Blind Deconvolution via Lower-Bounded Logarithmic Image Priors

摘要: In this work we devise two novel algorithms for blind deconvolution based on a family of logarithmic image priors. contrast to recent approaches, consider minimalistic formulation the problem where there are only energy terms: least-squares term data fidelity and an prior lower-bounded logarithm norm gradients. We show that is sufficient achieve state art in with good margin over previous methods. Much performance due chosen prior. On one hand, very effective favoring sparsity other non convex. Therefore, solutions can deal effectively local minima become necessary. iterative minimization at each iteration solve convex problems: obtained via primal-dual approach majorization-minimization. While former computationally efficient, latter achieves state-of-the-art public dataset.