Automatic lighting design

Abstract

A significant problem in the automatic design of 3D graphics is the configuration of the lighting for a scene. The number of lights included, and the properties of these lights, has an enormous impact on what a viewer can judge about the content (the objects), properties (the geometric characteristics and spatial relations of the objects) and other aesthetic qualities of a scene. The traditional approach to lighting design for image synthesis is based on manual design methods, whereby users interactively specify values of lighting parameters, render the scene, and modify the lighting parameters until the desired visual properties of the scene are achieved. Non-expert users encounter a number of difficulties in selecting the appropriate lighting parameters, as the process requires both a subtle technical and aesthetic understanding of lighting in computer graphics.

In this thesis, perceptual aspects such as contrast and the non-linear characteristics of our perceptual response to colour are combined with practical studio lighting techniques and a novel treatment of shadows, to yield an extension to existing perceptual approaches to lighting design. This so-called ideal lighting approach optimises the lighting configuration for a scene with respect to a set of absolute perceptual metrics. An intuitive approach to lighting design, lighting-by-example, is also proposed and extensively explored in forms that exploit both the perception-based lighting framework and a new wavelet formulation. User studies are conducted both to configure the perception-based lighting objective function and to evaluate the performance of the proposed lighting design approaches. Finally, we develop an interactive interface for the lighting design process that incorporates both the ideal lighting and lighting-by-example approaches.