The murals of Yongle Palace in Ruicheng, Shanxi, are an essential part of the Yellow River cultural heritage and have received widespread attention and support both domestically and internationally. However, over time, these murals have experienced information loss and gradual color degradation due to natural factors and human interventions. This has not only weakened their cultural value but also limited their further development in academic research and practical applications. To address these issues, a digital restoration method for mural images based on patch matching and color restoration models was proposed. Traditional algorithms based on patch matching were first employed for preliminary structural restoration. Subsequently, a deep learning-based restoration method was applied to further restore the color of the murals. A multi-feature fusion module was introduced to extract features from both gradient and feature levels, addressing the shortcomings of feature extraction and improving the color restoration quality and visual effects of mural images. Experimental results demonstrate that the proposed method achieves high subjective and objective evaluation metrics, effectively compensates for missing information in mural images, and enhances the quality of color restoration.
LEI Tong, LIU Hui-jie, LI Yi-xuan, ZHANG Ying-yi, ZHOU Wen.
Digital Restoration of the Murals in Yongle Palace, Ruicheng[J].
Science Technology and Engineering, 2026, 26(9): 3867-3875 DOI:10.12404/j.issn.1671-1815.2502087
图像修复是利用算法对受损、缺失的图像进行恢复的过程,能够提高图像的质量和呈现效果。近年来,除应用于数字摄影、视频图像和医学影像等领域中[1],壁画图像领域也成为热点。早期修复技术以传统算法为主,主要包括三类:基于扩散的方法、基于样本块匹配的方法和基于稀疏表示的方法,如Bertalmio等[2]基于扩散的方法,将三阶偏微分方程用于图像修复中,并提出了沿图像修复边界的等照度线方向,进行各向异性扩散的模型;在此之后,后人又进行了一系列的改进,提出了多种基于扩散的修复方法[3-5]并不断优化,但在修复高频纹理和大面积缺损时常出现边界模糊和细节丢失的问题。基于样本块匹配的方法以Criminisi等[6]提出的方法为代表,这类型的方法[7-10]主要为采用图像中的完整区域,与图像中待修复区域边缘相匹配,选取匹配度最高的样本块进行修复,能较好保留纹理连续性,但面对结构复杂或大范围缺失的区域,匹配精度下降,易产生视觉违和。而基于稀疏表示的方法,能够同时兼顾保留图像更多的原始信息和处理图像中失真等状况,Olshausen等[11]首次将稀疏表示引入图像处理中,伴随着大量研究,当前基于稀疏表示的方法已有很大改进,如Arya等[12]基于小波块和基于组的稀疏表示学习,利用两者的优势在小波分解分量上使用自适应K-奇异值分解(K-singular value decomposition,K-SVD)创建多个字典;陈永等[13]则通过非下采样轮廓波变换优化传统稀疏表示方法以适配壁画结构与纹理的异质性,但此类方法普遍依赖高质量字典构建,计算复杂度高,对不同图像内容的适应性有限。
ChenWenxiang, TianQichuan, LianLu, et al. Research progress of image inpainting methods based on deep learning[J]. Computer Engineering and Applications, 2024, 60(22): 58-73.
[3]
BertalmioM, SapiroG, CasellesV, et al. Image inpainting[C]// Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques. New York: ACM, 2000: 417-424.
[4]
ShenJ, ChanT F. Mathematical models for local nontexture inpaintings[J]. Siam Journal on Applied Mathematics, 2002, 62: 1019-1043.
ShiXiaobo, LinSuzhen. Restoring method of the tomb fracture based on sample patch and BSCB model[J]. Science Technology and Engineering, 2017, 17(35): 98-104.
[7]
TheljaniZ M M. High-order anisotropic diffusion operators in spaces of variable exponents and application to image inpainting and restoration problems[J]. Nonlinear Analysis: Real World Applications, 2019, 47: 251-271.
[8]
CriminisiA, PerezP, ToyamaK. Region filling and object removal by exemplar-based image inpainting[J]. IEEE Transactions on Image Processing, 2004, 13: 1200-1212.
[9]
BarnesC, ShechtmanE, FinkelsteinA, et al. PatchMatch: a randomized correspondence algorithm for structural image editing[J]. ACM Transactions on Graphics, 2009, 28: 1531330.
[10]
CaoJ F, LiY F, ZhangQ, et al. Restoration of an ancient temple mural by a local search algorithm of an adaptive sample block[J]. Heritage Science, 2019, 7(1). DOI: 10.1186/s40494-019-0281-y.
[11]
YaoF. Damaged region filling by improved criminisi image inpainting algorithm for thangka[J]. Cluster Computing, 2019, 22: 13683-13691.
[12]
LiK S, WeiY S, YangZ, et al. Image inpainting algorithm based on TV model and evolutionary algorithm[J]. Soft Computing, 2016, 20: 885-893.
[13]
OlshausenB A, FieldD J. Emergence of simple-cell receptive field properties by learning a sparse code for natural images[J]. Nature, 1996, 381: 607-609.
[14]
AryaA S, SahaA, MukhopadhyayS. ADMM optimizer for integrating wavelet-patch and group-based sparse representation for image inpainting[J]. The Visual Computer, 2024, 40: 345-372.
ChenYong, ZhaoMengxue, TaoMeifeng. Mural inpainting algorithm for group sparse based on multi-scale contourlet transform decomposition[J]. Journal of Xidian University, 2022(6): 120-128.
[17]
WangN, ZhangY P, ZhangL F. Dynamic selection network for image inpainting[J]. IEEE Transactions on Image Processing, 2021, 30: 1784-1798.
[18]
WadhwaG, DhallA, MuralaS, et al. Hyperrealistic Image Inpainting with Hypergraphs[C]//2021 IEEE Winter Conference on Applications of Computer Vision (WACV). Montreal: IEEE, 2021: 3911-3920.
[19]
ZengY, LinZ, LuH C, et al. CR-Fill: generative image inpainting with auxiliary contextual reconstruction[C]// 2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal: IEEE, 2021: 14144-14153.
[20]
DongQ L, CaoC J, FuY W. Incremental transformer structure enhanced image inpainting with masking positional encoding[J]. ArXiv: 2203. 00867, 2022.
XuZhigang, YinWenyu, ZhuXufeng. Color restoration of mural image based on double constrained convolutional neural network[J]. Journal of Huazhong University of Science and Technology(Natural Science Edition), 2020, 48(6): 6-12.
[23]
HeM M, LiaoJ, ChenD D, et al. Progressive color transfer with dense semantic correspondences[J]. ArXiv: 1710. 00756, 2017.
[24]
GatysL A, EckerA S, BethgeM. Image style transfer using convolutional neural networks[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 2414-2423.
[25]
SongG H, WangH. Artificial intelligence-assisted fresco restoration with multiscale line drawing generation[J]. Complexity, 2021(4): 1-12.
[26]
LiJ, WangH, DengZ Q, et al. Restoration of non-structural damaged murals in Shenzhen Bao’an based on a generator-discriminator network[J]. Heritage Science, 2021, 9(1): 6. DOI: 10.1186/s40494-020-00478-w.
[27]
CaoJ F, ZhangZ B, ZhaoA D, et al. Ancient mural restoration based on a modified generative adversarial network[J]. Heritage Science, 2020(1). DOI: 10.1186/S40494-020-0355-X.
CaoJianfang, JinMengyan, LiZhaoxia, et al. Mural color restoration algorithm based on cyclic generative adversarial network[J]. Journal of Shandong University of Science and Technology(Natural Science), 2023, 42(4): 101-112.
JiangChao, LeiTong, YuLin, et al. The digital restoration method for the murals in the Yuan Dynasty Tomb of Hansenzhai, Xi’an[J]. Science Technology and Engineering, 2020, 20(31): 12922-12926.
[32]
ZhouZ H, LiuX R, ShangJ Y, et al. Inpainting digital dunhuang murals with structure-guided deep network[J]. Journal on Computing and Cultural Heritage, 2022, 15: 1-25.
[33]
NazeriK, NgE, JosephT, et al. EdgeConnect: structure guided image inpainting using edge prediction[C]// 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW). Seoul: IEEE, 2019: 3265-3274.
[34]
刘欣然. 敦煌数字壁画修复算法研究[D]. 广州: 华南理工大学, 2023.
[35]
LiuXinran. Research on digital dunhuang mural inpainting algorithm[D]. Guangzhou, China: South China University of Technology, 2023.
[36]
TomasiC, ManduchiR. Bilateral filtering for gray and color images[C]// Sixth International Conference on Computer Vision (IEEE Cat. No. 98CH36271). New Delhi: Narosa Publishing House: 839-846.
[37]
BertalmioM, BertozziA L, SapiroG. Navier-Stokes, fluid dynamics, and image and video inpainting[C]// IEEE Computer Society Conference on Computer Vision & Pattern Recognition. Kauai: IEEE, 2001: 990497.
[38]
LiuG L, RedaA F, ShihK J, et al. Lecture notes in computer science[M]. Berlin: Springer International Publishing, 2018.