The University of Sheffield
Department of Computer Science

Joseph Stevens Undergraduate Dissertation 2000/01

"Subdivision Surfaces for Computer Games"

Supervised by A.Watt

Abstract

Along with the ever-increasing technological advances in computer hardware, come the demands of the media and public to push these improvements to and beyond their limits with the help of the latest software. This requires computer game developers to produce software that looks and feels more realistic, but also runs faster and more smoothly. As hardware capable of displaying and manipulating complex, three-dimensional geometric objects became more readily available, subdivision surfaces became much more significant to game developers with their ability to produce approximations of perfectly smooth, topologically arbitrary surfaces, using relatively simple recursive algorithms.

The paper presents a qualitative evaluation of four of the most common subdivision schemes (the Butterfly, Catmull-Clark, Doo-Sabin and Loop schemes), regarding their appropriateness and particular benefits, when used to model smooth surfaces that are subject to manipulation. In order to compete the aims of this project a tool that would enable the investigation of each of the different subdivision schemes was developed. The investigative work carried out using the tool that had been developed, along with the detailed knowledge of the workings of each subdivision schemes gained throughout the course of the project, enabled a detailed analysis of each of the schemes to be carried out; producing some interesting results.

When choosing a subdivision surface to model a deformable object it is essential to know in what situations the surface will be used, and how it will be deformed. Interpolating schemes, produce a surface that is more the smoothed vision of the base mesh and will have more well defined curves; meshes at low levels of surface division closely resemble the limit surface. Approximating schemes produce a surface that is a continuously curved interpretation of base mesh and will not have any well-defined features; meshes at low levels of surface division bear little resemblance to the limit surface. In general approximating schemes tend to smooth deformations into the surrounding surface as it is divided, although greater definition can be gained at higher levels of subdivision, whereas interpolating schemes produce a more well defined deformation.