Effective Compression Techniques for Precomputed Visibility

Abstract

In rendering large models, it is important to identify the small subset of primitives that is visible from a given viewpoint. One approach is to partition the viewpoint space into viewpoint cells, and then precompute a visibility table which explicitly records for each viewpoint cell whether or not each primitive is potentially visible. We propose two algorithms for compressing such visibility tables in order to produce compact and natural descriptions of potentially-visible sets. Alternatively, the algorithms can be thought of as techniques for clustering cells and clustering primitives according to visibility criteria. The algorithms are tested on three types of scenes which have very different structures: a terrain model, a building model, and a world consisting of curved tunnels. The results show that the natural structure of each type of scene can automatically be exploited to achieve a compact representation of potentially visible sets.