基于多方向梯度网络的自适应边缘信息图像去噪模型

doi:10.13229/j.cnki.jdxbgxb.20231453

吉林大学学报(工学版) ›› 2025, Vol. 55 ›› Issue (10): 3319-3328.doi: 10.13229/j.cnki.jdxbgxb.20231453

• 计算机科学与技术 • 上一篇

基于多方向梯度网络的自适应边缘信息图像去噪模型

王梓桐¹(),赵晶²,乔双³,朱芮⁴()

^1.吉林大学计算机科学与技术学院，长春 130012
^2.北京理工大学光电学院，北京 100081
^3.东北师范大学物理学院，长春 130024
^4.吉林大学通信工程学院，长春 130012

收稿日期:2023-12-28 出版日期:2025-10-01 发布日期:2026-02-03
通讯作者: 朱芮 E-mail:softlikeyou@163.com;zhurui@jlu.edu.cn
作者简介:王梓桐（2000-），女，博士研究生. 研究方向：深度学习及图像去噪. E-mail： softlikeyou@163.com
基金资助:
吉林省科技厅青年成长科技项目(20230508164RC)

Adaptive edge information image denoising model based on multi-directional gradient network

Zi-tong WANG¹(),Jing ZHAO²,Shuang QIAO³,Rui ZHU⁴()

^1.College of Computer Science and Technology，Jilin University，Changchun 130012，China
^2.School of Optics and Photonics，Beijing Institute of Technology，Beijing 100081，China
^3.College of Physics，Northeast Normal University，Changchun 130024，China
^4.College of Communication Engineerging，Jilin University，Changchun 130012，China

Received:2023-12-28 Online:2025-10-01 Published:2026-02-03
Contact: Rui ZHU E-mail:softlikeyou@163.com;zhurui@jlu.edu.cn

摘要/Abstract

摘要：

针对现有的基于学习的图像去噪算法不能很好地保留图像边缘和纹理信息的不足，本文提出了一种基于多方向梯度网络的自适应边缘信息图像去噪模型，能够分别捕获不同类别的图像信息。首先，采用多方向梯度算子过滤干净目标图像以获取无噪声梯度图，引导多方向梯度网络学习无噪声梯度图。其次，提出自适应梯度融合模块，自适应地融合梯度信息与噪声图像，提高去噪网络对边缘和纹理信息的关注度。实验结果表明，本文模型在PSNR和SSIM指标方面具有良好性能。此外，去噪后的图像始终具有更好的视觉质量，从而展示了其在图像去噪中的应用潜力。

关键词: 深度神经网络, 图像去噪, 梯度算子, 自适应融合

Abstract:

To address the limitation that existing learning-based image denoising algorithms struggle to preserve edges and textures， we propose an adaptive edge-aware denoising model built upon a multi-directional gradient network that can capture distinct image information separately. First， multi-directional gradient operators are applied to the clean target image to generate noise-free gradient maps， which then guide the network in learning gradient representations free from corruption. Second， an adaptive gradient-fusion module is introduced to fuse gradient cues with the noisy image adaptively， increasing the network’s attention to edge and texture details. Experimental results demonstrate that the proposed model achieves competitive PSNR and SSIM values. Moreover， the denoised images consistently exhibit superior visual quality， underscoring the model’s potential for practical image-denoising applications.

Key words: deep neural network, image denoising, gradient operator, adaptive fusion

中图分类号:

TP391.4

王梓桐,赵晶,乔双,朱芮. 基于多方向梯度网络的自适应边缘信息图像去噪模型[J]. 吉林大学学报(工学版), 2025, 55(10): 3319-3328.

Zi-tong WANG,Jing ZHAO,Shuang QIAO,Rui ZHU. Adaptive edge information image denoising model based on multi-directional gradient network[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(10): 3319-3328.

图/表 13

图1

图2

图3

表1

表2

表3

不同去噪算法在3个标准库上的平均PSNR（dB）/SSIM对比（σn=50）"

数据库	Set12		BSD68		Urban100
指标	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
DnCNN	27.18	0.782 9	26.23	0.718 9	26.28	0.787 4
FFDNet	27.32	0.790 3	26.29	0.724 5	26.52	0.805 7
MemNet	27.38	0.793 3	26.35	0.729 7	26.64	0.802 9
$N 3 N e t$	27.43	0.795 0	26.40	0.730 2	26.82	0.814 1
MWDCNN	27.36	0.790 4	26.40	0.729 7	27.12	0.806 3
LIGN	27.61	0.793 3	26.53	0.736 2	27.68	0.826 1
DRANet	27.62	0.800 3	26.47	0.731 6	27.90	0.829 4
本文算法	27.76	0.803 9	26.56	0.739 2	28.09	0.836 4

表3

表4

不同去噪算法在Set12数据集上的单张图像PSNR（dB）对比（σn=50）"

算法	C.man	House	Peppers	Starfish	Monarch	Airplane	Parrot	Lena	Barbara	Boat	Man	Couple
DnCNN	27.03	30.00	27.32	25.70	26.78	25.87	26.48	29.39	26.22	27.20	27.24	26.90
FFDNet	27.05	30.37	27.54	25.75	26.81	25.89	26.57	29.66	26.45	27.33	27.29	27.08
MemNet	27.17	30.52	27.47	25.81	26.98	25.95	26.52	29.68	26.67	27.34	27.30	27.14
$N 3 N e t$	27.18	30.60	27.63	26.13	27.02	25.94	26.47	29.65	26.92	27.33	27.26	27.03
MWDCNN	27.15	30.56	27.44	25.82	27.00	25.83	26.53	29.64	26.67	27.32	27.22	27.08
LIGN	27.30	30.86	27.67	26.39	27.13	26.03	26.67	29.87	27.34	27.49	27.30	27.26
DRANet	27.58	30.89	27.62	25.99	27.13	26.03	26.67	29.59	27.34	27.57	27.34	27.36
本文算法	27.42	31.09	27.81	26.69	27.22	26.04	26.67	29.96	27.82	27.58	27.37	27.38

表4

图4

图5

图6

表5

不同去噪网络模型参数量及在尺寸为256×256、512×512和1 024×1 024且噪声等级为50的噪声图像上的运行时间比较 (s)"

尺寸/资源消耗	DnCNN	MemNet	FFDNet	$N 3 N e t$	DRANet	MWDCNN	LIGN	MDGAENet
Size 256×256	0.01	0.88	0.01	0.17	0.08	0.06	0.02	0.01
Size 512×512	0.05	3.61	0.03	0.74	0.38	0.22	0.06	0.12
Size 1 024×1 024	0.16	14.69	0.11	3.25	1.24	3.89	2.10	1.33
Memory cost/M	0.554	0.667	0.490	0.706	1.211	0.525	3.749	3.845

表5

表6

图7

参考文献 42

[1]	Chen T, Ma K K, Chen L H. Tri-state median filter for image denoising[J]. IEEE Transactions on Image Processing, 1999, 8(12): 1834-1838.
[2]	Zhang X B, Zhang S L. Diffusion scheme using mean filter and wavelet coefficient magnitude for image denoising[J]. AEU-International Journal of Electronics and Communications, 2016, 70(7): 944-952.
[3]	Buades A, Coll B, Morel J M. A non-local algorithm for image denoising[C]∥IEEE Conference on Computer Vision and Pattern Recognition, Piscataway, LISA, 2005: 60-65.
[4]	Dabov K, Foi A, Katkovnik V, et al. Image denoising by sparse 3-D transform-domain collaborative filtering[J]. IEEE Transactions on Image Processing, 2007, 16(8): 2080-2095.
[5]	Dabov K, Foi A, Katkovnik V, et al. Color image denoising via sparse 3D collaborative filtering with grouping constraint in luminance-chrominance space[C]∥IEEE International Conference on Image Processing, Piscataway, LISA, 2007: 313-I-6.
[6]	Rudin L I, Osher S, Fatemi E. Nonlinear total variation based noise removal algorithms[J]. Physica D: nonlinear Phenomena, 1992, 60(1-4): 259-268.
[7]	Weiss Y, Freeman W T. What makes a good model of natural images[C]∥IEEE Conference on Computer Vision and Pattern Recognition, Piscataway, USA, 2007: 1-8.
[8]	张红英, 彭启琮. 全变分自适应图像去噪模型[J]. 光电工程, 2006, 33(3): 50-53.
	Zhang Hong-ying, Peng Qi-cong. Adaptive total variation image denoising model[J]. Opto-Electronic Engineering, 2006, 33(3): 50-53.
[9]	Elad M, Aharon M. Image denoising via sparse and redundant representations over learned dictionaries[J]. IEEE Transactions on Image Processing, 2006, 15(12): 3736-3745.
[10]	Mairal J, Elad M, Sapiro G. Sparse representation for color image restoration[J]. IEEE Transactions on Image Processing, 2007, 17(1): 53-69.
[11]	Xie Q, Zhao Q, Meng D Y, et al. Multispectral images denoising by intrinsic tensor sparsity regularization[C]. IEEE Conference on Computer Vision and Pattern Recognition, Piscataway, USA, 2016: 1692-1700.
[12]	郑毅贤. 基于稀疏表示理论的图像去噪方法研究[D]. 上海: 上海交通大学电子信息与电气工程学院, 2013.
	Zheng Yi-xian. Research on image denoising methods based on sparse representation theory[D]. Shanghai: School of Electronic Information and Electronical Engineering, Shanghai Jiao Tong University, 2013.
[13]	银壮辰. 基于稀疏表示和字典学习的图像去噪研究[D]. 武汉: 武汉理工大学电子信息学院, 2014.
	Yin Zhuang-chen. Research on image denoising based on sparse representation and dictionary learning [D]. Wuhan: School of Electronic Information Wuhan University of Technology, 2014.
[14]	Gu S H, Zhang L, Zuo W M, et al. Weighted nuclear norm minimization with application to image denoising[C]∥IEEE Conference on Computer Vision and Pattern Recognition, Piscataway, USA, 2014: 2862-2869.
[15]	Xu J, Zhang L, Zhang D, et al. Multi-channel weighted nuclear norm minimization for real color image denoising[C]∥IEEE International Conference on Computer Vision, Piscataway, USA, 2017: 1096-104.
[16]	Xie T, Li S T, Sun B. Hyperspectral images denoising via nonconvex regularized low-rank and sparse matrix decomposition[J]. IEEE Transactions on Image Processing, 2019, 29: 44-56.
[17]	Jia X X, Feng X C, Wang W W. Adaptive regularizer learning for low rank approximation with application to image denoising[C]∥IEEE International Conference on Image Processing, Piscataway, USA, 2016: 3096-3100.
[18]	王圳萍. 基于低秩矩阵恢复的图像去噪算法研究[D]. 成都: 西南交通大学信息科学与技术学院, 2015.
	Wang Zhen-ping. Research on image denoising algorithms based on low-rank matrix recovery[D]. Chengdu: School of Information Science and Technology, Southwest Jiaotong University, 2015.
[19]	Xie J Y, Xu L L, Chen E H. Image denoising and inpainting with deep neural networks[C]∥Advances in Neural Information Processing Systems. Red Hcok, NY: Curran Associates, Inc., 2012: 341-349.
[20]	Agostinelli F, Anderson M R, Lee H. Adaptive multi-column deep neural networks with application to robust image denoising[C]∥Advances in Neural Information Processing Systems, Red Hcok, NY: Curran Associates, Inc., 2013: 1493-1501.
[21]	Sivakumar K, Desai U B. Image restoration using a multilayer perceptron with a multilevel sigmoidal function[J]. IEEE Transactions on Signal Processing, 1993, 41(5): 2018-2022.
[22]	Burger H C, Schuler C J, Harmeling S. Image denoising: Can plain neural networks compete with BM3D[C]∥IEEE Conference on Computer Vision and Pattern Recognition, Piscataway, USA, 2012: 2392-2399.
[23]	Zhou Y T, Chellappa R, Vaid A, et al. Image restoration using a neural network[J]. IEEE Transactions on Acoustics, Speech, and Signal Processing, 1988, 36(7): 1141-1151.
[24]	Jain V, Seung H S. Natural image denoising with convolutional networks[C]∥Advances in Neural Information Processing Systems. Red Hcok, NY: Curran Associates, Inc., 2008, 21.
[25]	Chen Y J, Pock T. Trainable nonlinear reaction diffusion: A flexible framework for fast and effective image restoration[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(6): 1256-1272.
[26]	Zhang K, Zuo W M, Chen Y J, et al. Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising[J]. IEEE Transactions on Image Processing, 2017, 26(7): 3142-3155.
[27]	Mao X J, Shen C, Yang Y B. Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connections[C]∥Advances in Neural Information Processing Systems, Piscataway, USA, 2016: 379-387.
[28]	Tai Y, Yang J, Liu X M, et al. MemNet: A persistent memory network for image restoration[C]∥IEEE International Conference on Computer Vision, Piscataway, USA, 2017: 4539-4547.
[29]	Zhang K, Zuo W, Zhang L. FFDNet: Toward a fast and flexible solution for CNN-based image denoising[J]. IEEE Transactions on Image Processing, 2018, 27(9): 4608-4622.
[30]	Liu D, Wen B H, Fan Y C, et al. Non-local recurrent network for image restoration[C]∥Advances in Neural Information Processing Systems, Red Hcok, NY: Curran Associates, Inc., 2018: 1673-1682.
[31]	Plötz T, Roth S. Neural nearest neighbors networks[C]∥Advances in Neural Information Processing Systems, Red Hcok, NY: Curran Associates, Inc., 2018: 1095-1106.
[32]	Charbonnier P, Blanc-Feraud L, Aubert G, et al. Two deterministic half-quadratic regularization algorithms for computed imaging[C]∥Proceedings of 1st International Conference on Image Processing, Piscataway, USA, 1994, 2: 168-172.
[33]	Kingma D P, Ba J.Adam: a method for stochastic optimization[J]. Computer Science, 2014: 1-15.
[34]	Wang Z, Bovik A C, Sheikh H R, et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612.
[35]	Martin D, Fowlkes C, Tal D, et al. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics[C]∥Proceedings Eighth IEEE International Conference on Computer Vision, Piscataway, USA,2001: 416-423.
[36]	Timofte R, Agustsson E, Van Gool L, et al. Ntire 2017 challenge on single image super-resolution: Methods and results[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, Piscataway, USA, 2017: 114-125.
[37]	Zhang K, Zuo W M, Chen Y J, et al. Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising[J]. IEEE Transactions on Image Processing, 2017, 26(7): 3142-3155.
[38]	Roth S, Black M J. Fields of experts[J]. International Journal of Computer Vision, 2009, 82: 205-229.
[39]	Huang J B, Singh A, Ahuja N. Single image super-resolution from transformed self-exemplars[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Piscataway,USA, 2015: 5197-5206.
[40]	Qiao S, Yang J R, Zhang T, et al. Layered input GradiNet for image denoising[J]. Knowledge-Based Systems, 2022, 254: No.109587.
[41]	Tian C W, Zheng M H, Zuo W M, et al. Multi-stage image denoising with the wavelet transform[J]. Pattern Recognition, 2023, 134(1): No.109050.
[42]	Wu W C, Liu S J, Xia Y, et al. Dual residual attention network for image denoising[J]. Pattern Recognition, 2024, 149: No.110291.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

数据库	Set12		BSD68		Urban100
指标	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
DnCNN	32.86	0.903 1	31.73	0.890 7	32.68	0.925 5
FFDNet	32.75	0.902 7	31.63	0.890 2	32.43	0.927 3
MemNet	32.84	0.904 4	31.70	0.892 7	32.84	0.926 1
N³Net	-	-	-	-	-	-
MWDCNN	32.87	0.907 2	31.87	0.897 6	33.15	0.931 3
LIGN	33.03	0.909 9	31.77	0.896 4	33.46	0.935 6
DRANet	33.00	0.905 6	31.79	0.892 1	33.55	0.932 8
本文算法	33.19	0.911 8	31.93	0.900 6	33.86	0.938 9

数据库	Set12		BSD68		Urban100
指标	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
DnCNN	30.44	0.862 2	29.23	0.828 7	29.97	0.879 7
FFDNet	30.43	0.863 4	29.19	0.828 9	29.92	0.888 6
MemNet	30.43	0.860 2	29.20	0.826 2	30.45	0.882 9
N³Net	30.50	0.865 1	29.30	0.832 9	30.19	0.891 0
WMDCNN	30.52	0.867 0	29.34	0.837 6	30.64	0.892 9
LIGN	30.72	0.871 1	29.25	0.833 4	31.04	0.901 3
DRANet	30.69	0.868 1	29.36	0.832 6	31.18	0.899 9
本文算法	30.83	0.873 0	29.48	0.842 4	31.39	0.906 3

	σ=50		σ=25		σ=15
评估指标	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
案例1	27.76	0.802 4	30.82	0.873 6	33.19	0.911 8
案例2	27.31	0.785 0	30.52	0.867 2	32.82	0.906 7
案例3	27.60	0.800 0	30.71	0.869 4	33.06	0.909 9

基于多方向梯度网络的自适应边缘信息图像去噪模型

Adaptive edge information image denoising model based on multi-directional gradient network

RICH HTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 42

相关文章 14

Metrics

本文评价

推荐阅读 0

[1]	邓天民,谢鹏飞,余洋,陈月田. 双分支特征自适应融合的车道线检测方法[J]. 吉林大学学报(工学版), 2025, 55(12): 3840-3851.
[2]	张锦洲,姬世青,谭创. 融合卷积神经网络和双边滤波的相贯线焊缝提取算法[J]. 吉林大学学报(工学版), 2024, 54(8): 2313-2318.
[3]	巨涛,刘帅,火久元,张学军. 深度神经网络模型并行自适应计算任务调度方法[J]. 吉林大学学报(工学版), 2024, 54(12): 3601-3613.
[4]	刘鹏举. 基于深度神经网络的物联网安全态势自动辨识算法设计[J]. 吉林大学学报(工学版), 2023, 53(7): 2121-2126.
[5]	冀汶莉,田忠,柴敬,张丁丁,王斌. 多属性融合分布式光纤导水裂隙带高度预测方法[J]. 吉林大学学报(工学版), 2023, 53(4): 1200-1210.
[6]	成丽波,李新月,李喆,贾小宁. 基于曲波变换与拟合优度检验的遥感图像去噪方法[J]. 吉林大学学报(工学版), 2023, 53(11): 3207-3213.
[7]	宋现敏,杨舒天,刘明鑫,李志慧. 站点间公交行程时间波动特性及预测方法[J]. 吉林大学学报(工学版), 2022, 52(8): 1792-1799.
[8]	姜超, 耿则勋, 刘立勇, 潘映峰. 结合噪声去除的极大似然图像复原[J]. 吉林大学学报(工学版), 2015, 45(4): 1360-1366.
[9]	李抵非，田地，胡雄伟. 基于深度学习的中文标准文献语言模型[J]. 吉林大学学报(工学版), 2015, 45(2): 596-599.
[10]	祁冰, 孔韦韦. 基于改进型交叉视觉皮层模型的自适应图像去噪[J]. 吉林大学学报(工学版), 2014, 44(01): 184-190.
[11]	尹奎英¹,刘宏伟²,金林¹. 快速的Otsu双阈值SAR图像分割法[J]. 吉林大学学报(工学版), 2011, 41(6): 1760-1765.
[12]	陈玫玫, 郭树旭, 王瑶, 吴斌, 于思瑶, 邵向鑫. 压缩感知的指静脉图像去噪[J]. 吉林大学学报(工学版), 2011, 41(02): 559-0562.
[13]	王宏志，武伟，钟诚 . 基于非线性扩散与小波变换的混合图像去噪算法[J]. 吉林大学学报(工学版), 2009, 39(02): 525-0529.
[14]	崔艳秋；王珂；孙巍. 基于小波域自适应高斯混合模型的图像去噪方法[J]. 吉林大学学报(工学版), 2006, 36(06): 983-0988.