Rendering Filters for Controlling Detail and Creating Effects

This thesis demonstrates the way in which various methods for controlling detail and creating effects in computer graphics may be unified under the general theme of the rendering filter. Generally stated, such a filter is a passive, stateless operator that acts upon a decomposition of terms in the rendering equation. In the first Part, we present background that motivates this concept, and provides an understanding of the way in which the rendering filter follows logically from existing use in other domains. First, in Chapter 1, we discuss the general and historical use of the term "filter," especially as a useful metaphor that encapsulates various similar operations. We present examples of filters in photography, electronics, imaging and geometry processing. In Chapter 2, we provide background specific to rendering in graphics, examine the process of rendering as inherently related to filtering, and define the rendering filter itself. In the second Part, we see the application of these concepts by three specific examples of rendering filters. In addition to demonstrating the utility of the methods themselves, we show how these distinct algorithms are unified by the underlying rendering filter framework. In Chapter 3, we show various ways in which artists use "abstract," or otherwise unrealistic shadows to achieve compositional purposes; to allow similar control in real-time graphics, we present the stylized shadow filter. This filter accepts an intuitive set of controls, with which it converts an accurate shadow into a stylized shadow, and simulates many of the same artistic effects. Importantly, we present an algorithmic framework sufficiently efficient for real-time, interactive rendering. In Chapter 4, we present the subtractive shadow filter, which derives from the principle that rather than adding unshadowed light contributions from multiple lights in the scene to produce a shadowed result, we rather subtract extraneous light from an unshadowed rendering. By doing so, we enable user control over shadow level-of-detail in a progressive and flexible manner, so as to achieve a time/quality tradeoff over rendering, analogous to existing filtering techniques for other domains. In Chapter 5, we focus on improving the quality of realistic light transport through the path-density filter, which removes outlier points that degrade the quality of the scene. Their removal leads to significant improvement in rendering quality while retaining plausibility of the result. These particular filters serve as specific examples of rendering filter in practice, but by no means constitute the limit of what may be achieved according to this framework. It is hoped that by introducing these concepts we may stimulate the subsequent development of many additional methods.