Joint of discrete curvelet transform and nonlocal tensor sparse regularization for SAR image despeckling

This paper proposes a new joint sparsity regularization model (JSRM) for synthetic aperture radar (SAR) image despeckling by characterizing local sparsity and nonlocal self-similarity of SAR images simultaneously. The proposed model contains a data fidelity term and two regularization terms. One of the two regularization terms employs the discrete curvelet transform to depict the local smoothness of the SAR image, and the other one employs a tensor sparse transform of the three-dimensional (3D) tensor generated by stacking similar SAR image patches. The joint employment of these two regularization terms, which has not been utilized in existing methods of SAR image despeckling yet, aims to produce better despeckling performance and preserve more geometrical features of SAR images. To address the optimization problem in the proposed model, a new efficient algorithm is derived based on the split Bregman iterations framework. Experimental results show that the proposed model considerably outperforms some conventional and state-of-the-art techniques in terms of both subjective visual assessment of image quality and objective evaluation.