一种基于深度残差网络的图像修复算法设计与实现

doi:10.19640/j.cnki.jtau.2024.03.015

天津农学院学报 ›› 2024, Vol. 31 ›› Issue (3): 85-91.doi: 10.19640/j.cnki.jtau.2024.03.015

一种基于深度残差网络的图像修复算法设计与实现

吴金苗¹, 王育欣^{1,通信作者}, 韩江宁², 张家亮¹, 张志¹, 魏雨露¹

1.天津农学院计算机与信息工程学院,天津 300392;
2.联通视频科技有限公司业务管理部,天津 300300

收稿日期:2022-10-25 出版日期:2024-06-30 发布日期:2024-07-15
通讯作者: 王育欣（1976—）,男,教授,博士,主要从事能源大数据、农业大数据、人工智能方面研究。E-mail：261235654@qq.com。
作者简介:吴金苗（1999—）,男,本科在读,主要从事人工智能算法等方面研究。E-mail：wu_jinmiao@qq.com。
基金资助:
中国高校产学研创新基金重点项目（2022IT017）; 教育部产学合作协同育人项目（220900287260151）; 天津农学院人才资助项目计划（Y0400907）; 天津农学院研究生教育教学研究与改革重点项目（2021-YA-6）; 天津农学院教育教学研究与改革项目（2023-A-14）; 校企合作技术研发项目（TNHXKJ2022074,TNHXKJ2023009）; 天津农学院教材研究项目（2021-B-01）

Design and implementation of an image inpainting algorithm based on deep residual network

Wu Jinmiao¹, Wang Yuxin^{1,Corresponding Author}, Han Jiangning², Zhang Jialiang¹, Zhang Zhi¹, Wei Yulu¹

1. College of Computer and Information Engineering, Tianjin Agricultural University, Tianjin 300392, China;
2. Business Management Department, China Unicom Video Technology Co., Ltd., Tianjin 300300, China

Received:2022-10-25 Online:2024-06-30 Published:2024-07-15

摘要/Abstract

摘要： 随着数字图像在人们日常生活中使用越来越多,针对破损图像缺失区域的数字图像修复问题,本文提出一种基于深度残差网络的图像修复模型。该模型总体架构基于编码-解码结构,编码器采用不同深度的残差网络,解码器分别使用反卷积网络结构和上采样-卷积结构。通过实验探讨在本模型中不同结构的解码器、编码器以及不同的损失函数对图像修复效果的影响。实验结果表明：本文提出的基于深度残差网络的图像修复模型,采用修改后的Resnet 34-layer作为编码器,反卷积网络作为解码器,L1 Loss作为损失函数,能够达到较好的图像修复效果。

关键词: 图像修复, 残差网络, 反卷积, 编码器-解码器

Abstract: With the increasing use of digital images in people's daily life, restoring the missing areas of damaged images has become a problem, for which this paper proposes an image inpainting model based on a deep residual network. The overall architecture of the model is based on the encoder-decoder structure. The encoder uses residual networks of different depths, and the decoder uses a deconvolution network structure and an upsampling-convolution structure. The effects of different structure decoders, encoders and different loss functions on the image inpainting effect in this model were discussed through experiments. The experimental results showed that the image inpainting model based on the deep residual network proposed in this paper could achieve good image inpainting effect by adopting the modified Resnet 34-layer as the encoder, the deconvolution network as the decoder, and the L1 Loss as the loss function.

Key words: image inpainting, residual network, deconvolution, encoder-decoder

中图分类号:

TP391.4

吴金苗, 王育欣, 韩江宁, 张家亮, 张志, 魏雨露. 一种基于深度残差网络的图像修复算法设计与实现[J]. 天津农学院学报, 2024, 31(3): 85-91.

Wu Jinmiao, Wang Yuxin, Han Jiangning, Zhang Jialiang, Zhang Zhi, Wei Yulu. Design and implementation of an image inpainting algorithm based on deep residual network[J]. Journal of Tianjin Agricultural University, 2024, 31(3): 85-91.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://xuebao.tjau.edu.cn/CN/10.19640/j.cnki.jtau.2024.03.015

http://xuebao.tjau.edu.cn/CN/Y2024/V31/I3/85

参考文献

[1] 李权锋. 图像修复研究进展综述[J]. 计算机时代,2021(12):6-9.
[2] 赵露露,沈玲,洪日昌. 图像修复研究进展综述[J]. 计算机科学,2021,48(3):14-26.
[3] PATHAK D,KRAHENBUHL P,DONAHUE J,et al.Context encoders:Feature learning by inpainting[C]// IEEE. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington DC:IEEE Computer Society,2016:2536-2544.
[4] HE K M,ZHANG X Y,REN S Q,et al.Deep residual learning for image recognition[C]//IEEE. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington DC:IEEE Computer Society,2016:770-778.
[5] ODENA A,DUMOULIN V,OLAH C.Deconvolution and checkerboard artifacts[J]. Distill,2016,1(10):143.
[6] 张焕,张庆,于纪言. 激活函数的发展综述及其性质分析[J]. 西华大学学报(自然科学版),2021,40(4):1-10.
[7] PHILBIN J,CHUM O,ISARD M,et al.Lost in quantization:Improving particular object retrieval in large scale image databases[C]//IEEE. 2008 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2008:1-8.
[8] DOZAT T.Incorporating nesterov momentum into Adam[C]//ICLR. Proceedings of ICLR 2016 Workshop. Addis Ababa:ICLR,2016:1-4.
[9] YU J H,LIN Z,YANG J M,et al.Generative image inpainting with contextual attention[C]//IEEE. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City:IEEE,2018:5505-5514.
[10] CHUNG G,VESE L A.Image segmentation using a multilayer level-set approach[J]. Computing and Visualization in Science,2009,12(6):267-285.
[11] HORE A,ZIOU D.Image quality metrics:PSNR vs. SSIM[C]//IEEE. 20th International Conference on Pattern Recognition. Piscataway:IEEE,2010:2366-2369.
[12] SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[J]. Computer Science,2014,23:1409.

一种基于深度残差网络的图像修复算法设计与实现

Design and implementation of an image inpainting algorithm based on deep residual network

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 8

编辑推荐

Metrics

本文评价

[1]	安瑞钰, 郝志斌. 基于半监督主动学习的小麦叶片病虫害区域分割方法[J]. 天津农学院学报, 2024, 31(2): 87-93.
[2]	马含章, 王淼鑫, 钟拼扬, 刘华, 常若葵, 董晋峰. 视觉识别的生活垃圾分类系统设计[J]. 天津农学院学报, 2023, 30(4): 53-57.
[3]	郑海波, 宋爽, 刘同海. 基于改进Tiny YOLOv3的交通标志检测方法[J]. 天津农学院学报, 2023, 30(2): 62-68.
[4]	高佳颀, 吴芳. 基于机器学习的农村水质水色分类研究[J]. 天津农学院学报, 2022, 29(1): 61-65.
[5]	纪思琪, 吴芳, 李乃祥. 基于决策树的蔬菜病害静态预警模型[J]. 天津农学院学报, 2017, 24(2): 77-80.
[6]	王启欣, 邓胜琳. 一种基于计算机图像处理经典算法的染色体边缘检测方法[J]. 天津农学院学报, 2012, 19(1): 42-44.
[7]	赵娜, 赵平, 高轶军. 机器视觉技术在我国现代农业生产中的应用研究[J]. 天津农学院学报, 2015, 22(2): 55-58.
[8]	马国强, 马吉飞, 郭鹏, 宋志恒, 于娜. 改进的医学CT图像K-均值聚类分割算法的实现[J]. 天津农学院学报, 2013, 20(4): 28-35.