.png){kind=icon}
Return
Enhancing light field image super-resolution through Mamba-based spatial–angular correlation learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s00371-025-04178-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Light field cameras capture both spatial intensity and directional information of light rays, enabling applications such as depth perception, 3D reconstruction, and view rendering. However, the simultaneous capture of high spatial and angular resolutions remains challenging, densely angular sampling results in sub-aperture images (SAIs) with low spatial resolution. In this paper, we address this issue by proposing a novel approach for LF image SR that leverages Mamba, a state-space model (SSM) equipped with a selective scan mechanism, to effectively capture nonlocal spatial–angular correlation. Our method transforms 4D LF images into 2D epipolar plane images (EPIs) and employs an SSM-based block with a selective scan mechanism to explore the spatial–angular correlation within EPIs. Additionally, we propose a pyramid feature extraction module (PFEM) to extract richer information during the early processing stages. Experimental results on five public benchmark datasets demonstrate that our method not only outperforms state-of-the-art LF image SR approaches but also maintains competitive performance in real-world scenarios, achieving PSNR gains of up to 0.85 dB over the next-best method on the EPFL dataset at $$\times 4$$ upscaling. The source code and datasets used in this study are publicly available at https://github.com/ChenSorry/MambaLFSR .
Keywords:
Super-resolution
Light field
Mamba
Spatial–angular correlation
Journal
T
IF:
0
Papers:
369
Citations:
0
Organization
View more organizations
Cited Papers
ShareSave
ShareSaveShareSaveShareSaveHierarchical feature fusion network for light field spatial super-resolution.png)
VISUAL COMPUTER
VISUAL COMPUTERIF2.9
PREAI
ShareSave View more