Fusion of Orthogonal Matching Pursuit and Least Squares Pursuit for Robust Sparse Recovery

We propose two approaches for obtaining a more robust algorithm for sparse signal recovery based on combining two existing algorithms, Orthogonal Matching Pursuit and the related Least Squares Pursuit. The first approach relies on averaging the gradient vectors used in the two algorithms for atom selection at every iteration. The second approach is to use the atom selected by the algorithm which has the highest confidence in its choice. Simulation results show more robust and well-balanced recovery performance, even in testing scenarios which are adverse to one or the other of the underlying algorithms.