Description
GraFT (Graph-Filtered Temporal) is a signal extraction method for spatio-temporal data. It uses a diffusion map to learn a graph over spatial pixels that enables stochastic filtering of learned sparse representations over each pixel’s time-trace. The sparse representations are modeled as in a hierarchical dictionary learning framework with correlated decompositions over the graph.
Features
- Signal extraction for spatio-temporal data
- Uses diffusion maps for graph learning
- Stochastic filtering of learned sparse representations
- Hierarchical dictionary learning framework
- Correlated decompositions over the graph