Performance evaluation of MAP algorithms with different penalties, object geometries and noise levels

A new algorithm (LBFGS-B-PC) which combines ideas of two existing convergent reconstruction algorithms, relaxed separable paraboloidal surrogate (SPS) and limited-memory Broyden-Fletcher-Goldfarb-Shanno with boundary constraints (LBFGS-B), is proposed. Its performance is evaluated in terms of log-posterior value and regional recovery ratio. The results demonstrate the superior convergence speed of the proposed algorithm to relaxed SPS and LBFGS-B, regardless of the noise level, activity distribution, object geometry, and penalties.