Geometry driven local neighbourhood based predictors for dynamic mesh compression

Abstract

The task of dynamic mesh compression seeks to find a compact representation of a surface animation, while the artifacts introduced by the representation are as small as possible. In this paper we present two geometric predictors, which are suitable for PCA based compression schemes. The predictors exploit the knowledge about the geometrical meaning of the data, which allows a more accurate prediction, and thus a more compact representation. We also provide rate/distortion curves showing that our approach outperforms the current PCA-based compression methods by more than 20%.