Joint regularization and low-rank fusion for atmospheric turbulence removal

Atmospheric turbulence removal remains a challenging task, because it is very difficult to mitigate geometric distortion and remove spatially and temporally variant blur. This paper presents a novel strategy for atmospheric turbulence removal by characterizing local smoothness, nonlocal similarity and low-rank property of natural images. The main contributions are three folds. First, a joint regularization model is made which combines nonlocal total variation regularization and steering kernel regression total variation regularization in order that reference image enhancement and image registration are jointly implemented on geometric distortion reduction. Secondly, a fast split Bregman iteration algorithm is designed to address the joint variation optimization problem. Finally, a weighted nuclear norm is introduced to constrain the low-rank optimization problem to reduce blur variation and generate a fusion image. Extensive experimental results show that our method can effectively mitigate geometric deformation as well as blur variations and that it outperforms several other state-of-the-art turbulence removal methods.