Block-sparse signal recovery via forward-backward block pursuit algorithm

In this paper, a novel iterative greedy recovery algorithm for block-sparse signals, named the Forward-Backward Block Pursuit (FBBP), is proposed. It takes advantage of backtracking to refine the reliable blocks and reconstructs the block-sparse signals without the prior information of block sparsity. Unlike BMP and BOMP, FBBP iteratively enlarges the block support set by forward and backward steps. In each iteration, multiple blocks are added to the block support set by the forward step, then some of the blocks in the support set will be removed by the backward step. The number of blocks, which are selected by the forward step, is more than the number of blocks pruned by the backward step. Therefore, the block support set is enlarged at the end of each iteration. Several experiments are presented to demonstrate the excellent recovery performance of our proposed algorithm. © 2015, by Binary Information Press