[1]孙立辉,赵宜友.基于金字塔可形变卷积的多分支视频超分模型[J].常州大学学报(自然科学版),2025,37(01):28-36.[doi:10.3969/j.issn.2095-0411.2025.01.004]
　SUN Lihui,ZHAO Yiyou.Multi-branch video super-resolution model based on pyramid deformable convolution[J].Journal of Changzhou University(Natural Science Edition),2025,37(01):28-36.[doi:10.3969/j.issn.2095-0411.2025.01.004]
点击复制

基于金字塔可形变卷积的多分支视频超分模型()
分享到：

常州大学学报(自然科学版)[ISSN:2095-0411/CN:32-1822/N]

卷:
第37卷
期数:
2025年01期
页码:
28-36
栏目:
计算机与信息工程
出版日期:
2025-01-22

文章信息/Info

Title:
Multi-branch video super-resolution model based on pyramid deformable convolution
文章编号:
2095-0411(2025)01-0028-09
作者:
孙立辉1; 2; 赵宜友1
1.河北经贸大学 信息技术学院, 河北 石家庄 050051; 2.河北省跨境电商技术创新中心, 河北 石家庄 050051
Author(s):
SUN Lihui1; 2;  ZHAO Yiyou1
1.School of Information Technology, Hebei University of Economics and Business,Shijiazhuang 050051, China; 2.Hebei Cross-Border E-Commerce Technology Innovation, Shijiazhuang 050051, China
关键词:
超分辨率重建;  BasicVSR;  帧对齐;  可形变卷积;  级联融合
Keywords:
super-resolution reconstruction;  BasicVSR;  frame alignment;  deformable convolution;  cascade fusion
分类号:
TP 391
DOI:
10.3969/j.issn.2095-0411.2025.01.004
文献标志码:
A
摘要:
为利用帧间的空时相关性特点,提升红外视频超分辨率重建效果,提出了一种改进BasicVSR的超分辨率重建方法。首先,使用金字塔可形变对齐代替BasicVSR中使用的光流法进行帧对齐,将参考帧和相邻帧当作输入,使用可形变卷积对帧间的偏移量进行测量,使不同帧进行信息上的叠加,最大限度得到图像中的细节特征。其次,在上采样时,将参考图像与经过融合后的图像进行级联,通过浅层特征与深层特征的融合,增强特征表达能力。文章设计的模型具有轻量、运行时间短、重建图像主观视觉效果好等优点,且峰值信噪比(PSNR)与结构相似度(SSIM)以及模型运行时间等客观评价指标得到了改进。本文所提模型EbasicVSR比相关模型运行时间平均提升了19 s,信噪比提升了0.14 dB以上,结构相似度提升了2.9%以上,实验结果表明,相比于原BasicVSR模型,本文提出的模型取得了更好的重建效果。
Abstract:
In order to make use of the spatial and temporal correlation between frames to improve the performance of video super-resolution reconstruction, an improved BasicVSR super-resolution reconstruction method was proposed. Firstly, using deformable alignment to replace the optical flow method used in BasicVSR for frame alignment, taking the reference frame and adjacent frames as inputs, and using deformable convolution to measure the offset between frames, so that different frames can be superimposed on information to maximize the details of the image. Secondly, during the up-sampling, the reference image is cascaded with the fused image, and the feature expression ability was enhanced through the fusion of shallow features and deep features. The model designed in this paper has the advantages of light weight, short running time, good subjective visual effect of reconstructed images. In addition, objective evaluation indicators such as PSNR, SSIM and model running time have been improved. The EbasicVSR model proposed in this paper has an average improvement of 19 s in running time, more than 0.14 dB in signal-to-noise ratio and more than 2.9% in structural similarity compared with the relevant models. The experimental results show that the model proposed in this paper achieves better reconstruction results, compared with the original BasicVSR model.

参考文献/References:

[1] 伍利兵, 陈朝方, 林国生, 等. 红外热成像技术在安全检测领域的应用现状[J]. 品牌与标准化, 2021(4): 97-100.
[2] 李苗苗. 红外图像数字细节增强关键技术研究[D]. 南京: 南京邮电大学, 2019.
[3] 冷佳旭, 王佳, 莫梦竟成, 等. 基于深度学习的视频超分辨率重构进展综述[J]. 计算机科学, 2022, 49(2): 123-133.
[4] WOODCOCK J, BOCA P. ABZ2008 VSR-net workshop[C]//International Conference on Abstract State Machines. Heidelberg: Springer, 2008: 378-379.
[5] SHI W Z, CABALLERO J, HUSZÁR F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas: IEEE, 2016: 1874-1883.
[6] WANG X T, CHAN K C K, YU K, et al. EDVR: video restoration with enhanced deformable convolutional networks[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops(CVPRW). Long Beach: IEEE, 2019: 1954-1963.
[7] CHAN K C K, WANG X T, YU K, et al. BasicVSR: the search for essential components in video super-resolution and beyond[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Nashville: IEEE, 2021: 4945-4954.
[8] 吴昊月, 张惊雷, 赵俊亚. 基于导向型级联可形变卷积的电气设备红外图像检测[J]. 计算机应用与软件, 2023, 40(4): 204-210.
[9] 樊凌雁, 彭义辉, 王洪波.基于金字塔结构超分辨率网络的图像拼接方法: 202211547399.6[P]. 2022-12-05[2023-10-09].
[10] 薛培, 薛国新, 张亚洲, 等. 一种基于小波系数加权平均的Retinex图像增强算法[J]. 常州大学学报(自然科学版), 2010, 22(4): 39-42.
[11] NAH S, BAIK S, HONG S, et al. NTIRE 2019 challenge on video deblurring and super-resolution: dataset and study[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops(CVPRW). Long Beach: IEEE, 2019: 1996-2005.
[12] 郑德品, 沈海斌, 赵武峰. 一种抗模糊失真的结构相似度图像质量评价法[J]. 机电工程, 2007, 24(10): 82-84.

备注/Memo

备注/Memo:
收稿日期: 2024-06-15。
基金项目: 河北省重点研发计划“军民科技协同创新专项”资助项目(20350801D)。
作者简介: 孙立辉(1971—), 男, 河北石家庄人, 博士, 教授。E-mail: Sun-lh@163.com
更新日期/Last Update: 1900-01-01