Dynamic group allocation reconstruction for group sparse signals

The reconstruction process in compressive sensing is an underdetermined problem. Existing reconstruction algorithms in compressive sensing approaches this problem using the standard sparsity model, where nonzero terms are randomly isolated among the signal. However, natural signals are tended to show group sparsity which may help in the reduction of computational complexity and acceleration of recovery processes of signals if being incorporated in algorithm designs. GOMP, modified from OMP, includes the concept of groups, yet because of fixed group division, GOMP suffers low column selection accuracy. Besides, the group distribution is unknown in most cases, fixed group division in GOMP is impractical. In this paper, we propose a new reconstruction algorithm, dynamic group allocation reconstruction (DGAR). The results show that the proposed method has better column selection accuracy in the context of group sparsity compared to its predecessors. Proposed DGAR algorithm is 50% and 100% higher in accuracy compared to GOMP algorithm and StOMP respectively when sparsity is 40 and group sparsity is 2.