Graph Fourier transform based on ?1 norm variation minimization

The definition of graph Fourier transform is a fundamental issue in signal processing. Conventional defined through the eigenvectors Laplacian matrix, which minimize ?2 norm variation. In this paper, we propose generalized based on ?1 variation minimization. We obtain necessary condition satisfied by basis, and provide fast greedy algorithm to approximately construct basis. Numerical experiments are also provided compare differences between basis