基于高光谱技术的古书法文物隐藏字迹识别  

Hidden Handwriting Recognition of Calligraphy Artifact Based on Hyperspectral Technology

作　　者：高宇 孙雪剑 李广华 张立福[1,3] 曲亮 张东辉 GAO Yu;SUN Xue-jian;LI Guang-hua;ZHANG Li-fu;QU Liang;ZHANG Dong-hui(National Engineering Laboratory for Satellite Remote Sensing Applications,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China;University of Chinese Academy of Sciences,Beljing 100049,China;China-Greece Belt and Road Joint Laboratory on Cultural Heritage Conservation Technology,Beijing 100009,China;The Palace Museum,Beijing 100009,China;Institute of Remote Sensing Satellite,China Academy of Space Technology,Beijing 100095,China)

机构地区：[1]中国科学院空天信息创新研究院,遥感卫星应用国家工程实验室,北京100094 [2]中国科学院大学,北京100049 [3]中国-希腊文物保护技术“一带一路”联合实验室,北京100009 [4]故宫博物院,北京100009 [5]中国空间技术研究院,卫星遥感总体部,北京100095

出　　处：《光谱学与光谱分析》2024年第12期3485-3493,共9页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2022YFF0904400,2020YFE0204600)资助。

摘　　要：古书法文物中的字迹内容是记载中国古人社会文明的重要信息,获取并识别其纸面中的隐藏字迹对于文物的历史背景研究和人文典籍搜集具有重要的意义。然而,传统的书法文物隐藏字迹识别方法依赖于人工经验判读,对研究人员的经验积累要求较高,且分析过程耗时较长,易对文物产生二次损伤。针对这一困难,高光谱遥感凭借非接触、高效的特点,在采集纸质文物表面空间属性的同时,也获取了其文物丰富的光谱信息,从光谱维实现了书法文物的数字化存储。首先,使用最小噪声分离(MNF)变换发现了书法文物中隐藏的模糊信息,设计出了一种基于线性差异增强(LDE)的光谱变换方法对其进行进一步识别,并统计其变换前后的光谱特征参量、以及MNF提取的成分图像,采用熵评价以获取其中信息丰度最高的光谱图像,有效识别出了该文物的隐藏字迹内容。研究结果表明:(1)书法文物高光谱影像在进行MNF变换处理后,得到的第一信息成分图像中显示出了隐藏在文物字迹中的模糊图案,其光谱形态与该文物其他内容要素之间存在相似性,但幅值存在差异;(2)通过LDE算法能有效扩大文物中隐藏信息与字迹光谱之间各个波段的相对差异,并且增强后隐藏字迹的大部分光谱特征也得到了显著增强;(3)文物数据经过LDE处理后,分别在波长范围、光谱特征和MNF子成分中的图像熵值得到了一定程度的提升,其中以LDE处理后的光谱方差(SV)图像的熵值最高,为6.74 bit;(4)在文物经过LDE处理后SV图像中,有效识别出隐藏在文物中的字迹内容,为不属于《心经》中的佛经,证明了该文物所用纸张为写经纸。研究结果有效发现并识别出了隐藏在乾隆墨宝背后的隐藏字迹,丰富了该文物的历史人文背景,为后续古书法文物中的隐藏字迹信息提取提供了科学的理论与技术支撑。The content of ancient calligraphy artifacts contains crucial information documenting the social civilization of ancient China.Obtaining and identifying hidden inscriptions within the paper surface is important for historical background research on cultural relics and the collection of humanistic classics.However,traditional methods for identifying hidden inscriptions in calligraphy artifacts rely heavily on manual interpretation,which requires extensive expertise from researchers and results in a time-consuming analysis that may inadvertently cause secondary damage to the artifacts.To address this challenge,hyperspectral remote sensing,with its non-contact and efficient characteristics,captures the spatial properties of paper-based artifacts and acquires rich spectral information.This enables the digitization and storage of calligraphy artifacts.Initially,the Minimum Noise Fraction(MNF)transformation technique was utilized to reveal latent blurry information in calligraphy artifacts.Subsequently,a spectral transformation method based on Linear Difference Enhancement(LDE)was developed to identify these details further,and statistical analysis was conducted on the spectral parameters before and after transformation and the component images extracted by MNF.By utilizing entropy evaluation,we obtained the most information-rich spectral image,ultimately enabling the successful identification of the hidden inscriptions within the calligraphy artifact.The research results demonstrate the following:(1)The MNF transformation of the hyperspectral image of the calligraphy artifact reveals the blurred patterns hidden in the inscriptions.These patterns exhibit similarities in spectral morphology with other content elements of the calligraphy artifact but with differences in reflectivity.(2)The LDE algorithm effectively amplifies the relative differences between hidden information within the calligraphy artifact and the spectral bands of the inscriptions.LDE significantly enhances most spectral features of the hidden inscript

关 键 词：书法文物 高光谱遥感 隐藏字迹 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]