Compression of isosurfaces for structured volumes with context modelling

Abstract

In many applications surfaces with a large number of primitives occur. Geometry compression reduces storage space and transmission time for such models. A special case is given by polygonal isosurfaces generated from gridded volume data. However, most current state-of-the-art geometry compression systems do not capitalize on the structure that is characteristic of such isosurfaces, namely that the surfaces are defined by a set of vertices on edges of the grid. In a previous paper we had proposed a compression method for isosurfaces that is designed to exploit this feature. In this paper we use the same coding approach, however, including context models for the encoding of the symbol streams. We report improved compression ratios by about 20% for complex isosurfaces from a CT scan of a human head. For this data set our new coder outperformed state-of-the-art general purpose geometry compression methods by a factor of 2.6 to 3.4 in terms of compression ratio. We also report results obtained by two predictive coding methods based on least squares function fitting and a surface relaxation algorithm.