"Matching Pursuit with Localized Frames"Holböck, JakobMatching Pursuit is a greedy algorithm that is used to approximate signals by finite linear combinations of atoms from a given dictionary. During each iteration, one selects the atom that has the largest correlation with the residual and adds a suitable multiple to the current approximation. In order to reduce the computational, we assume that the dictionary is a localized frame. Thereby we can approximate the Gramian by a band matrix and introduce an approximate updating step for the correlations. This is a generalization of an analogous approach for multi-window Gabor dictionaries. The poster is based on joint work with Peter Balazs. |
« back