NLTV-Gabor-based models for image decomposition and denoising

By using the nonlocal total variation (NLTV) as the regularization and Gabor functions as the fidelity, this paper proposes two novel models for image decomposition and denoising. The presented models closely incorporate the advantages of the NLTV and Gabor wavelets-based methods. These improvements are aimed at overcoming the drawbacks of staircase artifacts and loss of edge details caused by the traditional variational frameworks. Furthermore, on the basis of Chambolle's projection algorithm, we introduce two extremely efficient numerical methods to solve the resulting optimization problems. Finally, compared with several popular and powerful numerical methods, this article confirms the superiorities of the developed strategies for image decomposition and denoising in terms of visual quality and quantitative assessments.