Subsampling of Multivariate Time-Vertex Graph Signals.

Pierre Humbert, Laurent Oudre, Nicolas Vayatis

Abstract: This article presents a new approach for processing and subsampling multivariate time-vertex graph signals. The main idea is to model the relationships within each dimension (time, space, feature space) with different graphs and to merge these structures. A new technique based on tensor formalism is provided, which aims to identify the frequency support of the graph signal in order to preserve its content after subsampling. Results are provided on real EEG data for data interpolation and reconstruction.