Quantification of Information Content in Different Frames of Reference
H. Jänicke, Gerik Scheuermann · 1981
Automatic detection of relevant structures in scientific data-sets can be achieved using techniques based on information theoretic measures. The methods that have been proposed so far, however, were restricted to Cartesian grids. In this paper we introduce a new information theoretic measure called linear local statistical complexity (LinearLSC). Locally, this new measure relies only on an individual position’s past and future, which makes it suitable for unstructured grids, point-based data and the analysis of boundary surfaces. LinearLSC is applicable to any kind and combination of data types (scalar, vector and tensor-valued) and computes for each position a single scalar value telling how informative this position is with respect to the entire data-set. Thus, relevant structures are automatically highlighted in an application independent, purely mathematic manner. Different data-sets from flow simulation are used to evaluate the properties of LinearLSC.