Anisotropic total variation model for removing oblique stripe noise in remote sensing image

Abstract

Destriping is one of typical problems in remote sensing image processing which is crucial in subsequent applications. The orientation of the stripe in remote sensing image plays a key role in the destriping methods. However, the direction of the oblique stripe noises is uncertain which makes the destriping methods for vertical/horizontal stripes no longer applicable for the oblique ones. To address the issue, we propose an anisotropic total variation (TV) model for oblique stripe noise removal. The model is constructed by combining L1 norm regularization and anisotropic total variation regularization, where the L1 norm regularizations represent the global characteristics and orientation of stripe noise, anisotropic total variation regularizations describe the different effects of oblique stripe noise on the smoothness of the original clear image along the vertical and horizontal directions, which can preserve the strong edges and geometric features while suppressing stripes. In order to solve this model, we design an effective alternating direction method of multipliers (ADMM) algorithm with guaranteed convergence. The experimental results over simulated and real datasets demonstrate that the proposed approach can not only effectively remove oblique stripe noise, but also for vertical/horizontal ones, and outperform the related state-of-the-art methods.