HFPnet: a hybrid feature perception network for high-quality low-light light field enhancement
Abstract Existing algorithms for low-light light field enhancement (L3FE) generally suffer from detail loss, severe noise interference and poor cross-view consistency, which greatly degrade the reconstruction quality of L3FE. To solve these problems, the hybrid feature perception network (HFPnet) is proposed for high-quality L3FE. Firstly, a hybrid decoupling module (HDM) is designed to extract spatial features, angular features and omnidirectional EPI geometric features of light field (LF) images. Based on HDM, three dedicated optimization modules are constructed, including a reflection estimation module (REM), a gradient prior module (GPM) and an illumination estimation module (IEM). These modules collaboratively compensate for missing texture details, suppress image noise and restore distorted illumination information to achieve high-fidelity L3FE and stable cross-view consistency. Extensive experimental results verify that the proposed HFPnet outperforms current state-of-the-art (SOTA) algorithms.
Authors
- Daming Wang (ORCID: https://orcid.org/0000-0002-1897-2874)
- Xuanwei Liu
- Shangshang Yan
- Ziji Liu
- Yiran Liu
- Hongbo Zhang
Institutions
- University of Electronic Science and Technology of China (CN)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1038/s41598-026-80289-4
- Primary Topic
- Image Enhancement Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00