摘要
图像修复算法多依赖先验信息推断并恢复缺损区域,对于无参考信息下结构复杂的非规则图案,容易出现结构线断联和块效应现象。为此,本文提出一种基于结构线拟合的无参照丝织品文物图像虚拟修复算法。在缺损区域图案信息修复总体框架下,首先对丝织文物图像进行超分辨率重建,获取细节较为清晰的图像;其次,对图像待修复区域进行掩膜处理,在交互引导下,采用反距离加权法插值获取缺损轮廓边界关键点像素颜色值,利用Bezier曲线连通纹理结构线;最后,利用改进优先权的图像修复算法对纹理合成填充待修复区域进行迭代修复,直至完成缺损区域内容修补。实验结果表明,文物图像中纹样结构上基本保持了原图的路径趋势,能够完好地对内容进行修复。修复结果取得的SSIM均值为0.98,PSNR均值为36.76,高度再现了丝织品图案原貌。
Due to the long-term erosion in moist and microbial environment,silk cultural relics are very fragile and prone to be damaged,and the surface patterns become faded,aging,and dirty,which is not conducive to the connotation excavation and value dissemination of the relics.Compared with contact body repair,virtual restoration based on the image inpainting technology can reduce the risk of physical restoration,and it is commonly used for online collection restoration display,ontology restoration scheme reference,etc.Image inpainting algorithms rely on prior information to infer the information of the defective area,so repair is very challenging when the image has little reference information.However,as most silk cultural relics are unique and their patterns are irregular,there are structural line disconnection and block effect when they are repaired by the existing image restoration algorithms.In order to make repaired images more realistic and improve the digital display effect of relics,an image restoration method based on pattern structure line fitting is proposed to realize the virtual restoration of silk cultural relics.Firstly,silk cultural relic images are reconstructed in super-resolution to obtain sharp details.Besides,the defective areas are masked under the guidance of experts.Secondly,the color values of key points on the structure lines to be fitting are obtained interactively and the missing pattern structure lines are reconstructed.Finally,the Criminisi algorithm based on improved priority is used to iteratively repair until the information of the defective areas is all filled.This method differs from existing methods in that it reconstructs the structure lines according to the pattern design and guides the image inpainting.In this study,Bezier curve is utilized to form the structure lines.The inverse distance weighted interpolation method is introduced to color the pixel points on the structure lines,and the control points are adaptively decided by the slope formula to fit the missing structural lines.Further,the boundary function is introduced to improve the priority,which changes the priority of the patch to be repaired to avoid error repair and block effect.Results show that the pattern structure of the cultural relics basically maintains the path trend of the original image and can completely repair the missing content.The average SSIM and PSNR obtained from the restoration results are 0.98 and 36.76,respectively,which highly reproduces the original appearance of the silk fabric pattern.Virtual restoration,as a supplement to physical restoration,on the one hand,assists the physical restoration of cultural relics,which can provide reference for the effect of physical restoration of cultural relics.On the other hand,it provides technical support for online museum virtual display.The method in this paper is suitable for images with obvious structural trend in the missing area.In view of the problem of automatic repair in the state of blind repair with few reference samples,it is difficult to generate images with complex patterns and insignificant structural trend based on the peripheral information.In the future,it is necessary to focus on the interactive adjustment of the parameters to meet the multiply recommended recovery schemes needs.
作者
孙晓婉
贾静
徐平华
曹竟文
李碧
SUN Xiaowan;JIA Jing;XU Pinghua;CAO Jingwen;LI Bi(School of Fashion Design&Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China;Zhejiang Provincial Research Center of Fashion Engineering Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China;MOC Key Laboratory of Silk Culture Heritage and Product Design Digital Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China;School of Art and Design,Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《丝绸》
CAS
CSCD
北大核心
2023年第1期1-8,共8页
Journal of Silk
基金
国家自然科学基金青年基金项目(61702460)
国家社会科学基金重点项目(19AMZ009)
浙江理工大学科研业务费专项资金资助项目(2021Q057)
服装设计国家级虚拟仿真实验教学中心项目(zx20212004)
浙江理工大学教育教学改革研究重点项目(jgzd202202)
浙江省服装工程技术研究中心开放基金项目(2021FZKF05)
浙江省教育厅科研基金项目(Y202250618)
浙江理工大学优秀研究生学位论文培育基金项目(LW-YP2021053)。
关键词
曲线拟合
虚拟修复
图像重建
文物修复
丝织品
图像补全
curve fitting
virtual restoration
image reconstruction
cultural relic restoration
silk fabric
image inpainting
