Group sparsity based regularization model for remote sensing image stripe noise removal

Stripe noise degradation is a common phenomenon in remote sensing image, which largely affects the visual quality and brings great difficulty for subsequent processing. In contrast to existing stripe noise removal (destriping) models in which the reconstruction is performed to directly estimate the clean image from the striped one, the proposed model achieves the destriping by estimating the stripe component firstly. Since the stripe component possesses column sparse structure, the group sparsity is employed in this study. In addition, difference-based constraints are used to describe the direction information of the stripes. Then, we build a novel convex optimization model which consists of a unidirectional total variation term, a group sparsity term and a gradient domain fidelity term solved by an efficient alternating direction method of multiplier. Compared with the state-of-the-art methods, experiment results on simulated and real data are reported to demonstrate the effectiveness of the proposed method.