East Asian J. Appl. Math., 8 (2018), pp. 44-69.
Published online: 2018-02
Cited by
- BibTex
- RIS
- TXT
Minimisation of the total variation regularisation for linear operators under $ℓ_1$-constraints applied to image restoration is considered, and relationships between the Lagrange multiplier for a constrained model and the regularisation parameter in an unconstrained model are established. A constrained $ℓ_1$-problem reformulated as a separable convex problem is solved by the alternating direction method of multipliers that includes two sequences, converging to a restored image and the “optimal" regularisation parameter. This allows blurry images to be recovered, with a simultaneous estimation of the regularisation parameter. The noise level parameter is estimated, and numerical experiments illustrate the efficiency of the new approach.
}, issn = {2079-7370}, doi = {https://doi.org/10.4208/eajam.210117.060817a}, url = {http://global-sci.org/intro/article_detail/eajam/10884.html} }Minimisation of the total variation regularisation for linear operators under $ℓ_1$-constraints applied to image restoration is considered, and relationships between the Lagrange multiplier for a constrained model and the regularisation parameter in an unconstrained model are established. A constrained $ℓ_1$-problem reformulated as a separable convex problem is solved by the alternating direction method of multipliers that includes two sequences, converging to a restored image and the “optimal" regularisation parameter. This allows blurry images to be recovered, with a simultaneous estimation of the regularisation parameter. The noise level parameter is estimated, and numerical experiments illustrate the efficiency of the new approach.