结构引导的渐进式生成对抗壁画修复

陈永; 陈锦; 陶美风

doi:10.13700/j.bh.1001-5965.2021.0440

结构引导的渐进式生成对抗壁画修复

doi: 10.13700/j.bh.1001-5965.2021.0440

陈永^{1, 2, ,},
陈锦¹,
陶美风¹

1.
兰州交通大学电子与信息工程学院，兰州 730070
2.
甘肃省人工智能与图形图像处理工程研究中心，兰州 730070

基金项目: 国家自然科学基金(61963023)；教育部人文社会科学研究青年基金(19YJC760012)；兰州交通大学基础研究拔尖人才项目(2022JC36)；兰州交通大学天佑创新团体(TY202003)

详细信息

通讯作者:
E-mail：edukeylab@126.com

中图分类号: TN911.73
计量
- 文章访问数: 576
- HTML全文浏览量: 130
- PDF下载量: 91
- 被引次数: 0
出版历程
- 收稿日期: 2021-08-04
- 录用日期: 2021-11-05
- 网络出版日期: 2021-11-16
- 整期出版日期: 2023-06-30

Mural inpainting progressive generative adversarial networks based on structure guided

CHEN Yong^{1, 2
, ,},
CHEN Jin¹,
TAO Meifeng¹

1.
School of Electronic and Information Engineering，Lanzhou Jiaotong University，Lanzhou 730070，China
2.
Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphics & Image Processing，Lanzhou 730070，China

Funds: National Natural Science Foundation of China (61963023); Ministry of Education in China Project of Humanities and Social Sciences Youth Foundation (19YJC760012); Basic Top-Notch Personnel Project of Lanzhou Jiaotong University (2022JC36); Tianyou Innovation Team of Lanzhou Jiaotong University (TY202003)

More Information

Corresponding author: E-mail：edukeylab@126.com

摘要

摘要:
针对破损壁画图像修复过程中存在的结构修复不当及修复后壁画细节重构丢失等问题，提出了一种基于结构引导的渐进式生成对抗壁画修复深度学习模型。设计结构生成器对壁画缺失结构内容进行修复，得到修复的壁画结构图像。通过壁画生成器生成对抗学习，结合改进的双池化SKNet多尺度特征提取模块，利用修复后的结构图像引导破损壁画实现渐进式修复，以提高壁画的细节特征学习能力。通过局部判别器和全局判别器，完成对结构图像和壁画图像的重构判别，增强壁画修复效果的全局一致性。通过对真实敦煌壁画数字化修复的实验表明：所提方法能够有效修复破损的敦煌壁画，修复后的壁画具有更好的结构及细节信息，在主客观评价指标上均优于比较方法。
- 图像重构 /
- 壁画修复 /
- 结构引导 /
- 双池化特征选择 /
- 生成对抗网络
Abstract:
Aiming at the problems of improper structural repair and loss of mural detail reconstruction after repairing during the process of damaged mural image inpainting, mural inpainting progressive generative adversarial networks based on structure guided is proposed. Firstly, a structure generator is designed to generate the missing structure content of the mural. Secondly, the mural generator is used to generate adversarial learning, and combined with the improved double pooling SKNet multi-scale feature extraction modular, the repaired structure image is used to guide the damaged mural to achieve progressive repair, which improves the detailed feature learning ability of the mural. Lastly, the reconstruction of the structural picture and the mural image is finished using the local discriminator and the global discriminator, which improves the overall consistency of the mural restoration result. Experiments on digital restoration of real Dunhuang murals show that the proposed method can effectively repair damaged Dunhuang murals, and the restored murals have a stronger structure and high-quality texture details than other comparison algorithms. Meanwhile, the proposed has better both subjective and objective evaluation.
- image reconstruction /
- mural inpainting /
- structure guidance /
- double pooling feature selection /
- generative adversarial networks

HTML全文

图 1 本文总体模型框架

Figure 1. Model framework of the proposed method

下载: 全尺寸图片幻灯片

图 2 结构生成器示意图

Figure 2. Schematic diagram of structure generator

下载: 全尺寸图片幻灯片

图 3 编码器和解码器结构

Figure 3. Structure of auto encoder and decoder

下载: 全尺寸图片幻灯片

图 4 壁画生成器结构

Figure 4. Structure of mural generative network

下载: 全尺寸图片幻灯片

图 5 特征提取效果示意图

Figure 5. Schematic diagram of feature extraction effect

下载: 全尺寸图片幻灯片

图 6 平均池化

Figure 6. Average pooling

下载: 全尺寸图片幻灯片

图 7 最大池化

Figure 7. Max pooling

下载: 全尺寸图片幻灯片

图 8 双池化示意图

Figure 8. Schematic diagram of double pooling

下载: 全尺寸图片幻灯片

图 9 双池化SKNet多尺度特征提取模块结构

Figure 9. Framework of double pooling SKNet multi-scale feature extraction module

下载: 全尺寸图片幻灯片

图 10 全局判别器和局部判别器

Figure 10. Global discriminator and local discriminator

下载: 全尺寸图片幻灯片

图 11 中心掩膜修复实验对比

Figure 11. Comparison of center mask inpainting experiments

下载: 全尺寸图片幻灯片

图 12 随机掩膜修复实验对比

Figure 12. Comparison of random mask inpainting experiments

下载: 全尺寸图片幻灯片

图 13 真实破损壁画修复实验对比

Figure 13. Comparison of inpainting experiments of real broken murals

下载: 全尺寸图片幻灯片

图 14 大幅复杂结构破损壁画修复实验对比

Figure 14. Comparison of inpainting experiments of large-scale and complex broken murals

下载: 全尺寸图片幻灯片

图 15 不同方法训练样本比例与PLCC关系

Figure 15. Relationship between proportion of different algorithm training samples and PLCC

下载: 全尺寸图片幻灯片

表 1 中心掩膜修复结果PSNR和SSIM对比

Table 1. Comparison of PSNR and SSIM of center mask inpainting results

图像	PSNR/dB			SSIM
图像	文献[11] 方法	文献[13] 方法	本文方法	文献[11] 方法	文献[13] 方法	本文方法
1	28.97	29.55	31.12	0.8488	0.8670	0.9381
2	24.21	27.36	33.82	0.8152	0.8492	0.9565
3	25.75	26.57	28.95	0.8281	0.8651	0.9035
4	21.57	28.19	33.42	0.8368	0.8690	0.9558
5	27.01	28.39	31.72	0.8866	0.8850	0.9156

下载: 导出CSV

表 2 随机掩膜修复结果PSNR和SSIM对比

Table 2. Comparison of PSNR and SSIM of random mask inpainting results

图像	PSNR/dB			SSIM
图像	文献[11] 方法	文献[13] 方法	本文方法	文献[11] 方法	文献[13] 方法	本文方法
1	17.31	23.21	23.45	0.6135	0.8614	0.8947
2	21.31	24.00	24.10	0.5846	0.6547	0.6771
3	21.29	27.63	29.46	0.6431	0.7935	0.8495
4	23.98	23.73	23.89	0.7550	0.7849	0.8034
5	23.98	27.89	27.95	0.4791	0.5799	0.5745

下载: 导出CSV

表 3 不同方法模型平均准确率比较

Table 3. Comparison of average accuracy of different algorithm models

方法	均方误差
文献[11]	377.774
文献[13]	147.384
本文	136.939

下载: 导出CSV

表 4 不同方法修复平均时间对比

Table 4. Comparison of average repair time of different algorithms

方法	修复平均时间/s
文献[11]	8.21
文献[13]	10.62
本文	7.58

下载: 导出CSV

参考文献(22)

[1]	WANG H, LI Q Q, JIA S. A global and local feature weighted method for ancient murals inpainting[J]. International Journal of Machine Learning and Cybernetics, 2020, 11(6): 1197-1216. doi: 10.1007/s13042-019-01032-2
[2]	SHAO H, WANG Y X. Generative image inpainting with salient prior and relative total variation[J]. Journal of Visual Communication and Image Representation, 2021, 79: 103231. doi: 10.1016/j.jvcir.2021.103231
[3]	BRKIC A L, MITROVIC D, NOVAK A. On the image inpainting problem from the viewpoint of a nonlocal Cahn-Hilliard type equation[J]. Journal of Advanced Research, 2020, 25: 67-76. doi: 10.1016/j.jare.2020.04.015
[4]	YANG X H, GUO B L, XIAO Z L, et al. Improved structure tensor for fine-grained texture inpainting[J]. Signal Processing:Image Communication, 2019, 73: 84-95. doi: 10.1016/j.image.2018.02.006
[5]	FAN Y. Damaged region filling by improved criminisi image inpainting algorithm for thangka[J]. Cluster Computing, 2019, 22(6): 13683-13691.
[6]	LI P, CHEN W G, NG M K. Compressive total variation for image reconstruction and restoration[J]. Computers and Mathematics with Applications, 2020, 80(5): 874-893. doi: 10.1016/j.camwa.2020.05.006
[7]	BINI A A. Image restoration via DOST and total variation regularisation[J]. IET Image Processing, 2019, 13(3): 458-468. doi: 10.1049/iet-ipr.2018.5504
[8]	WAN W, HUANG H Y, LIU J. Local block operators and TV regularization based image inpainting[J]. Inverse Problems & Imaging, 2018, 12(6): 1389-1410.
[9]	陈永, 艾亚鹏, 郭红光. 改进曲率驱动模型的敦煌壁画修复算法[J]. 计算机辅助设计与图形学学报, 2020, 32(5): 787-796. CHEN Y, AI Y P, GUO H G. Improved curvature-driven model of Dunhuang mural restoration algorithm[J]. Journal of Computer Aided Design and Graphics, 2020, 32(5): 787-796(in Chinese).
[10]	QIN J, BAI H H, ZHAO Y. Multi-scale attention network for image inpainting[J]. Computer Vision and Image Understanding, 2021, 204: 103155. doi: 10.1016/j.cviu.2020.103155
[11]	XIE C H, LIU S H, LI C, et al. Image inpainting with learnable bidirectional attention maps[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 8858-8867.
[12]	DU W C, CHEN H, YANG H. Learning invariant representation for unsupervised image restoration[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 11224-11233.
[13]	DEMIR U, UNAL G. Patch-based image inpainting with generative adversarial networks[EB/OL]. (2018-03-20)[2021-08-01].https://arxiv.org/abs/1803.07422.
[14]	FANG Y C, LI Y F, TU X K, et al. Face completion with hybrid dilated convolution[J]. Signal Processing:Image Communication, 2020, 80: 115664. doi: 10.1016/j.image.2019.115664
[15]	HE X, YIN Y. Non-local and multi-scale mechanisms for image inpainting[J]. Sensors, 2021, 21(9): 3281. doi: 10.3390/s21093281
[16]	SHI Y, FAN Y, ZHANG N. A generative image inpainting network based on the attention transfer network across layer mechanism[J]. Optik, 2021, 242: 167101. doi: 10.1016/j.ijleo.2021.167101
[17]	胡凯, 赵健, 刘昱, 等. 结构引导的图像修复[J]. 北京航空航天大学学报, 2022, 48(7): 1269-1277. doi: 10.13700/j.bh.1001-5965.2021.0004 HU K, ZHAO J, LIU Y, et al. Structure-guided image restoration[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(7): 1269-1277(in Chinese). doi: 10.13700/j.bh.1001-5965.2021.0004
[18]	李清泉, 王欢, 邹勤. 一种基于稀疏表示模型的壁画修复算法[J]. 武汉大学学报(信息科学版), 2018, 43(12): 1847-1853. LI Q Q, WANG H, ZOU Q. A mural restoration algorithm based on sparse representation model[J]. Journal of Wuhan University (Information Science Edition), 2018, 43(12): 1847-1853(in Chinese).
[19]	REN Y R, YU X M, ZHANG R N, et al. StructureFlow: Image inpainting via structure-aware appearance flow[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 181-190.
[20]	LIU H Y, JIANG B, SONG Y B, et al. Rethinking image inpainting via a mutual encoder-decoder with feature equalizations[C]//Computer Vision-ECCV 2020. Berlin: Springer, 2020, 12347: 725-741.
[21]	LI X, WANG W H, HU X L, et al . Selective kernel networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 510-519.
[22]	HOU Q B, ZHANG L, CHENG M M, et al. Strip pooling: Rethinking spatial pooling for scene parsing[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 4002- 4011.