Matching Pursuit for Sparse Signal Reconstruction Based on Dual Thresholds

Abstract

Anumberofsparserecoveryapproacheshaveappearedintheliterature based on Orthogonal Matching Pursuit (OMP) algorithms because of its low computationalComplexity. Thismanuscriptintroducesanoveladaptive forward-back greedy approach, called Dual Threshold Matching Pursuit (DTMP), which select atoms based on two appropriate thresholds. During forward atom increasing process, DTMP picks out new candidate atoms based on the forward threshold under Restricted Isometry Constant (RIC) condition. In backward atom decreasing process, DTMP deletes wrong atoms based on the backward threshold according tothe principal of energy concentration. Like forward-backward pursuit (FBP), DTMP does not need the sparsity level in contrast to the Subspace Pursuit (SP) or Compressive Sampling Matching pursuit (CoSa MP) algorithms. Experimental results show that the reconstruction accuracy of DTMP surpasses SP, FBP and other greedy algorithms obviously and its complexity is comparable with those of OMP and SP.