Construction of undersampled graph filter banks via row subset selection
Akie Sakiyama, Yuichi Tanaka · 2016
This paper introduces a construction method of M-channel under-sampled spectral graph filter banks. They can be applied to any kind of undirected graphs, use arbitrary critically sampled or oversampled analysis filters, and obtain low redundancy, which is less than 1, regardless of the number of the analysis filters. We formulate the construction problem as a row subset selection method of the transform matrix of the prototype (critically sampled or oversampled) filter banks. In the experiment, a graph signal on Minnesota Traffic Graph is decomposed to examine the performance of our spectral graph filter banks.