Interactive stylized silhouette for point-sampled geometry

Silhouette is a drawing feature that is popular in illustrations and line-drawing artworks and is an important focus of research in the field of Non-Photorealistic Rendering (NPR). Current work on silhouette extraction and rendering tends to assume a polygonal mesh input, where connectivity information is available and the formation of silhouettes involves tracking of edges. On the other hand, the availability of 3D scanning devices nowadays leads to the proliferation of dense point data set for which no connectivity information is initially available. We present in this paper a hybrid image/object-space method to directly detect and render stylized silhouettes for such data set. Our method is simple to implement and runs at an interactive frame rate. Its basic form involves rendering silhouette points, each with a unique color value, to a color buffer. We show how lines can then be quickly fitted to the points projected. We further show how stylized silhouette can be used effectively with two other NPR techniques that happen to work very naturally with point-sampled geometry: stippling, and direct drawing onto surfaces.