Two dictionaries matching pursuit for sparse decomposition of signals

Signal may be decomposed sparsely and power focally in an over-complete dictionary with matching pursuit (MP). In this paper, proposed is a modified NIP method named two dictionaries MP (TDMP) to decompose signal more sparsely. In the iteration procedure of TDMP, the over-complete dictionary is classified into two separate dictionaries with the selected and unselected atoms, and in each iteration, the algorithm was designed to have more chances than the original NIP to choose the atom in the selected atom dictionary as the optimal atom by the constraint of a simulate annealing threshold function, thus the algorithm avails for a more sparse decomposition. The decomposition results for a cosine-modulated exponential signal and an actual speech signal showed that the proposed TDMP could decompose signal more sparsely. (c) 2006 Elsevier B.V. All rights reserved.