基于改进生成式对抗网络与矢量绘制技术的古蜀锦纹样数字化研究  

Research on digitalization of ancient Shu brocade patterns based on improved generativeadversal network and vector rendering technology

作　　者：王维杰 刘毅 肖露[1] 方佳[1] 王金羽 王佳丽[1] WANG Weijie;LIU Yi;XIAO Lu;FANG Jia;WANG Jinyu;WANG Jiali(Sichuan Academy of Silk Sciences Co.,Ltd.,Chengdu 610031,China;College of Biomass Science and Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川省丝绸科学研究院有限公司,成都610031 [2]四川大学轻工科学与工程学院,成都610065

出　　处：《丝绸》2023年第11期18-27,共10页Journal of Silk

基　　金：四川省科技计划项目(2021YFS0348)。

摘　　要：目前古蜀锦纹样的数字化保存方法多基于数字照片的形式,其可编辑性差,不利于针对纹样的二次开发与应用,更有纹样因拍摄年代久远而导致分辨率低,从而导致在传播过程中损失关键信息。文章针对古纹样的原始图片分辨率低与位图图像的可编辑性差等问题,提出一种基于改进生成式对抗网络与矢量绘制技术的古蜀锦纹样的复原方法。以《四天王狩狮纹锦》为例,通过对原始图像的超分辨率重建与矢量化绘制完成矢量化建模。实验结果表明,本文方法可以实现针对古蜀锦纹样的高精度矢量化重建,纹样复原效果的模糊综合评价为比较好的隶属度,即为47.78%,说明了本文方法的有效性。通过该方法生成的古蜀锦矢量图像可以广泛应用于文旅产品开发与虚拟数字博物馆建设。同时,本文方法对其他类型古纹样的矢量化重建具有重要的借鉴意义。Silk textiles represent one of the most challenging categories of artifacts to preserve.Deterioration caused by bacteria and other microorganisms targeting fibroin proteins leads to the gradual loss of the original color and pattern characteristics of silk fabrics over time.Consequently,the imperative to effectively safeguard the patterns on silk textile artifacts is paramount.Taking Shu brocade as an example,inadequate conservation efforts directed towards ancient remnants of Shu brocade have resulted in the gradual disappearance of certain patterns,thereby negatively impacting the lineage of Shu brocade patterns.Historically,strategies for safeguarding ancient patterns can be categorized into traditional and modern methods.Among these,traditional approaches encompass the preservation of fragmentary patterns through techniques such as digital imaging or manual illustration.However,the lack of integration with image preprocessing technologies has led to varying degrees of pattern detail loss during the rendering process,undermining the authenticity of pattern propagation.Modern methods predominantly rely on digital image processing techniques and deep learning algorithms to address issues associated with the damage,blurriness,noise,and low resolution inherent in ancient patterns.These methods also involve the computer-generated production of vector outlines for the patterns.Nevertheless,there is currently a dearth of automated completion algorithms tailored to intricate patterns.Additionally,the utilization of conventional digital image processing techniques falls short in achieving super-resolution reconstruction of degraded images.Limited research has hitherto employed deep learning-based algorithms to realize super-resolution reconstruction of patterns on silk textile artifacts.To address the challenges posed by the current preservation techniques for ancient patterns,including issues of poor applicability to complex patterns and unsatisfactory results in super-resolution reconstruction,we presented a digital

关 键 词：生成式对抗网络 超分辨率重建 古蜀锦纹样 矢量绘制 数字化保护 

分 类 号：TS941.2[轻工技术与工程—服装设计与工程]