面向低空场景的光场高斯泼溅表示及空间-角度同时超分辨率方法

孟佳涵; 杨驰; 李传君; 林丽莉; 周文晖

doi:10.13700/j.bh.1001-5965.2025.0797

面向低空场景的光场高斯泼溅表示及空间-角度同时超分辨率方法

doi: 10.13700/j.bh.1001-5965.2025.0797

1.
杭州电子科技大学计算机学院，杭州 310018
2.
浙江工商大学信息与电子工程学院，杭州 310018

基金项目:

国家自然科学基金(62471437)；浙江省“尖兵”“领雁” 科技计划项目(2023C01030)

详细信息

通讯作者:
E-mail：zhouwenhui@hdu.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 250
- HTML全文浏览量: 92
- PDF下载量: 16
- 被引次数: 0
出版历程
- 收稿日期: 2025-11-17
- 录用日期: 2025-12-12
- 网络出版日期: 2026-02-02
- 整期出版日期: 2026-09-01

Light field gaussian splatting representation and joint spatial-angular super-resolution for low-altitude scenarios

1.
School of Computer Science and Technology，Hangzhou Dianzi University，Hangzhou 310018，China
2.
School of Information and Electronic Engineering，Zhejiang Gongshang University，Hangzhou 310018，China

Funds:

National Natural Science Foundation of China (62471437); ‘‘Pioneer’’ and ‘‘Leading Goose’’ Research and Development Program of Zhejiang (2023C01030)

More Information

Corresponding author: E-mail：zhouwenhui@hdu.edu.cn

摘要

摘要:
光场成像技术可同时捕获光线的空间位置和角度信息，在无人机感知、环境监测等低空经济应用场景中具有广阔应用前景。然而，现有光场相机成像及超分辨率方法难以兼顾空间分辨率与角度分辨率，制约了其在精确感知和高精度建模等任务中的应用。为此，提出一种面向低空场景的光场高斯泼溅表示及光场空间-角度同时超分辨率（LFASSR）方法。将传统3D高斯泼溅建模方法拓展到4D光场表示空间中，提出一种4D光场高斯泼溅表示方法，并利用光场极平面（EPI）的空间-角度几何特性，将其分解为水平与垂直方向上的2D高斯泼溅级联建模，以更有效地表征光场几何特征。基于此，构建了一个通用的光场图像超分辨率框架，在提取并融合光场空间、角度及EPI域特征后，构建水平和垂直方向的光场几何特征空间，并分别进行2D高斯泼溅建模，最终经EPI渲染，生成致密视角且高空间分辨率的高质量光场图像。在公开光场图像数据集及低空场景光场数据上的实验结果表明：所提方法在LFASSR任务中显著优于现有方法，在峰值信噪比（PSNR）和结构相似性指标（SSIM）等客观指标上均达到当前最优水平，在视觉质量上展现出清晰的细节恢复能力与自然的视差过渡效果。在独立的光场空间超分辨率（LFSSR）任务和光场角度超分辨率（LFASR）任务中，所提方法同样展现出良好的泛化能力与竞争力。
- 光场 /
- 光场超分辨率 /
- 光场高斯泼溅 /
- 空间-角度同时超分辨率 /
- 极平面几何
Abstract:
Unmanned aerial vehicle (UAV) perception and environmental monitoring are two low-altitude commercial applications where light field imaging technology shows significant promise by simultaneously capturing the spatial and angular information of light rays. However, existing light field cameras and super-resolution methods struggle to jointly optimize spatial and angular resolutions, limiting their effectiveness in tasks requiring precise perception and high-fidelity reconstruction. To address this challenge, this paper proposes a light field Gaussian splatting representation and a light field angular spatial super-resolution (LFASSR) method tailored for low-altitude scenarios. Specifically, we extend the conventional 3D Gaussian splatting framework into the 4D light field domain and propose a 4D light field Gaussian splatting representation. Leveraging the geometric structure of light field epipolar plane images (EPIs), we decompose the 4D representation into cascaded 2D Gaussian splatting models along horizontal and vertical directions, enabling more efficient modeling of light field geometry. Based on this representation, we develop a unified light field super-resolution framework. The framework first extracts and fuses features from the spatial, angular, and EPI domains, then constructs directional geometric feature spaces for horizontal and vertical views, where 2D Gaussian splatting is applied separately. Finally, high-quality dense-view light field images with enhanced spatial resolution are rendered via EPI-based synthesis. The suggested methodology greatly outperforms state-of-the-art methods in joint spatial-angular super-resolution, as shown by extensive experimentation on public light field benchmarks and a recently gathered low-altitude light field dataset. It achieves the best performance in terms of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), while also producing visually superior results with sharp detail recovery and smooth, natural disparity transitions. Moreover, the method exhibits strong generalization capability and competitive performance in standalone light field spatial super-resolution (LFSSR) and light field angular super-resolution (LFASR) tasks.
- light field /
- light field super-resolution /
- light field Gaussian splatting /
- spatial-angular joint super-resolution /
- epipolar plane geometry

HTML全文

图 1 整体网络结构框

Figure 1. Overall framework of the proposed network

下载: 全尺寸图片幻灯片

图 2 光场几何编码模块和注意力模块

Figure 2. Light field geometry encoding module and attention modules

下载: 全尺寸图片幻灯片

图 3 低分辨率 EPI 网格上构建的 2D 高斯泼溅模型示意

Figure 3. Illustration of the 2D Gaussian splatting model constructed on a low-resolution EPI grid

下载: 全尺寸图片幻灯片

图 4 低空场景(建筑、桥梁、交通、道路)下的光场图像及其 EPI 图像

Figure 4. Light field images and their EPIs in low-altitude scenarios (e.g., buildings, bridges, traffic, and roads)

下载: 全尺寸图片幻灯片

图 5 空间(×2)-角度(2×2→5×5)同时超分辨率任务下不同模型生成的视觉效果对比

Figure 5. Visual comparisons of different models on the joint spatial(×2) and angular(2×2→5×5) super-resolution task

下载: 全尺寸图片幻灯片

图 6 低空场景光场图像的空间(×2)-角度(2×2→5×5)同时超分辨率重建视觉结果

Figure 6. Visual results of joint spatial (×2) and angular (2×2→5×5) super-resolution on low-altitude light field images task (the second scene in Fig.6)

下载: 全尺寸图片幻灯片

图 7 空间(×2)-角度(2×2→5×5)同时超分辨率任务下PSNR和SSIM值分布的可视化比对(图6中第2个场景)

Figure 7. Visual comparison of PSNR and SSIM distributions under joint spatial (×2) and angular (2×2→5×5) super-resolution

下载: 全尺寸图片幻灯片

表 1 本文模型与7种模型在空间(×2)-角度(2×2→5×5)同时超分辨率任务上的性能对比

Table 1. Performance comparisons of this paper with seven models on the joint spatial(×2) and angular(2×2→5×5) super-resolution task

LFASR模型	LFSSR模型	PSNR/dB，SSIM
LFASR模型	LFSSR模型	30Scenes	Occlusion	Reflective	HCInew	HCIold	EPFL
DistgASR^[16]	DistgSSR^[16]	41.93 / 0.992 0	38.36 / 0.985 4	38.94 / 0.977 7	34.74 / 0.961 7	41.56 / 0.990 5	32.90 / 0.969 5
DistgASR^[16]	EPIT^[25]	41.85 / 0.991 8	38.27 / 0.985 1	38.89 / 0.976 8	35.21 / 0.965 3	42.12 / 0.991 4	33.21 / 0.970 3
DistgASR^[16]	MLFSR^[14]	41.99 / 0.992 1	38.41 / 0.985 7	39.01 / 0.977 2	34.87 / 0.962 8	41.98 / 0.991 2	33.37 / 0.970 8
LF-EASR^[7]	DistgSSR^[16]	41.86 / 0.991 9	38.21 / 0.985 1	38.94 / 0.977 6	34.66 / 0.961 2	41.67 / 0.990 4	33.17 / 0.969 7
LF-EASR^[7]	EPIT^[25]	41.80 / 0.991 7	38.15 / 0.984 9	38.93 / 0.977 4	35.06 / 0.964 4	42.05 / 0.991 1	33.32 / 0.969 5
LF-EASR^[7]	MLFSR^[14]	41.77 / 0.991 7	38.12 / 0.984 9	38.91 / 0.977 2	34.67 / 0.961 3	41.68 / 0.990 7	33.28 / 0.970 0
LFSR^[19]	LFSR^[19]	41.89 / 0.991 9	38.30 / 0.985 3	38.86 / 0.977 2	34.46 / 0.959 9	41.72 / 0.990 7	32.98 / 0.969 2
本文模型	本文模型	42.15 / 0.992 3	38.64 / 0.986 2	39.22 / 0.978 2	35.35 / 0.965 3	42.81 / 0.992 5	33.67 / 0.971 9
注：指标最优和次优分别作加粗和下划线处理。

下载: 导出CSV

表 2 本文与7种模型在LFASR（2×2→7×7）任务上的PSNR和SSIM对比

Table 2. Performance comparisons on PSNR and SSIM of this paper with seven models in the LFASR task(2×2→7×7)

模型	发表年份	PSNR/dB，SSIM
模型	发表年份	HCInew	HCIold	30Scenes	Occlusion	Reflective
LF-EASR^[7]	2022	35.86 / 0.975	41.54 / 0.971	43.44 / 0.995	39.80 / 0.992	39.35 / 0.981
DistgASR^[16]	2022	34.70 / 0.974	42.18 / 0.978	43.67 / 0.995	39.46 / 0.991	39.11 / 0.978
HLFSR-ASR^[17]	2023	36.43 / 0.982	43.25 / 0.990	44.03 / 0.996	39.87 / 0.992	39.60 / 0.981
Liu等^[26]	2023	35.29 / 0.954	41.92 / 0.969	43.33 / 0.988	39.67 / 0.983	39.20 / 0.963
Li和Zhong^[27]	2024	35.21 / 0.975	42.36 / 0.979	43.95 / 0.996	39.68 / 0.991	39.36 / 0.981
Liu等^[28]	2024	35.41 / 0.954	42.02 / 0.969	43.38 / 0.995	39.71 / 0.991	39.31 / 0.982
ViewFormer^[29]	2025	35.37 / 0.977	41.93 / 0.975	43.82 / 0.995	39.86 / 0.992	39.57 / 0.981
本文		37.71 / 0.983	43.99 / 0.993	44.15 / 0.995	40.07 / 0.992	39.45 / 0.984
注：指标最优和次优分别作加粗和下划线处理。

下载: 导出CSV

表 3 本文与9种模型在LFSSR任务(放大2倍和4倍)上的PSNR和SSIM对比

Table 3. Performance comparisons on PSNR and SSIM of this paper with nine models in the LFSSR tasks(×2 and ×4)

模型	发表年份	PSNR/dB，SSIM
		EPFL		HCInew		HCIold		INRIA		STFgantry
		放大2倍	放大4倍	放大2倍	放大4倍	放大2倍	放大4倍	放大2倍	放大4倍	放大2倍	放大4倍
DPT^[30]	2022	34.49	28.94	37.36	31.20	44.30	37.41	36.41	30.96	39.43	31.15
DPT^[30]	2022	0.975 8	0.917 0	0.977 1	0.918 8	0.994 3	0.972 1	0.984 3	0.952 3	0.992 6	0.948 8
DistgSSR^[16]	2022	34.81	28.99	37.96	31.38	44.94	37.56	36.59	30.99	40.40	31.65
DistgSSR^[16]	2022	0.978 7	0.919 5	0.979 6	0.921 7	0.994 9	0.973 2	0.985 9	0.951 9	0.994 2	0.953 5
EPIT^[25]	2023	34.83	29.34	38.23	31.51	45.08	37.68	36.67	31.37	42.17	32.18
EPIT^[25]	2023	0.977 5	0.919 7	0.981 0	0.923 1	0.994 9	0.973 7	0.985 3	0.952 6	0.995 7	0.957 1
HLFSR-SSR^[17]	2023	35.31	29.20	38.32	31.57	44.98	37.78	37.06	31.24	40.85	31.64
HLFSR-SSR^[17]	2023	0.980 0	0.922 2	0.980 7	0.923 8	0.995 0	0.972 4	0.986 7	0.953 4	0.994 7	0.953 7
LF-DET^[31]	2023	35.26	29.47	38.31	31.56	44.99	37.84	36.95	31.39	41.76	32.14
LF-DET^[31]	2023	0.979 7	0.923 0	0.980 7	0.923 5	0.995 0	0.974 4	0.986 4	0.953 4	0.995 5	0.957 3
徐欣宜等^[32]	2023	34.49	28.70	37.47	31.13	44.25	37.21	36.37	30.73	38.89	30.61
徐欣宜等^[32]	2023	0.976 7	0.914 8	0.979 0	0.920 8	0.994 3	0.972 4	0.984 3	0.950 3	0.991 6	0.943 2
MLFSR^[14]	2024	35.22	29.28	38.14	31.56	44.90	37.83	36.92	31.24	40.98	32.03
MLFSR^[14]	2024	0.980 1	0.921 8	0.980 3	0.923 5	0.995 0	0.974 5	0.986 5	0.953 1	0.994 9	0.956 7
LF-TGUnet^[33]	2024	34.67	28.96	37.80	31.42	44.78	37.63	36.46	31.28	39.95	31.36
LF-TGUnet^[33]	2024	0.977 6	0.918 9	0.979 1	0.921 5	0.994 8	0.973 4	0.985 7	0.952 9	0.993 6	0.951 1
LF-GIANet^[34]	2025	34.56	29.14	38.01	31.40	44.80	37.60	36.45	31.25	40.45	31.30
LF-GIANet^[34]	2025	0.978 2	0.919 8	0.979 8	0.921 8	0.994 8	0.973 3	0.985 8	0.952 2	0.994 5	0.950 5
本文		35.70	29.47	38.49	31.61	45.31	37.86	37.28 /	31.45	41.45	31.72
本文		0.980 9	0.924 4	0.981 5	0.924 4	0.995 3	0.974 8	0.987 2	0.955 0	0.995 3	0.955 3
注：指标最优和次优分别作加粗和下划线处理。

下载: 导出CSV

表 4 低空场景光场图像的空间(×2)-角度(2×2→5×5)同时超分辨率性能对比

Table 4. Performance comparisons on the joint spatial (×2) and angular (2×2→5×5) super-resolution for low-altitude light field images

场景编号	PSNR/dB，SSIM
场景编号	DistgASR^[16] +DistgSSR^[16]	DistgASR^[16] +EPIT^[25]	DistgASR^[16] +MLFSR^[14]	LF-EASR^[7] +DistgSSR^[16]	LF-EASR^[7] +EPIT^[25]	LF-EASR^[7] + MLFSR^[14]	LFSR^[19]	本文模型
Image36	37.09/0.981 7	37.06/0.981 6	37.15/0.981 9	36.85/0.980 4	37.01/0.981 3	37.00/0.980 9	36.82/0.980 4	37.20/0.982 3
Image62	36.34/0.978 5	36.41/0.979 1	36.64/0.979 3	36.07/0.977 7	36.42/0.978 8	36.34/0.978 2	36.04/0.977 5	36.64/0.979 3
Image63	36.08/0.983 4	36.25/0.984 0	36.30/0.984 1	36.17/0.983 9	36.19/0.983 8	36.13/0.983 6	35.89/0.982 8	36.44/0.984 5
Image81	37.87/0.980 7	37.97/0.981 2	37.92/0.980 9	37.95/0.981 2	37.87/0.980 8	37.85/0.980 6	37.68/0.980 0	38.04/0.981 4
Image82	36.87/0.978 7	37.01/0.979 3	37.03/0.979 2	36.68/0.977 7	36.93/0.978 4	36.71/0.977 8	36.50/0.977 1	37.30/0.980 5
Image85	37.56/0.986 1	37.79/0.9865	37.74/0.9863	37.41/0.9857	37.76/0.9860	37.42/0.9855	37.10/0.9847	38.10/0.9872
Image86	39.56/0.978 9	39.57/0.9789	39.48/0.9786	39.59/0.9790	39.52/0.9787	39.55/0.9788	39.46/0.9785	39.64/0.9792
Image89	37.26/0.978 9	37.23/0.9790	37.33/0.9792	37.20/0.9787	37.26/0.9789	37.23/0.9787	37.03/0.9782	37.37/0.9795
Image98	37.21/0.983 7	37.41/0.9844	37.29/0.9839	37.06/0.9831	37.29/0.9838	37.11/0.9831	37.07/0.9832	37.36/0.9843
Image128	39.52/0.991 0	39.69/0.9920	39.56/0.9921	39.51/0.9919	39.69/0.9918	39.53/0.9916	39.14/0.9901	39.99/0.9928
Image187	37.23/0.9914	37.38/0.9924	37.21/0.9920	37.48/0.9921	37.25/0.9916	37.10/0.9914	36.80/0.9910	37.69/0.9926
Image264	35.68/0.9812	36.04/0.9826	36.19/0.9830	34.69/0.9775	36.02/0.9819	34.62/0.9763	35.34/0.9798	36.01/0.9829
注：指标最优和次优分别作加粗和下划线处理；从左到右8个模型的平均值分别为37.36/0.9829、37.48/0.9834、37.49/0.9834、37.22/0.9824、37.43/0.9830、37.22/0.9822、37.07/0.9820、37.65/0.9839。

下载: 导出CSV

表 5 LFASR和LFSSR消融实验

Table 5. Ablation study on the LFASR and LFSSR tasks

超分辨率任务类型	数据集	PSNR (dB)/SSIM
超分辨率任务类型	数据集	w/o 4DLFGS	本文方法
LFASR	HCInew	34.27/0.970 0	37.71/0.982 9
LFASR	HCIold	42.27/0.979 0	43.99/0.993 1
LFSSR	HCInew	38.32/0.981 0	38.49/0.981 5
LFSSR	HCIold	45.01/0.9950	45.31/0.9953

下载: 导出CSV

表 6 LFASSR任务消融实验

Table 6. Ablation study on the LFASSR task

交叉注意力模块	自注意力模块	4DLFGS	PSNR/dB，SSIM						参数量	浮点运算速度/ (10⁹·s⁻¹)
交叉注意力模块	自注意力模块	4DLFGS	HCInew	HCIold	EPFL	30Scenes	Occlusions	Reflective	参数量	浮点运算速度/ (10⁹·s⁻¹)
√			34.55/0.960 4	41.88/0.991 2	33.22/0.970 2	41.84/0.991 8	38.28/0.985 3	38.96/0.977 3	1.483×10⁷	76.12
	√		34.41/0.959 6	41.70/0.991 2	33.26/0.969 9	41.90/0.991 9	38.34/0.985 5	39.06/0.978 1	1.485×10⁷	76.30
√	√		35.12/0.963 9	42.40/0.992 0	33.51/0.971 1	41.96/0.992 0	38.45/0.985 7	39.12/0.978 6	1.638×10⁷	85.63
	√	√	35.09/0.9639	42.42/0.9920	33.51/0.9715	42.11/0.9922	38.53/0.9858	39.13/0.9780	1.525×10⁷	119.93
√	√	√	35.35/0.9653	42.81/0.9925	33.67/0.9719	42.15/0.9923	38.64/0.9862	39.22/0.9782	1.678×10⁷	129.26

下载: 导出CSV

参考文献(35)

[1]	Ng R, Levoy M, Brédif M, et al. Light field photography with a hand-held plenoptic camera[D]. Palo Alto: Stanford University, 2005: 5-11.
[2]	Lyu W Q, Sheng H, Ke W, et al. Advances in light field spatial super-resolution: a comprehensive literature survey[J]. IEEE Access, 2025, 13: 18470-18497.
[3]	Hu H, Guo M T, Li H D, et al. Revisiting spatio-angular trade-off in light field cameras and extended applications in super-resolution[J]. IEEE Transactions on Visualization and Computer Graphics, 2021, 27(6): 3019-3033.
[4]	Kalantari N K, Wang T C, Ramamoorthi R. Learning-based view synthesis for light field cameras[J]. ACM Transactions on Graphics, 2016, 35(6): 1-10.
[5]	Zhang S, Chang S, Shen Z Q, et al. Micro-lens image stack upsampling for densely-sampled light field reconstruction[J]. IEEE Transactions on Computational Imaging, 2021, 7: 799-811.
[6]	Jin J, Hou J H, Yuan H, et al. Learning light field angular super-resolution via a geometry-aware network[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 11141-11148.
[7]	Liu G S, Yue H J, Wu J M, et al. Efficient light field angular super-resolution with sub-aperture feature learning and macro-pixel upsampling[J]. IEEE Transactions on Multimedia, 2023, 25: 6588-6600.
[8]	Wang L Y, Ren L F, Wei X Y, et al. Light field angular super-resolution based on intrinsic and geometric information[J]. Knowledge-Based Systems, 2023, 270: 110553.
[9]	Mao Y F, Xiao Z Y, An P, et al. Deep sparse-to-dense inbetweening for multi-view light fields[J]. IEEE Transactions on Image Processing, 2025, 34: 6302-6317.
[10]	Yuan Y, Cao Z Q, Su L J. Light-field image superresolution using a combined deep CNN based on EPI[J]. IEEE Signal Processing Letters, 2018, 25(9): 1359-1363.
[11]	Wang Y L, Liu F, Zhang K B, et al. LFNet: a novel bidirectional recurrent convolutional neural network for light-field image super-resolution[J]. IEEE Transactions on Image Processing, 2018, 27(9): 4274-4286.
[12]	Zhang S, Lin Y F, Sheng H. Residual networks for light field image super-resolution[C]// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 11038-11047.
[13]	Liang Z Y, Wang Y Q, Wang L G, et al. Light field image super-resolution with transformers[J]. IEEE Signal Processing Letters, 2022, 29: 563-567.
[14]	Gao R S, Xiao Z Y, Xiong Z W. Mamba-based light field super-resolution withEfficient subspace scanning[C]//Proceedings of the Computer Vision–ACCV 2024. Berlin: Springer, 2024: 421-437.
[15]	Chen S R, Chen L, Wu D F, et al. Enhancing light field image super-resolution through Mamba-based spatial–angular correlation learning[J]. The Visual Computer, 2025, 41(15): 12649-12662.
[16]	Wang Y Q, Wang L G, Wu G C, et al. Disentangling light fields for super-resolution and disparity estimation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(1): 425-443.
[17]	Van Duong V, Huu T N, Yim J, et al. Light field image super-resolution network via joint spatial-angular and epipolar information[J]. IEEE Transactions on Computational Imaging, 2023, 9: 350-366.
[18]	Yoon Y, Jeon H G, Yoo D, et al. Learning a deep convolutional network for light-field image super-resolution[C]// Proceedings of the 2015 IEEE International Conference on Computer Vision Workshop. Piscataway: IEEE Press, 2016: 57-65.
[19]	Liu G S, Yue H J, Li K, et al. Adaptive pixel aggregation for joint spatial and angular super-resolution of light field images[J]. Information Fusion, 2024, 104: 102183.
[20]	Liu D Y, Li S Z, Mao Y F, et al. Learning implicit and detail-enhanced network for light field image spatial-angular super-resolution[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2026, 36(2): 1544-1557.
[21]	Wu T, Yuan Y J, Zhang L X, et al. Recent advances in 3D Gaussian splatting[J]. Computational Visual Media, 2024, 10(4): 613-642.
[22]	Hu J T, Xia B, Chen B, et al. GaussianSR: high fidelity 2D Gaussian splatting for arbitrary-scale image super-resolution[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2025, 39(4): 3554-3562.
[23]	Zhang H, Zhou W H, Lin L L, et al. Cascade residual learning based adaptive feature aggregation for light field super-resolution[J]. Pattern Recognition, 2025, 165: 111616.
[24]	Sheng H, Cong R X, Yang D, et al. UrbanLF: a comprehensive light field dataset for semantic segmentation of urban scenes[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(11): 7880-7893.
[25]	Liang Z Y, Wang Y Q, Wang L G, et al. Learning non-local spatial-angular correlation for light field image super-resolution[C]// Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2023: 12342-12352.
[26]	Liu D Y, Mao Y F, Zhou X F, et al. Learning a multilevel cooperative view reconstruction network for light field angular super-resolution[C]// Proceedings of the 2023 IEEE International Conference on Multimedia and Expo. Piscataway: IEEE Press, 2023: 1271-1276.
[27]	Li D C, Zhong R. Light field image angular super-resolution using edge features[C]//Proceedings of the 2024 6th International Conference on Robotics and Computer Vision. Piscataway: IEEE Press, 2024: 132-138.
[28]	Liu D Y, Mao Y F, Zuo Y F, et al. Light field angular super-resolution network based on convolutional transformer and deep deblurring[J]. IEEE Transactions on Computational Imaging, 2024, 10: 1736-1748.
[29]	Wang S Z, Lu Y, Xia W, et al. Light field angular super-resolution by view-specific queries[J]. The Visual Computer, 2025, 41(5): 3565-3580.
[30]	Wang S Z, Zhou T F, Lu Y, et al. Detail-preserving transformer for light field image super-resolution[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(3): 2522-2530.
[31]	Cong R X, Sheng H, Yang D, et al. Exploiting spatial and angular correlations with deep efficient transformers for light field image super-resolution[J]. IEEE Transactions on Multimedia, 2023, 26: 1421-1435.
[32]	徐欣宜, 邓慧萍, 向森, 等. 基于特征交互融合与注意力的光场图像超分辨率[J]. 激光与光电子学进展, 2023, 60(14): 1410017. Xu X Y, Deng H P, Xiang S, et al. Light field image super-resolution based on feature interaction fusion and attention mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1410017(in Chinese).
[33]	黄莉, 吕天琪, 武迎春, 等. 基于双路引导更新的光场图像超分网络[J]. 光电工程, 2024, 51(12): 240222. Huang L, Lv T Q, Wu Y C, et al. Two-way guided updating network for light field image super-resolution[J]. Opto-Electronic Engineering, 2024, 51(12): 240222(in Chinese).
[34]	Zhang W, Li H J, Ke W. LF-GIANet: cascaded global-view information adaptation-guided network for light field image super-resolution[J]. Multimedia Systems, 2025, 31(2): 73.
[35]	Chen Y B, Liu S F, Wang X L. Learning continuous image representation with local implicit image function[C]// Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2021: 8624-8634.