An optimized version of non-negative OMP

This article addresses least-squares minimization under sparsity and non-negativity constraints. We propose a recursive implementation of Non-Negative Orthogonal Matching Pursuit (NNOMP) based on the active set method for solving least-squares subproblems. We further propose an improvement of NNOMP, named support-Shrinkage NNOMP (SNNOMP), based on the shrinkage of the support of iterates when some coordinates vanish. SNNOMP is compared with the existing versions of NNOMP for a sparse deconvolution problem.