Analysis and implementation of the silhouette map data structure and its application to computer graphics.

摘要

The real-time texture mapping pipeline on graphics hardware consists of a series of sampling and filtering steps that introduces artifacts in our final image. In this work, we develop a novel data structure called a silhouette map that augments a standard texture with boundary information so that we can reconstruct textures with boundaries more accurately. Its compact representation is designed to run on graphics hardware at real-time rates. The resulting renderings are of improved perceptual quality because boundary edges provide important visual cues in many kinds of textures.; Since the silhouette map technique can allow us to match the reconstructed quality of a texture of much larger size, it can be considered a form of image compression. In this work, we compare our algorithm against the standard compression techniques currently used in graphics hardware and show that we are able to achieve similar quality at a fraction of the size. We also extend our work to an adaptive representation, which further reduces the storage overhead by adapting the sampling rate to the detail of the original texture signal.; To generate our silhouette maps, we present an automated algorithm that allows us to synthesize them directly from bitmapped images. We use this to demonstrate the effectiveness of silhouette maps particularly for textures with a significant amount of geometrical information (e.g. manmade textures such as logos and signs).; The flexibility of our structure means that silhouette maps can replace standard texture maps in many applications. As an example, we apply silhouette maps to shadow maps to reduce the artifacts commonly associated with this technique. As the results presented indicate, all of the implementations of silhouette maps run in real time on current graphics hardware.