"Silhouette Detection and Enhancement in Realtime Nonphotorealistic Rendering"

Supervised by S.Maddock

Abstract

The purpose of this project is to extract the silhouette from existing polygon mesh models and enhancing them. It does this by bringing together NPR work done in object space, (more specifically in the field of silhouette edge detection) and then by using ideas from previous image-based approaches seeks to enhance the silhouette images produced. Building on the work done by Lee Markosian et al at Brown University, this project first builds a data structure to allow it to implement some of the algorithms explained in their work and then seeks to solve some of the problems associated with them. It takes two approaches in attempting to abstract out strokes, one by directly searching the existing data structure and finding nearby silhouette edges and the other by first building a map of edges to traverse. The project compares and contrasts these two techniques and then shows how by altering parameters within them we can make silhouette detection both more efficient and silhouette images more suitable for enhancement and manipulation. Finally, it applies a number of enhancements to the strokes that make them look more visually appealing and then concludes by suggesting ways the techniques developed can be used for future work.