A Two-Pass Realistic Image Synthesis Method for

Complex Scenes

Kurt Zimmerman and Peter Shirley

Department of Computer Science
Indiana University
Bloomington, IN 47405

Program of Computer Graphics
Cornell University
Ithaca, NY 14853

Abstract

This paper presents a survey of two-pass global illumination algorithms that use a local pass (gather) and shadow ray optimization techniques as well as a new two-pass algorithm which relies on spatial coherence to efficiently handle the problem of numerous luminaires. The algorithm presented operates on scenes with Lambertian reflectors and luminaires with arbitrary emission distributions. A radiosity prepass is performed on a simplified version of the scene. Bright reflecting surfaces are reclassified as light sources. A spatial data structure is then built that associates bright visible surfaces to regions in the environment. The visibility of these surfaces is probabilistically estimated. Finally, a view-dependent gather" is performed at each pixel. This gather is made significantly more efficient than previous algorithms because of careful integration based on the contents of the spatial data structure.
CR Categories and Subject Descriptors: I.3.0 [Computer Graphics]: General; I.3.6 [Computer Graphics]: Methodology and Techniques.
Additional Key Words and Phrases: Radiosity, realistic image synthesis, global illumination, Monte Carlo Integration.

1 Introduction

In 1986 Kajiya introduced the rendering equation[8] and with it introduced an algorithm that, given sufficient computing time, can accurately solve all rendering problems that assume geometric optics. This has led to some speculation that global illumination is a solved problem[4]. However, a sufficient amount of computing time for Kajiya's algorithm is not practical for most rendering applications. Thus, since the introduction of Kajiya's algorithm, the goal of global illumination community has been a simple one: develop algorithms which can render realistic images to some accuracy goal in a practical amount of time. While there have been successful algorithms for simple scenes, most of these algorithms do not scale well when the environment becomes complex. So in spite of the volume of literature devoted to realistic rendering, a general and practical algorithm for producing physically accurate images of complex environments does not exist.