EUVIMEDEUVIMED entdecken
Crossref

Deep image prior regularized by weighted complex-valued total variation for in-line holography

Jing Ling, Jun Ma

Physica Scripta · 05.06.2026 · 101 (22) · 226001

DOI: 10.1088/1402-4896/ae7210 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Physica Scripta, online erschienen am 02.06.2026. Verfasst von Jing Ling und 1 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Innovation Project of GUET Graduate Education.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Lensless in-line digital holography involves an ill-posed single-shot phase retrieval problem and is highly susceptible to twin-image artifacts. Physics-informed deep image prior (DIP) methods enable single-shot phase retrieval without paired training data, but their performance remains sensitive to the choice of a single global regularization weight, making it difficult to balance artifact suppression and detail preservation. In this work, we propose DIP-weights complex-valued total variation (wCTV), a physics-informed single-shot phase retrieval method that incorporates a complex-valued total variation (CTV) prior with pixel-adaptive weights into the DIP framework. The resulting optimization problem is solved using the alternating direction method of multipliers, where the per-pixel weights are dynamically updated according to the global data misfit and the local gradient magnitude, thereby enabling edge-adaptive regularization and implicit annealing during optimization. Experiments on both simulated and experimental in-line holograms show that DIP-wCTV improves twin-image suppression and phase reconstruction quality while preserving fine structural details, compared with classical iterative methods and existing DIP-based approaches. Additional analyses on convergence and parameter sensitivity further support the practical effectiveness and robustness of the proposed method.

Förderung

  • National Natural Science Foundation of China
  • Innovation Project of GUET Graduate Education

Zitieren

APA

Ling, J., & Ma, J. (2026). Deep image prior regularized by weighted complex-valued total variation for in-line holography. Physica Scripta, 101(22), 226001. https://doi.org/10.1088/1402-4896/ae7210

Vancouver

Ling J, Ma J. Deep image prior regularized by weighted complex-valued total variation for in-line holography. Physica Scripta. 2026;101(22):226001. doi:10.1088/1402-4896/ae7210

BibTeX
@article{Ling2026,
  author = {Ling, Jing and Ma, Jun},
  title = {Deep image prior regularized by weighted complex-valued total variation for in-line holography},
  journal = {Physica Scripta},
  year = {2026},
  volume = {101},
  number = {22},
  pages = {226001},
  doi = {10.1088/1402-4896/ae7210}
}
Kostenlos starten →