course of history. Medieval and premedieval art is very much iconic in nature: persons and objects are displayed in simplified, often two-dimensional forms, and sizes and shapes are used to reflect the importance of the person displayed, relative positioning in a scene, or other properties.

The real beginning of realistic rendering probably starts with the first use and study of perspective drawings. Especially in Italy during the Renaissance period, various artists were involved in discovering the laws of perspective. Brunelleschi (1377–1446), da Vinci (1452–1519), and Dürer (1471–1528) (to name a few) are well known for their contributions. Later, painters also started to pay attention to the shading aspects. By carefully studying shadows and light reflections, very accurate renderings of real scenes could be produced using classic artistic materials.

Much of the knowledge of photorealistic painting was collected by British landscape artist Joseph Turner (1775–1851), appointed Professor of Perspective at the Royal Academy of Arts in London. He designed a course of six lectures, covering principles such as accurate drawing of light, reflections, and refractions. Some of his sketches show reflections in mirrored and transparent spheres, a true precursor of ray tracing almost 300 years later. In his book, Secret Knowledge, British artist David Hockney [73] develops an interesting thesis: Starting in the 15th century, artists began using optical tools to display reality very accurately. Mostly using a camera lucida, they projected imagery onto the canvas and traced the silhouettes and outlines very carefully. Afterwards, several such drawings were composed in a bigger painting, which explains the different perspectives found in various well-known paintings of the era.

It is certainly no coincidence that the trend and developments towards more photorealism in art were somewhat halted with the invention of photography at the beginning of the 19th century (Nicéphore Niépce, 1765–1833). Capturing an image accurately is suddenly not a difficult process anymore. After the invention of photography, art evolved into modern art, with its various novel ways, not necessarily photorealism, of looking at reality (pointillism, impressionism, cubism, etc.).

Primitive Shading Algorithms

The birth of computer graphics is usually accredited to SketchPad [188], the Ph.D. thesis of Ivan Sutherland at the Massachusetts Institute of Technology (M.I.T.) in 1963. Early computer graphics were mostly line drawings, but with the advent of raster graphics, shading algorithms became widely available. Primitive shading algorithms usually attributed a single color to a single polygon, the color being determined by the incident angle of light on the surface. This type of shading gives some cues about shape and orientation but is far away from realistically illuminated objects.

A breakthrough was achieved by Henri Gouraud and Bui Tui Phong, who realized that by interpolation schemes, additional realism in shading can be easily achieved. Gouraud shading [58] computes illumination values at vertices and interpolates these values over the area of a polygon. Phong shading [147] interpolates the normal vectors over the area of a polygon and computes illumination values afterwards, thus better preserving highlights caused by nondiffuse reflection functions. Both techniques are longstanding shading algorithms in computer graphics and are still widely used.

Another major breakthrough for more realism in computer-generated imagery was the use of texture mapping. Using a local two-dimensional coordinate system on an object, it is possible to index a texture map and attribute a color to the local coordinate. Integration in the rendering process involves a two-dimensional coordinate transform from the local coordinate system on the object to the local coordinate system of the texture map. Once texture mapping was able to change the color of points on a surface, it was fairly straightforward to change other attributes as well. Thus, the techniques of bumpmapping, displacement mapping, environment mapping, etc., were added. Texturing remains one of the building blocks for rendering in general.

Additional research was also performed in the area of light-source modeling. Originally only point light sources or directional light sources were used in the earliest rendering algorithms, but fairly soon spotlights, directional lights, and other types of light sources, sometimes emulating those found in lighting design, were introduced. Together with the modeling of light sources, the accurate portrayal of shadows has received much attention. When using point light sources, the computation of shadows can be reduced to a simple visibility problem from a single point of view, but the shadows are sharp and hard. The use of shadow volumes and shadow maps are among the best-known algorithms and still receive attention for improvement.

Ray Tracing

In 1980, ray tracing, probably the most popular algorithm in rendering, was introduced by Turner Whitted [194]. Although the principle of tracing rays was used before to generate correct perspective and shadows in the conventional arts, the tracing of rays in computer graphics was a major idea for generating all sorts of photorealistic effects. The original paper used rays for determining visibility through a single pixel (also known as ray casting) but also used rays to compute direct illumination and perfect specular and refractive illumination effects. As such, this seminal paper described a major new tool in generating images.

The ray-tracing algorithm has been researched and implemented extensively during the last two decades. Initially, much attention was on efficiency, using well-known techniques such as spatial subdivision and bounding volumes. More and more, the focus was also on lighting effects themselves. By treating ray tracing as a tool for computing integrals, effects such as diffuse reflections and refractions, motion blur, lens effects, etc. could be computed within a single framework. For a nice overview, the reader is referred to [52].

The original paper did not solve the entire global illumination problem but was very influential for later developments. To make a distinction with more modern ray-tracing algorithms, the first algorithm is sometimes referred to as Whitted-style ray tracing or classic ray tracing. Many present-day global illumination algorithms at the core are ray tracers, in the sense that the basic tool still is a procedure that traces a ray through a threedimensional scene.

Since a basic ray tracer is rather easy to implement, it is a very popular algorithm to serve as the first step into photorealistic rendering. It is traditional to have undergraduate students implement a ray tracer in many graphics courses. Many computer graphics enthusiasts post their ray tracers on the internet, and many of the more popular rendering packages have ray-tracing roots.

Radiosity

With ray tracing being well underway in the first half of the eighties as the algorithm of choice for realistic rendering, it became clear that ray tracing also had severe limitations. Indirect illumination effects such as color bleeding and diffuse reflections were very difficult to simulate. It was clear that a solution needed to be found if one wanted to produce photorealistic pictures. The answer came in the form of ...