梯度引导的JPEG压缩图像超分辨率重建

曹坪; 林树冉; 张淳杰; 郑晓龙; 赵耀

doi:10.16383/j.aas.c240517

梯度引导的JPEG压缩图像超分辨率重建

doi: 10.16383/j.aas.c240517 cstr: 32138.14.j.aas.c240517

曹坪^{1, 2,},
林树冉^{1, 2,},
张淳杰^{1, 2,},
郑晓龙^{3, 4, 5,},
赵耀^{1, 2,}

1.
北京交通大学计算机科学与技术学院信息科学研究所北京 100044
2.
北京交通大学视觉智能交叉创新教育部国际合作联合实验室北京 100044
3.
中国科学院自动化研究所多模态人工智能系统全国重点实验室北京 100190
4.
中国科学院自动化研究所复杂系统管理与控制国家重点实验室北京 100190
5.
中国科学院大学北京 101408

基金项目: 国家自然科学基金(62476021, 72225011, 72434005, 62072026), 多模态人工智能系统全国重点实验室开放课题基金(MAIS2024106)

详细信息

作者简介:
曹坪：北京交通大学计算机科学与技术学院信息科学研究所博士研究生. 主要研究方向为目标检测, 语义分割和低质量视觉重建. E-mail: pingcao@bjtu.edu.cn

林树冉：北京交通大学计算机科学与技术学院信息科学研究所硕士研究生. 主要研究方向为图像超分辨率重建. E-mail: 21120292@bjtu.edu.cn

张淳杰：北京交通大学计算机科学与技术学院信息科学研究所教授. 主要研究方向为图像处理与理解, 计算机视觉和多媒体数据处理与分析. 本文通讯作者. E-mail: cjzhang@bjtu.edu.cn

郑晓龙：中国科学院自动化研究所研究员. 主要研究方向为大数据与社会计算, 多模态数据感知与理解. E-mail: xiaolong.zheng@ia.ac.cn

赵耀：北京交通大学计算机科学与技术学院信息科学研究所教授. 主要研究方向为图像/视频编码, 数字水印和取证, 视频分析和理解. E-mail: yzhao@bjtu.edu.cn

计量
- 文章访问数: 79
- HTML全文浏览量: 50
- 被引次数: 0
出版历程
- 收稿日期: 2024-07-22
- 录用日期: 2025-04-22
- 网络出版日期: 2025-05-14

Gradient-guided Super-resolution Reconstruction for JPEG-compressed Images

CAO Ping^{1, 2
,},
LIN Shu-Ran^{1, 2
,},
ZHANG Chun-Jie^{1, 2
,},
ZHENG Xiao-Long^{3, 4, 5
,},
ZHAO Yao^{1, 2
,}

1.
Institute of Information Science, School of Computer Science and Technology, Beijing Jiaotong University, Beijing 100044
2.
Visual Intelligence + X International Cooperation Joint Laboratory of Ministry of Education, Beijing Jiaotong University, Beijing 100044
3.
State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190
4.
State Key Laboratory for Management and Control of Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190
5.
University of Chinese Academy of Sciences, Beijing 101408

Funds: Supported by National Natural Science Foundation of China (62476021, 72225011, 72434005, 62072026) and Open Projects Program of State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS2024106)

More Information

Author Bio:
CAO Ping Ph. D. candidate at the Institute of Information Science, School of Computer Science and Technology, Beijing Jiaotong University. Her research interest covers object detection, semantic segmentation and low-quality visual reconstruction

LIN Shu-Ran Master student at the Institute of Information Science, School of Computer Science and Technology, Beijing Jiaotong University. His main research interest is image super-resolution reconstruction

ZHANG Chun-Jie Professor at the Institute of Information Science, School of Computer Science and Technology, Beijing Jiaotong University. His research interest covers image processing and understanding, computer vision and multimedia data processing and analysis. Corresponding author of this paper

ZHENG Xiao-Long Professor at the Institute of Automation, Chinese Academy of Sciences. His research interest covers big data and social computing, multimodal data perception and understanding

ZHAO Yao Professor at the Institute of Information Science, School of Computer Science and Technology, Beijing Jiaotong University. His research interest covers image/video coding, digital watermarking and forensics, video analysis and understanding

摘要

摘要: 在真实场景中, 图像往往同时遭受低分辨率、压缩失真及噪声等多种退化因素影响. 现有方法通常聚焦于单一退化类型, 难以应对复杂的复合退化情况. 为解决真实场景中普遍存在的低分辨率与JPEG压缩伪影复合退化问题, 提出一种梯度引导的联合JPEG压缩伪影去除和超分辨率重建网络. 该网络以超分辨率分支为主导, 融合JPEG压缩伪影去除分支与梯度引导分支的非对称特征, 实现了高质量图像重建. JPEG压缩伪影去除分支专注于压缩伪影抑制, 缓解了主导分支的重建负担. 梯度引导分支则精准估计图像梯度, 引导主导分支恢复更多细节与纹理. 实验结果表明, 该方法提升了低分辨率JPEG压缩图像的重建质量.
- JPEG压缩 /
- 超分辨率 /
- 图像重建 /
- 梯度先验
Abstract: In real-world scenarios, images are often affected by multiple degradation factors simultaneously, such as low resolution, compression distortions, and noise. Existing methods typically focus on addressing a single type of degradation, making them less effective when dealing with complex compound degradations. To tackle the commonly encountered compound degradation issue of low resolution and JPEG compression artifacts in real-world scenarios, we propose a gradient-guided joint JPEG compression artifact removal and super-resolution reconstruction network. The proposed network adopts the super-resolution branch as the leading branch, which asymmetrically integrates features from the JPEG compression artifact removal and gradient-guided branches to achieve high-quality image reconstruction. The JPEG compression artifact removal branch focuses on suppressing compression artifacts, thereby alleviating the reconstruction burden on the leading branch. The gradient-guided branch accurately estimates image gradients to guide the leading branch in restoring fine details and textures. Experimental results demonstrate that the proposed method improves the reconstruction quality of low-resolution JPEG-compressed images.
- JPEG compression /
- super-resolution /
- image reconstruction /
- gradient prior

HTML全文

图 1 JPEG压缩图像示例

Fig. 1 Examples of JPEG-compressed images

下载: 全尺寸图片幻灯片

图 2 JCARSR的网络结构

Fig. 2 Network architecture of JCARSR

下载: 全尺寸图片幻灯片

图 3 多尺度无非线性激活模块

Fig. 3 Multi-scale nonlinear activation free block

下载: 全尺寸图片幻灯片

图 4 混合注意力模块

Fig. 4 Mix attention module

下载: 全尺寸图片幻灯片

图 5 多尺度交叉卷积组

Fig. 5 Muti-scale cross convolutional group

下载: 全尺寸图片幻灯片

图 6 不同模型的性能与运行时间对比

Fig. 6 Comparison of performance and runtime across different models.

下载: 全尺寸图片幻灯片

图 7 JPEG压缩图像超分辨率视觉对比

Fig. 7 Visual comparison of JPEG-compressed image super-resolution

下载: 全尺寸图片幻灯片

图 8 两阶段JPEG压缩图像超分辨率视觉对比

Fig. 8 Visual comparison of two-stage JPEG-compressed image super-resolution

下载: 全尺寸图片幻灯片

图 9 在真实数据集上的超分辨率视觉对比

Fig. 9 Visual comparison of super-resolution on real-world datasets

下载: 全尺寸图片幻灯片

图 10 梯度引导视觉分析

Fig. 10 Visual analysis of gradient guidance.

下载: 全尺寸图片幻灯片

表 1 JCARSR与经典图像超分辨率方法的定量比较

Table 1 Quantitative comparison of JCARSR with classical image super-resolution methods

对比方法	尺度因子	模型参数(M)	Set5	Set14	Urban100	B100	Manga109
对比方法	尺度因子	模型参数(M)	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
EDSR^[27]	2	40.73	28.97/0.8235	26.85/0.7136	25.07/0.7368	26.14/0.6623	28.04/0.8560
RDN^[28]	2	22.12	28.92/0.8227	26.82/0.7126	24.97/0.7335	26.11/0.6613	27.98/0.8548
RCAN^[29]	2	15.44	29.00/0.823 6	26.83/0.7125	25.08/0.7355	26.13/0.6610	28.02/0.8550
MDCN^[30]	2	13.12	28.91/0.8220	26.80/0.7120	24.92/0.7311	26.11/0.6607	27.91/0.8530
SwinIR^[46]	2	11.75	28.95/0.8231	26.87/0.714 1	25.11/0.737 7	26.13/0.6622	28.03/0.856 5
DRCT ^[55]	2	13.34	28.86/0.8207	26.76/0.7103	24.87/0.7275	26.08/0.6597	27.78/0.8510
HMA^[56]	2	66.49	28.79/0.8183	26.71/0.7077	24.73/0.7214	26.04/0.6575	27.64/0.8471
JCARSR	2	10.10	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609
EDSR^[27]	4	43.09	25.03/0.7118	23.91/0.6005	21.87/0.5891	23.86/0.5588	23.26/0.7270
RDN^[28]	4	22.27	25.04/0.7128	23.89/0.6012	21.84/0.5881	23.84/0.5593	23.25/0.7276
RCAN^[29]	4	15.59	25.07/0.715 5	23.90/0.6010	21.93/0.5910	23.85/0.5592	23.26/0.7276
MDCN^[30]	4	13.71	25.02/0.7128	23.84/0.5995	21.80/0.5885	23.82/0.5585	23.20/0.7251
SwinIR^[46]	4	11.90	25.03/0.7145	23.93/0.602 2	21.94/0.592 8	23.86/0.560 0	23.29/0.730 7
DRCT ^[55]	4	13.48	24.95/0.7111	23.86/0.5988	21.79/0.5842	23.82/0.5580	23.16/0.7244
HMA^[56]	4	66.63	24.99/0.7120	23.87/0.5984	21.89/0.5881	23.83/0.5581	23.18/0.7254
JCARSR	4	10.40	25.14/0.7174	24.03/0.6039	22.11/0.5967	23.92/0.5607	23.50/0.7338

下载: 导出CSV

表 2 JCARSR与JCAR-SR方法的定量比较

Table 2 Quantitative comparison of JCARSR with JCAR-SR methods

对比方法	尺度因子	Set5	Set14	Urban100	B100	Manga109
对比方法	尺度因子	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
ARCNN^[40] + SwinIR^[46]	2	28.24/0.7988	26.34/0.6936	24.16/0.6947	25.84/0.6480	26.68/0.8184
DnCNN3^[41] + SwinIR^[46]	2	28.54/0.8103	26.50/0.7017	24.40/0.7070	25.96/0.6542	27.01/0.8304
QGCN^[43] + SwinIR^[46]	2	28.93/0.822 7	26.80/0.712 5	25.04/0.735 3	26.13/0.661 8	27.88/0.852 8
FBCNN^[44] + SwinIR^[46]	2	28.87/0.8209	26.80/0.712 0	24.95/0.7311	26.13/0.6614	27.83/0.8527
JDEC^[12] + SwinIR^[46]	2	27.76/0.7920	25.70/0.6678	23.63/0.6724	25.39/0.6236	25.61/0.8024
JCARSR	2	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609
ARCNN^[40] + SwinIR^[46]	4	24.60/0.6806	23.57/0.5826	21.41/0.5563	23.68/0.5485	22.55/0.6881
DnCNN3^[41] + SwinIR^[46]	4	24.77/0.6911	23.70/0.5883	21.55/0.5640	23.74/0.5518	22.71/0.6965
QGCN^[43] + SwinIR^[46]	4	25.04/0.708 5	23.93/0.5995	21.93/0.587 5	23.86/0.5577	23.17/0.7200
FBCNN^[44] + SwinIR^[46]	4	25.02/0.7080	23.94/0.599 9	21.88/0.5851	23.87/0.557 8	23.20/0.723 6
JDEC^[12] + SwinIR^[46]	4	24.58/0.6966	23.38/0.5786	21.10/0.5454	23.54/0.5437	21.82/0.6771
JCARSR	4	25.14/0.7174	24.03/0.6039	22.11/0.5967	23.92/0.5607	23.50/0.7338

下载: 导出CSV

表 3 JCARSR与SR-JCAR方法的定量比较

Table 3 Quantitative comparison of JCARSR with SR-JCAR methods

对比方法	尺度因子	Set5	Set14	Urban100	B100	Manga109
对比方法	尺度因子	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
SwinIR^[46] + ARCNN^[40]	2	27.23/0.759 2	25.59/0.671 7	23.38/0.6634	25.36/0.634 4	25.47/0.778 3
SwinIR^[46] + DnCNN3^[41]	2	26.75/0.7355	25.23/0.6573	23.01/0.6442	25.13/0.6253	24.88/0.7507
SwinIR^[46] + QGCN^[43]	2	26.97/0.7495	25.43/0.6661	23.25/0.6571	25.22/0.6293	25.30/0.7721
SwinIR^[46] + FBCNN^[44]	2	27.14/0.7577	25.55/0.6717	23.38/0.664 2	25.30/0.6333	25.43/0.7781
SwinIR^[46] + JDEC^[12]	2	27.25/0.7601	25.62/0.6701	23.46/0.6619	25.34/0.6315	25.71/0.7880
JCARSR	2	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609
SwinIR^[46] + ARCNN^[40]	4	23.46/0.6134	22.56/0.536 8	20.56/0.508 9	23.02/0.518 1	21.48/0.627 8
SwinIR^[46] + DnCNN3^[41]	4	23.23/0.5936	22.40/0.5223	20.44/0.4956	22.86/0.5066	21.35/0.6079
SwinIR^[46] + QGCN^[43]	4	23.45/0.6124	22.56/0.5350	20.57/0.5068	23.00/0.5172	21.50/0.6246
SwinIR^[46] + FBCNN^[44]	4	23.46/0.613 8	22.58/0.5361	20.58/0.5079	23.01/0.518 1	21.51/0.6253
SwinIR^[46] + JDEC^[12]	4	23.60/0.6170	22.71/0.5381	20.70/0.5102	23.09/0.5196	21.68/0.6286
JCARSR	4	25.14/0.7174	24.03/0.6039	22.11/0.5967	23.92/0.5607	23.50/0.7338

下载: 导出CSV

表 4 不同JPEG压缩因子下的定量比较

Table 4 Quantitative comparison under different JPEG compress factors

对比方法	压缩因子	尺度因子	Set5	Set14	Urban100	B100	Manga109
对比方法	压缩因子	尺度因子	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
SwinIR^[46]	10	2	28.95/0.8231	26.87/0.7141	25.11/0.7377	26.13/0.6622	28.03/0.8565
	20	2	30.91/0.8638	28.31/0.7657	26.53/0.7911	27.33/0.7167	30.23/0.8943
	30	2	31.92/0.8822	29.11/0.7929	27.40/0.8205	27.98/0.7464	31.48/0.9120
	40	2	32.53/0.8926	29.63/0.8092	28.05/0.8385	28.40/0.7651	32.19/0.9214
	80	2	34.65/0.9230	31.48/0.8653	30.09/0.8881	30.19/0.8341	35.26/0.9497
JCARSR	10	2	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609
	20	2	31.01/0.8660	28.41/0.7669	26.69/0.7943	27.38/0.7175	30.55/0.8972
	30	2	32.01/0.8836	29.18/0.7934	27.53/0.8215	28.02/0.7469	31.80/0.9137
	40	2	32.63/0.8939	29.69/0.8098	28.12/0.8386	28.47/0.7659	32.66/0.9235
	80	2	34.74/0.9233	31.50/0.8653	30.13/0.8883	30.20/0.8345	35.51/0.9502
SwinIR^[46]	10	4	25.03/0.7145	23.93/0.6022	21.94/0.5928	23.86/0.5600	23.29/0.7307
	20	4	26.50/0.7577	24.95/0.6369	22.88/0.6377	24.70/0.5917	24.77/0.7754
	30	4	27.48/0.7825	25.56/0.6572	23.42/0.6631	25.13/0.6095	25.59/0.7986
	40	4	28.04/0.7954	25.89/0.6695	23.78/0.6806	25.42/0.6216	26.14/0.8141
	80	4	29.89/0.8409	27.24/0.7229	25.02/0.7360	26.45/0.6733	28.33/0.8626
JCARSR	10	4	25.14/0.7174	24.03/0.6039	22.11/0.5967	23.92/0.5607	23.50/0.7338
	20	4	26.66/0.7608	25.07/0.6387	22.99/0.6386	24.76/0.5925	24.94/0.7767
	30	4	27.54/0.7826	25.60/0.6578	23.51/0.6637	25.18/0.6104	25.78/0.8006
	40	4	28.07/0.7957	25.97/0.6715	23.87/0.6813	25.47/0.6234	26.39/0.8171
	80	4	29.94/0.8424	27.31/0.7246	25.06/0.7362	26.50/0.6745	28.41/0.8637

下载: 导出CSV

表 5 网络模块消融研究

Table 5 Ablation study on the network block

对比方法	Set5	Set14	Urban100	B100	Manga109
对比方法	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
w/o MSNAFB	29.00/0.8232	26.84/0.7122	25.00/0.7311	26.12/0.6608	28.06/0.8545
w/o MA	29.01/0.8237	26.85/0.7124	25.02/0.7316	26.13/0.6610	28.08/0.8545
w/o MSCCB	29.02/0.8240	26.90/0.7136	25.11/0.7359	26.15/0.6618	28.19/0.8570
JCARSR	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609

下载: 导出CSV

表 6 网络分支结构消融研究

Table 6 Ablation study on the network branch structure

对比方法	Set5	Set14	Urban100	B100	Manga109
对比方法	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
ISRB	29.02/0.8242	26.89/0.7137	25.11/0.7366	26.15/0.6619	28.20/0.8575
ISRB$ + $GGB	29.05/0.8242	26.90/0.7139	25.14/0.7373	26.15/0.6621	28.24/0.8578
ISRB$ + $GGB$ + $CARB	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609

下载: 导出CSV

表 7 多分支融合方式消融研究

Table 7 Ablation study on multi-branch fusion methods

信息融合方式	Set5	Set14	Urban100	B100	Manga109
信息融合方式	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
双向	29.04/0.8242	26.89/0.7136	25.10/0.7356	26.15/0.6618	28.19/0.8572
单向	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609

下载: 导出CSV

表 8 梯度引导策略消融研究

Table 8 Ablation study on gradient-guided strategy

GL-ISRB	GL-GGB	Set5	Set14	Urban100	B100	Manga109
GL-ISRB	GL-GGB	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM	PSNR (dB)/SSIM
√	×	29.06/0.8258	26.91/0.7160	25.20/0.7406	26.17/0.6645	28.27/0.8591
×	√	29.13/0.8269	26.96/0.7158	25.28/0.7423	26.19/0.6637	28.37/0.8609
√	√	29.11/0.8268	26.94/0.7161	25.25/0.7412	26.18/0.6642	28.31/0.8595

下载: 导出CSV

参考文献(58)

[1]	Zhang C J, Zheng X L. Decomposing visual and semantic correlations for both fully supervised and few-shot image classification. IEEE Transactions on Artificial Intelligence, 2024, 5(4): 1658−1668 doi: 10.1109/TAI.2023.3329457
[2]	Tang Y P, Wang W N, Zhang C J, Liu J, Zhao Y. Temporal action proposal generation with action frequency adaptive network. IEEE Transactions on Multimedia, 2024, 26: 2340−2353 doi: 10.1109/TMM.2023.3295090
[3]	周登文, 刘子涵, 刘玉铠. 基于像素对比学习的图像超分辨率算法. 自动化学报, 2024, 50(1): 181−193 Zhou Deng-Wen, Liu Zi-Han, Liu Yu-Kai. Pixel-wise contrastive learning for single image super-resolution. Acta Automatica Sinica, 2024, 50(1): 181−193
[4]	Yue Z S, Wang J Y, Loy C C. ResShift: Efficient diffusion model for image super-resolution by residual shifting. In: Proceedings of the 37th International Conference on Neural Information Processing System. New Orleans, USA: Curran Associates Inc., 2023. 13294-13307
[5]	王云涛, 赵蔺, 刘李漫, 陶文兵. 基于组−信息蒸馏残差网络的轻量级图像超分辨率重建. 自动化学报, 2024, 50(10): 2063−2078 Wang Yun-Tao, Zhao Lin, Liu Li-Man, Tao Wen-Bing. G-IDRN: A group-information distillation residual network for lightweight image super-resolution. Acta Automatica Sinica, 2024, 50(10): 2063−2078
[6]	Wang X Y, Hu Q, Cheng Y S, Ma J Y. Hyperspectral image super-resolution meets deep learning: A survey and perspective. IEEE/CAA Journal of Automatica Sinica, 2023, 10(8): 1668−1691 doi: 10.1109/JAS.2023.123681
[7]	Dong C, Loy C C, He K M, Tang X O. Image super-resolution using deep convolutional networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(2): 295−307 doi: 10.1109/TPAMI.2015.2439281
[8]	Hu Q, Ma J Y, Gao Y, Jiang J J, Yuan Y X. Maun: MEMORY-augmented deep unfolding network for hyperspectral image reconstruction. IEEE/CAA Journal of Automatica Sinica, 2024, 11(5): 1139−1150 doi: 10.1109/JAS.2024.124362
[9]	Ma C, Rao Y M, Cheng Y, Chen C, Lu J W, Zhou J. Structure-preserving super resolution with gradient guidance. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE, 2020. 7769−7778
[10]	Fang F M, Li J C, Zeng T Y. Soft-edge assisted network for single image super-resolution. IEEE Transactions on Image Processing, 2020, 29: 4656−4668 doi: 10.1109/TIP.2020.2973769
[11]	Ma C, Rao Y M, Lu J W, Zhou J. Structure-preserving image super-resolution. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(11): 7898−7911 doi: 10.1109/TPAMI.2021.3114428
[12]	Han W K, Im S, Kim J, Jin K H. JDEC: JPEG decoding via enhanced continuous cosine coefficients. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE, 2024. 2784−2793
[13]	Zuo Y F, Xie J C, Wang H, Fang Y M, Liu D Y, Wen W Y. Gradient-guided single image super-resolution based on joint trilateral feature filtering. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33(2): 505−520 Zuo Y F, Xie J C, Wang H, Fang Y M, Liu D Y, Wen W Y. Gradient-guided single image super-resolution based on joint trilateral feature filtering. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33(2): 505−520
[14]	Zhu Y, Zhang Y N, Bonev B, Yuille A L. Modeling deformable gradient compositions for single-image super-resolution. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, USA: IEEE, 2015. 5417−5425
[15]	周圆, 王明非, 杜晓婷, 陈艳芳. 基于层次特征复用的视频超分辨率重建. 自动化学报, 2024, 50(9): 1736−1746 Zhou Yuan, Wang Ming-Fei, Du Xiao-Ting, Chen Yan-Fang. Video super-resolution via hierarchical feature reuse. Acta Automatica Sinica, 2024, 50(9): 1736−1746
[16]	Yang W H, Feng J S, Yang J C, Zhao F, Liu J Y, Guo Z M, et al. Deep edge guided recurrent residual learning for image super-resolution. IEEE Transactions on Image Processing, 2017, 26(12): 5895−5907 doi: 10.1109/TIP.2017.2750403
[17]	Luo Z W, Gustafsson F K, Zhao Z, Sjölund J, Schön T B. Photo-realistic image restoration in the wild with controlled vision-language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, USA: IEEE, 2024. 6641−6651
[18]	Özdenizci O, Legenstein R. Restoring vision in adverse weather conditions with patch-based denoising diffusion models. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(8): 10346−10357 doi: 10.1109/TPAMI.2023.3238179
[19]	Chen X Y, Liu Y H, Pu Y D, Zhang W L, Zhou J T, Qiao Y, et al. Learning a low-level vision generalist via visual task prompt. In: Proceedings of the 32nd ACM International Conference on Multimedia. Melbourne, Australia: ACM, 2024. 2671−2680
[20]	Wu R Y, Yang T, Sun L C, Zhang Z Q, Li S, Zhang L, et al. SeeSR: Towards semantics-aware real-world image super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE, 2024. 25456−25467
[21]	Zhang C J, Bai H H, Zhao Y. Fine-grained image classification by class and image-specific decomposition with multiple views. IEEE Transactions on Multimedia, 2023, 25: 6756−6766 doi: 10.1109/TMM.2022.3214431
[22]	Zhang C J, Cheng J, Tian Q. Multi-view image classification with visual, semantic and view consistency. IEEE Transactions on Image Processing, 2020, 29: 617−627 doi: 10.1109/TIP.2019.2934576
[23]	李金新, 黄志勇, 李文斌, 周登文. 基于多层次特征融合的图像超分辨率重建. 自动化学报, 2023, 49(1): 161−171 Li Jin-Xin, Huang Zhi-Yong, Li Wen-Bin, Zhou Deng-Wen. Image super-resolution based on multi-hierarchical features fusion network. Acta Automatica Sinica, 2023, 49(1): 161−171
[24]	Dong C, Loy C C, Tang X O. Accelerating the super-resolution convolutional neural network. In: Proceedings of the 14th European Conference on Computer Vision. Amsterdam, Netherlands: Springer, 2016. 391−407
[25]	Shi W Z, Caballero J, Huszár F, Totz J, Aitken A P, Bishop R, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, USA: IEEE, 2016. 1874−1883
[26]	Liu Y Q, Jia Q, Fan X, Wang S S, Ma S W, Gao W. Cross-SRN: Structure-preserving super-resolution network with cross convolution. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(8): 4927−4939 doi: 10.1109/TCSVT.2021.3138431
[27]	Lim B, Son S, Kim H, Nah S, Mu Lee K. Enhanced deep residual networks for single image super-resolution. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. Honolulu, USA: IEEE, 2017. 136−144
[28]	Zhang Y L, Tian Y P, Kong Y, Zhong B N, Fu Y. Residual dense network for image super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 2472−2481
[29]	Zhang Y L, Li K P, Li K, Wang L C, Zhong B N, Fu Y. Image super-resolution using very deep residual channel attention networks. In: Proceedings of the 15th European Conference on Computer Vision. Munich, Germany: Springer, 2018. 286−301
[30]	Li J C, Fang F M, Li J Q, Mei K F, Zhang G X. MDCN: Multi-scale dense cross network for image super-resolution. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 31(7): 2547−2561 doi: 10.1109/TCSVT.2020.3027732
[31]	Lai W S, Huang J B, Ahuja N, Yang M H. Deep laplacian pyramid networks for fast and accurate super-resolution. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, USA: IEEE, 2017. 624−632
[32]	Haris M, Shakhnarovich G, Ukita N. Deep back-projection networks for super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 1664−1673
[33]	Zakhor A. Iterative procedures for reduction of blocking effects in transform image coding. IEEE Transactions on Circuits and Systems for Video Technology, 1992, 2(1): 91−95 doi: 10.1109/76.134377
[34]	Zhang X F, Xiong R Q, Fan X P, Ma S W, Gao W. Compression artifact reduction by overlapped-block transform coefficient estimation with block similarity. IEEE Transactions on Image Processing, 2013, 22(12): 4613−4626 doi: 10.1109/TIP.2013.2274386
[35]	Zhang J, Xiong R Q, Zhao C, Zhang Y B, Ma S W, Gao W. CONCOLOR: Constrained non-convex low-rank model for image deblocking. IEEE Transactions on Image Processing, 2016, 25(3): 1246−1259 doi: 10.1109/TIP.2016.2515985
[36]	Foi A, Katkovnik V, Egiazarian K. Pointwise shape-adaptive DCT for high-quality denoising and deblocking of grayscale and color images. IEEE Transactions on Image Processing, 2007, 16(5): 1395−1411 doi: 10.1109/TIP.2007.891788
[37]	Liu X M, Cheung G, Wu X L, Zhao D B. Random walk graph laplacian-based smoothness prior for soft decoding of JPEG images. IEEE Transactions on Image Processing, 2017, 26(2): 509−524 doi: 10.1109/TIP.2016.2627807
[38]	Zhang C J, Liang C, Zhao Y. Exemplar-based, semantic guided zero-shot visual recognition. IEEE Transactions on Image Processing, 2022, 31: 3056−3065 doi: 10.1109/TIP.2021.3120319
[39]	Chen L Y, Chu X J, Zhang X Y, Sun J. Simple baselines for image restoration. In: Proceedings of the 17th European Conference on Computer Vision. Tel Aviv, Israel: Springer, 2022. 17−33
[40]	Dong C, Deng Y B, Loy C C, Tang X O. Compression artifacts reduction by a deep convolutional network. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV). Santiago, Chile: IEEE, 2015. 576−584
[41]	Zhang K, Zuo W M, Chen Y J, Meng D Y, Zhang L. Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising. IEEE Transactions on Image Processing, 2017, 26(7): 3142−3155 doi: 10.1109/TIP.2017.2662206
[42]	Fu X Y, Zha Z J, Wu F, Ding X H, Paisley J. JPEG artifacts reduction via deep convolutional sparse coding. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Korea (South): IEEE, 2019. 2501−2510
[43]	Li J W, Wang Y T, Xie H H, Ma K K. Learning a single model with a wide range of quality factors for JPEG image artifacts removal. IEEE Transactions on Image Processing, 2020, 29: 8842−8854 doi: 10.1109/TIP.2020.3020389
[44]	Jiang J X, Zhang K, Timofte R. Towards flexible blind JPEG artifacts removal. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, Canada: IEEE, 2021. 4997−5006
[45]	Liu Z, Lin Y T, Cao Y, Hu H, Wei Y X, Zhang Z, et al. Swin transformer: Hierarchical vision transformer using shifted windows. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, Canada: IEEE, 2021. 10012−10022
[46]	Liang J Y, Cao J Z, Sun G L, Zhang K, Van Gool L, Timofte R. SwinIR: Image restoration using swin transformer. In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Montreal, Canada: IEEE, 2021. 1833−1844
[47]	Zamir S W, Arora A, Khan S, Hayat M, Khan F S, Yang M H, et al. Learning enriched features for fast image restoration and enhancement. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(2): 1934−1948 doi: 10.1109/TPAMI.2022.3167175
[48]	Hendrycks D, Gimpel K. Gaussian error linear units (GELUs). arXiv preprint arXiv: 1606.08415, 2016.
[49]	Agustsson E, Timofte R. NTIRE 2017 challenge on single image super-resolution: Dataset and study. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Honolulu, USA: IEEE, 2017. 126−135
[50]	Bevilacqua M, Roumy A, Guillemot C, Alberi-Morel M L. Low-complexity single-image super-resolution based on nonnegative neighbor embedding. In: Proceedings of the British Machine Vision Conference. Surrey, UK: BMVC, 2012. 1−10
[51]	Zeyde R, Elad M, Protter M. On single image scale-up using sparse-representations. In: Proceedings of the 7th International Conference on Curves and Surfaces. Avignon, France: Springer, 2012. 711−730
[52]	Huang J B, Singh A, Ahuja N. Single image super-resolution from transformed self-exemplars. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, USA: IEEE, 2015. 5197−5206
[53]	Martin D, Fowlkes C, Tal D, Malik J. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In: Proceedings of the 8th IEEE International Conference on Computer Vision. Vancouver, Canada: IEEE, 2001. 416−423
[54]	Matsui Y, Ito K, Aramaki Y, Fujimoto A, Ogawa T, Yamasaki T, et al. Sketch-based manga retrieval using manga109 dataset. Multimedia Tools and Applications, 2017, 76(20): 21811−21838 doi: 10.1007/s11042-016-4020-z
[55]	Hsu C C, Lee C M, Chou Y S. DRCT: Saving image super-resolution away from information bottleneck. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, USA: IEEE, 2024. 6133−6142
[56]	Chu S C, Dou Z C, Pan J S, Weng S W, Li J B. HMANet: Hybrid multi-axis aggregation network for image super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, USA: IEEE, 2024. 6257−6266
[57]	Wang X T, Xie L B, Dong C, Shan Y. Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data. In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Montreal, Canada: IEEE, 2021. 1905−1914
[58]	Luo Z X, Huang Y, Li S, Wang L, Tan T N. End-to-end alternating optimization for real-world blind super resolution. International Journal of Computer Vision, 2023, 131(12): 3152−3169 doi: 10.1007/s11263-023-01833-7