Sparse Bayesian Learning Via Variational Bayes Fused With Orthogonal Matching Pursuit

We address here the non-sparse signal reconstruction behavior of the Gaussian-inverse-Gamma model, in the context of compressive sensing using sparse Bayesian learning with variational Bayes (VB) inference. We estimate the numerical sparsity level of the signal of interest using sparse Bayesian learning and VB inference. Then, we feed the estimated sparsity level along with the estimated variance on the components of the sparse signal to the orthogonal matching pursuit algorithm to refine the reconstruction results. The results show the performance improvement of sparse signal recovery, with a reasonable computation cost.