EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Scene‐Adaptive Nonlinear Tone Mapping for Low‐Light Novel View Synthesis

Mingzhe Lyu, Jinqiang Cui, Hong Zhang

IET Image Processing · 2026-01 · 20 (1)

DOI: 10.1049/ipr2.70505 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Image Processing, online erschienen am 11.11.2026. Verfasst von Mingzhe Lyu und 2 weiteren Autor:innen. Die Arbeit nennt 61 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Science, Technology and Innovation Commission of Shenzhen Municipality, Southern University of Science and Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Low‐light novel view synthesis (NVS) remains difficult because dark multi‐view inputs suffer from amplified sensor noise and compressed dynamic range, while paired normal‐light references are typically unavailable. Prevailing pipelines therefore synthesise a normal‐light pseudo ground‐truth from the dark input as the training target, which provides the dominant gradient signal. Despite this critical role, existing research focuses on architecture, rendering or loss design, leaving the target to simple linear scaling: a gain that lifts shadows inevitably clips highlights, yielding zero gradients and lost texture. Nonlinear tone mapping is standard in 2D low‐light enhancement, yet has not been transferred to training‐target design for 3D NVS. To address this gap, we propose a scene‐adaptive nonlinear tone‐mapping framework combining percentile‐based normalisation, an adaptive black‐level offset and two complementary curves, adaptive SoftExp and adaptive Poly3, as a drop‐in replacement with the reconstruction backbone unchanged. Across three benchmarks covering 21 scenes, a unified offset configuration consistently improves PSNR over the linear baseline, confirmed over three seeds, gaining up to 4.34 dB on LOM and 3.25 dB on RealX3D.

Förderung

  • Science, Technology and Innovation Commission of Shenzhen Municipality
  • Southern University of Science and Technology

Zitieren

APA

Lyu, M., Cui, J., & Zhang, H. (2026). Scene‐Adaptive Nonlinear Tone Mapping for Low‐Light Novel View Synthesis. IET Image Processing, 20(1). https://doi.org/10.1049/ipr2.70505

Vancouver

Lyu M, Cui J, Zhang H. Scene‐Adaptive Nonlinear Tone Mapping for Low‐Light Novel View Synthesis. IET Image Processing. 2026;20(1). doi:10.1049/ipr2.70505

BibTeX
@article{Lyu2026,
  author = {Lyu, Mingzhe and Cui, Jinqiang and Zhang, Hong},
  title = {Scene‐Adaptive Nonlinear Tone Mapping for Low‐Light Novel View Synthesis},
  journal = {IET Image Processing},
  year = {2026},
  volume = {20},
  number = {1},
  doi = {10.1049/ipr2.70505}
}
Kostenlos starten →