基于多重注意力和感知加权学习的单图像高动态范围重建

柯德涨; 陈晔曜; 徐海勇; 金充充; 蒋刚毅

doi:10.12263/DZXB.20241055

您当前的位置：

首页 >

文章列表页 >

基于多重注意力和感知加权学习的单图像高动态范围重建

学术论文 | 更新时间：2025-10-16

- 基于多重注意力和感知加权学习的单图像高动态范围重建
- Single-Image High Dynamic Range Reconstruction Based on Multi-Attention and Perceptual Weighted Learning
- 电子学报 2025年53卷第6期页码：2063-2078
- 作者机构：
  
  1.宁波大学信息科学与工程学院，浙江宁波 315211
  2.宁波大学数学与统计学院，浙江宁波 315211
  3.宁波大学科学技术学院，浙江宁波 315300
- 作者简介：
  
  [ "柯德涨男，2000年10月生，浙江瑞安人.宁波大学信息科学与工程学院硕士研究生.主要研究方向为高动态范围成像、光场图像处理等. E-mail: 15869640695@163.com" ]
  [ "陈晔曜男，1994年10月生，浙江嘉兴人.宁波大学信息科学与工程学院讲师.主要研究方向为高动态范围成像、光场成像、视频处理等.中国电子学会会员编号：E190158389M. E-mail: chenyeyao@nbu.edu.cn" ]
  [ "徐海勇男，1981年12月生，浙江金华人.宁波大学数学与统计学院教授、博士生导师.主要研究方向为水下图像增强、视觉感知等.中国电子学会会员编号：E190013185M. E-mail: xuhaiyong@nbu.edu.cn" ]
  [ "金充充女，1994年10月生，浙江绍兴人.宁波大学科学技术学院讲师.主要研究方向为视频图像质量评价、视觉感知等. E-mail: jinchongchong@nbu.edu.cn" ]
  [ "蒋刚毅男，1964年8月生，浙江绍兴人.宁波大学信息科学与工程学院教授、博士生导师.主要研究方向为多媒体信号处理、图像处理与视频压缩、计算成像与视觉感知等. E-mail: jianggangyi@nbu.edu.cn" ]
- 基金信息：
  
  国家自然科学基金(62401301;62271276;62171243);浙江省自然科学基金(LQ24F010002)
- DOI：10.12263/DZXB.20241055
  中图分类号： TP391;
- 收稿：2024-11-23，
  
  修回：2025-03-13，
  
  纸质出版：2025-06-25
- 稿件说明：
移动端阅览
柯德涨, 陈晔曜, 徐海勇, 等. 基于多重注意力和感知加权学习的单图像高动态范围重建[J]. 电子学报, 2025, 53(06): 2063-2078.

KE De-zhang, CHEN Ye-yao, XU Hai-yong, et al. Single-Image High Dynamic Range Reconstruction Based on Multi-Attention and Perceptual Weighted Learning[J]. Acta Electronica Sinica, 2025, 53(06): 2063-2078.
柯德涨, 陈晔曜, 徐海勇, 等. 基于多重注意力和感知加权学习的单图像高动态范围重建[J]. 电子学报, 2025, 53(06): 2063-2078. DOI：10.12263/DZXB.20241055

KE De-zhang, CHEN Ye-yao, XU Hai-yong, et al. Single-Image High Dynamic Range Reconstruction Based on Multi-Attention and Perceptual Weighted Learning[J]. Acta Electronica Sinica, 2025, 53(06): 2063-2078. DOI：10.12263/DZXB.20241055

摘要

单图像高动态范围（High Dynamic Range，HDR）重建能够避免多曝光HDR成像可能造成的鬼影伪像，正受到广泛研究.然而，现有方法由于缺乏对重要信息的关注，尚不能很好地恢复曝光不良区域的细节信息.为解决该问题，本文提出了一种基于多重注意力和感知加权学习的单图像HDR重建方法，旨在从单幅低动态范围图像中推断出高保真的HDR图像.具体而言，考虑到恢复曝光不良区域需参考其他区域的补偿信息，本文设计了具有全局-局部感受野的多重注意力视觉Transformer （Multi-Attention Vision Transformer，MA-ViT），其将深度可分离卷积和注意力机制相结合，从而实现更有效的全局和局部特征提取与交互.此外，还提出了一种损失感知加权图以引导网络聚焦曝光不良区域，进一步提升HDR重建质量.本文在多个基准数据集上构建了全面的对比实验，结果表明所提出方法相较于目前最先进的方法在平均峰值信噪比（Peak Signal to Noise Ratio，PSNR）上提高了0.23 dB，同时生成了具有更高视觉质量的HDR重建结果.

Abstract

Single-image high dynamic range (HDR) reconstruction can avoid ghosting artifacts that may be caused by multi-exposure HDR imaging

and is receiving widespread research. However

existing methods still struggle to effectively restore detail information in poorly exposed regions due to a lack of focus on critical information. To address this issue

this paper proposes a single-image HDR reconstruction method based on multi-attention and perceptual weighted learning

which aims to infer a high-fidelity HDR image from a single low dynamic range (LDR) image. Specifically

considering that the restoration of poorly exposed regions requires reference to compensation information from other regions

a multi-attention vision transformer (MA-ViT) with global-local receptive fields is designed. It combines depthwise separable convolution and attention mechanisms to achieve more effective global and local feature extraction and interaction. In addition

a loss aware weighted map is proposed to guide the network to focus on poorly exposed regions

further enhancing the quality of HDR reconstruction. Comprehensive comparative experiments are conducted on multiple benchmark datasets

and the results show that the proposed method improves the average peak signal to noise ratio (PSNR) by 0.23 dB compared to the state-of-the-art method

while generating HDR reconstruction results with higher visual quality.

关键词

Keywords

references

LI H F , YANG Z M , ZHANG Y F , et al . Single-image HDR reconstruction assisted ghost suppression and detail preservation network for multi-exposure HDR imaging [J ] . IEEE Transactions on Computational Imaging , 2024 , 10 : 429 - 445 .

张雨童 , 邓欣 , 徐迈 . 动态场景下深度自监督多曝光图像融合方法 [J ] . 电子学报 , 2024 , 52 ( 1 ): 264 - 273 .

ZHANG Y T , DENG X , XU M . Deep self-supervised multi-exposure image fusion for dynamic scenes [J ] . Acta Electronica Sinica , 2024 , 52 ( 1 ): 264 - 273 . (in Chinese)

LIU S Z , ZHANG X D , SUN L C , et al . Joint HDR denoising and fusion: A real-world mobile HDR image dataset [C ] // 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2023 : 13966 - 13975 .

印佳丽 , 韩津 , 陈斌 , 等 . 基于一致性感知特征融合的高动态范围成像方法 [J ] . 计算机学报 , 2024 , 47 ( 10 ): 2352 - 2367 .

YIN J L , HAN J , CHEN B , et al . High dynamic range imaging based on coherence-aware feature aggregation [J ] . Chinese Journal of Computers , 2024 , 47 ( 10 ): 2352 - 2367 . (in Chinese)

HU T , YAN Q S , QI Y K , et al . Generating content for HDR deghosting from frequency view [C ] // 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2024 : 25732 - 25741 .

王仙峰 , 刘世本 , 田建东 , 等 . 基于双重注意力网络的高动态范围图像重建 [J ] . 激光与光电子学进展 , 2024 , 61 ( 12 ): 1237005 .

WANG X F , LIU S B , TIAN J D , et al . High dynamic range image reconstruction based on dual-attention network [J ] . Laser Optoelectronics Progress , 2024 , 61 ( 12 ): 1237005 . (in Chinese)

郭露露 , 易红伟 . 基于局域加权叠加的高动态范围图像融合算法 [J ] . 光子学报 , 2022 , 51 ( 11 ): 344 - 353 .

GUO L L , YI H W . High dynamic range image fusion algorithm based on local weighted superposition [J ] . Acta Photonica Sinica , 2022 , 51 ( 11 ): 344 - 353 . (in Chinese)

LANDIS H . Production-ready global illumination [J ] . Siggraph Course Notes , 2002 , 10 : 87 - 101

HUO Y Q , YANG F , DONG L , et al . Physiological inverse tone mapping based on retina response [J ] . The Visual Computer , 2014 , 30 ( 5 ): 507 - 517 .

EILERTSEN G , KRONANDER J , DENES G , et al . HDR image reconstruction from a single exposure using deep CNNs [J ] . ACM Transactions on Graphics , 2017 , 36 ( 6 ): 1 - 15 .

MARNERIDES D , BASHFORD-ROGERS T , HATCHETT J , et al . ExpandNet: A deep convolutional neural network for high dynamic range expansion from low dynamic range content [J ] . Computer Graphics Forum , 2018 , 37 ( 2 ): 37 - 49 .

KHAN Z , KHANNA M , RAMAN S . FHDR: HDR image reconstruction from a single LDR image using feedback network [C ] // 2019 IEEE Global Conference on Signal and Information Processing (GlobalSIP) . Piscataway : IEEE , 2019 : 1 - 5 .

WANG H , YE M , ZHU X T , et al . KUNet: Imaging knowledge-inspired single HDR image reconstruction [C ] // Proceedings of the 31st International Joint Conference on Artificial Intelligence . Virtual Event : IJCAI Organization , 2022 : 1408 - 1414 .

ZHANG W , JIANG G Y , CHEN Y Y , et al . Multi-stage coarse-to-fine progressive enhancement network for single-image HDR reconstruction [J ] . Displays , 2024 , 84 : 102791 .

AKYÜZ A , FLEMING R , RIECKE B , et al . Do HDR displays support LDR content?: A psychophysical evaluation [J ] . ACM Transactions on Graphics , 2007 , 26 ( 3 ): 1 - 8 .

MASIA B , AGUSTIN S , FLEMING R W , et al . Evaluation of reverse tone mapping through varying exposure conditions [J ] . ACM Transactions on Graphics , 2009 , 28 ( 5 ): 1 - 8 .

MEYLAN L , DALY S , SÜSSTRUNK S . The reproduction of specular highlights on high dynamic range displays [J ] . Color and Imaging Conference , 2006 , 14 ( 1 ): 333 - 338 .

KUO P H , TANG C S , CHIEN S Y . Content-adaptive inverse tone mapping [C ] // 2012 Visual Communications and Image Processing . Piscataway : IEEE , 2012 : 1 - 6 .

KOVALESKI R P , OLIVEIRA M M . High-quality reverse tone mapping for a wide range of exposures [C ] // 2014 27th SIBGRAPI Conference on Graphics, Patterns and Images . Piscataway : IEEE , 2014 : 49 - 56 .

WANG T H , CHIU C W , WU W C , et al . Pseudo-multiple-exposure-based tone fusion with local region adjustment [J ] . IEEE Transactions on Multimedia , 2015 , 17 ( 4 ): 470 - 484 .

SANTOS M S , REN T I , KALANTARI N K . Single image HDR reconstruction using a CNN with masked features and perceptual loss [J ] . ACM Transactions on Graphics , 2020 , 39 ( 4 ): 1 - 10 .

CHEN X Y , LIU Y H , ZHANG Z W , et al . HDRUNet: Single image HDR reconstruction with denoising and dequantization [C ] // 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) . Piscataway : IEEE , 2021 : 354 - 363 .

贝悦 , 王琦 , 程志鹏 , 等 . 基于条件生成对抗网络的HDR图像生成方法 [J ] . 北京航空航天大学学报 , 2022 , 48 ( 1 ): 45 - 52 .

BEI Y , WANG Q , CHENG Z P , et al . HDR image generation method based on conditional generative adversarial network [J ] . Journal of Beijing University of Aeronautics and Astronautics , 2022 , 48 ( 1 ): 45 - 52 . (in Chinese)

MORIWAKI K , YOSHIHASHI R , KAWAKAMI R , et al . Hybrid loss for learning single-image-based HDR reconstruction [EB/OL ] . ( 2018-12-18 )[ 2024-11-23 ] . https://arxiv.org/abs/1812.07134v1 https://arxiv.org/abs/1812.07134v1 .

LIU Y L , LAI W S , CHEN Y S , et al . Single-image HDR reconstruction by learning to reverse the camera pipeline [C ] // 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2020 : 1651 - 1660 .

SHE L , YE M , LI S , et al . Single-image HDR reconstruction by dual learning the camera imaging process [J ] . Engineering Applications of Artificial Intelligence , 2023 , 120 : 105947 .

ENDO Y , KANAMORI Y , MITANI J . Deep reverse tone mapping [J ] . ACM Transactions on Graphics , 2017 , 36 ( 6 ): 1 - 10 .

LEE S , AN G H , KANG S J . Deep recursive HDRI: Inverse tone mapping using generative adversarial networks [C ] // Computer Vision-ECCV 2018 . Cham : Springer International Publishing , 2018 : 613 - 628 .

LEE S , JO S Y , AN G H , et al . Learning to generate multi-exposure stacks with cycle consistency for high dynamic range imaging [J ] . IEEE Transactions on Multimedia , 2020 , 23 : 2561 - 2574 .

JO S Y , LEE S , AHN N , et al . Deep arbitrary HDRI: Inverse tone mapping with controllable exposure changes [J ] . IEEE Transactions on Multimedia , 2021 , 24 : 2713 - 2726 .

ZHANG N , YE Y Y , ZHAO Y , et al . Revisiting the stack-based inverse tone mapping [C ] // 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2023 : 9162 - 9171 .

CHEN S K , YEN H L , LIU Y L , et al . Learning continuous exposure value representations for single-image HDR reconstruction [C ] // 2023 IEEE/CVF International Conference on Computer Vision (ICCV) . Piscataway : IEEE , 2023 : 12944 - 12954 .

陈羽中 , 陈友昆 , 林闽沪 , 等 . 基于边缘辅助和多尺度Transformer的无参考屏幕内容图像质量评估 [J ] . 电子学报 , 2024 , 52 ( 7 ): 2242 - 2256 .

CHEN Y Z , CHEN Y K , LIN M H , et al . No-reference screen content image quality assessment based on edge assistance and multi-scale transformer [J ] . Acta Electronica Sinica , 2024 , 52 ( 7 ): 2242 - 2256 . (in Chinese)

雷涛 , 翟钰杰 , 许叶彤 , 等 . 基于边缘引导和动态可变形Transformer的遥感图像变化检测 [J ] . 电子学报 , 2024 , 52 ( 1 ): 107 - 117 .

LEI T , ZHAI Y J , XU Y T , et al . Edge guided and dynamically deformable transformer network for remote sensing images change detection [J ] . Acta Electronica Sinica , 2024 , 52 ( 1 ): 107 - 117 . (in Chinese)

LIANG J Y , CAO J Z , SUN G L , et al . SwinIR: Image restoration using swin transformer [C ] // 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) . Piscataway : IEEE , 2021 : 1833 - 1844 .

HU J , SHEN L , ALBANIE S , et al . Squeeze-and-excitation networks [J ] . IEEE Transactions on Pattern Analysis and Machine Intelligence , 2020 , 42 ( 8 ): 2011 - 2023 .

SIMONYAN K , ZISSERMAN A . Very deep convolutional networks for large-scale image recognition [C ] // 2015 International Conference on Learning Representations . San Diego : ICLR , 2015 : 1 - 14 .

JANG H , BANG K , JANG J , et al . Dynamic range expansion using cumulative histogram learning for high dynamic range image generation [J ] . IEEE Access , 2020 , 8 : 38554 - 38567 .

LEE S , AN G H , KANG S J . Deep chain HDRI: Reconstructing a high dynamic range image from a single low dynamic range image [J ] . IEEE Access , 2018 , 6 : 49913 - 49924 .

KALANTARI N K , RAMAMOORTHI R . Deep high dynamic range imaging of dynamic scenes [J ] . ACM Transactions on Graphics , 2017 , 36 ( 4 ): 1 - 12 .

NEMOTO H , KORSHUNOV P , HANHART P , et al . Visual attention in LDR and HDR images [C ] // 2015 9th International Workshop on Video Processing and Quality Metrics for Consumer Electronics . Scottsdale : VPQM , 2015 : 1 - 6 .

HANJI P , MANTIUK R , EILERTSEN G , et al . Comparison of single image HDR reconstruction methods: The caveats of quality assessment [C ] // Special Interest Group on Computer Graphics and Interactive Techniques Conference Proceedings . New York : ACM , 2022 : 1 - 8 .

MANTIUK R K , AZIMI M . PU21: A novel perceptually uniform encoding for adapting existing quality metrics for HDR [C ] // 2021 Picture Coding Symposium (PCS) . Piscataway : IEEE , 2021 : 1 - 5 .

EILERTSEN G , HAJISHARIF S , HANJI P , et al . How to cheat with metrics in single-image HDR reconstruction [C ] // 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) . Piscataway : IEEE , 2021 : 3981 - 3990 .

MANTIUK R K , HAMMOU D , HANJI P . HDR-VDP-3: A multi-metric for predicting image differences, quality and contrast distortions in high dynamic range and regular content [EB/OL ] . ( 2023-08-26 )[ 2024-11-23 ] . https://arxiv.org/abs/2304.13625v1 https://arxiv.org/abs/2304.13625v1 .

REINHARD E , STARK M , SHIRLEY P , et al . Photographic tone reproduction for digital images [J ] . ACM Transactions on Graphics , 2002 , 21 ( 3 ): 267 - 276 .

MANTIUK R , DALY S , KEROFSKY L . Display adaptive tone mapping [J ] . ACM Transactions on Graphics , 2008 , 27 ( 3 ): 1 - 10 .

DRAGO F , MYSZKOWSKI K , ANNEN T , et al . Adaptive logarithmic mapping for displaying high contrast scenes [J ] . Computer Graphics Forum , 2003 , 22 ( 3 ): 419 - 426 .

ZHANG R , ISOLA P , EFROS A A , et al . The unreasonable effectiveness of deep features as a perceptual metric [EB/OL ] . ( 2018-08-10 )[ 2024-11-23 ] . https://arxiv.org/abs/1801.03924v2 https://arxiv.org/abs/1801.03924v2 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于EIMYOLO的高分遥感图像目标检测

基于感知和记忆的视频动态质量评价

复杂噪声环境下基于轻量化模型的车内交互语音增强和识别方法

MoGE：基于图上下文增强的多任务推荐算法