新书推荐|《中世纪手稿的数字化:圣查德福音书的物质性、复原与2D&3D再现》
2026-06-12 14:25 湖北
![]()
Bill Endres 的Digitizing Medieval Manuscripts: The St. Chad Gospels, Materiality, Recoveries, and Representation in 2D&3D 以八世纪 St. Chad Gospels 为核心个案,讨论中世纪手稿数字化如何服务于文本恢复、物质性研究、保存实践与知识再呈现。
一
内容介绍
![]()
Bill Endres 的《中世纪手稿的数字化:圣查德福音书的物质性、复原与2D & 3D再现(Digitizing Medieval Manuscripts: The St. Chad Gospels, Materiality, Recoveries, and Representation in 2D & 3D )》以八世纪 St. Chad Gospels 为核心个案,讨论中世纪手稿数字化如何服务于文本恢复、物质性研究、保存实践与知识再呈现。全书的研究目的在于说明,数字图像并非单纯的复制品,而是一种能够揭示手稿历史层累的研究工具。作者关注羊皮纸、墨水、颜料、磨损、擦除、褪色、装订变形等物质痕迹,试图通过光学成像与后期处理,重新读取肉眼难以辨识的信息,并由此推进古文献学、书籍史和数字人文之间的对话。
本书的创新点首先体现在方法整合上。作者将多光谱成像、替代色彩空间、反射变换成像、历史照片比对和三维呈现结合起来,展示不同技术如何分别处理文字恢复、表面细节观察、颜料损耗追踪和页面形态分析等问题。其次,作者强调开放源代码软件和可操作流程,使高级成像不只属于大型实验室,也能进入普通中世纪研究者的工作方法。最后,本书把技术问题放回手稿的宗教、文化和伦理语境中,讨论文物、开放获取、社区互惠和数字授权等问题,避免把数字化理解为纯粹技术工程。
从研究方向看,本书属于数字手稿学、计算古文献学和物质文本研究的交叉领域。它既面向损毁文本的恢复,也面向手稿长期保存和数字替代物的批判评估。作者尤其关心二维图像的限制,并进一步探索三维成像和虚拟现实作为新型知识空间的可能性。其价值在于提醒研究者,数字化手稿研究的核心并非追求更清晰的图像本身,而是通过图像、算法和物质性分析,重新理解手稿作为历史物、宗教物和知识媒介的复杂生命。
二
作者介绍
![]()
Bill Endres
比尔·恩德雷斯(Bill Endres),数字人文学者,2009年获亚利桑那州立大学博士学位,现任美国俄克拉荷马大学英语系副教授。其核心研究方向是用数字技术挖掘中世纪早期手稿的物质性,不只关注文字内容,更聚焦纸张、刻印、笔迹等肉眼难以察觉的实体特征。至今仍保留着诗歌与修辞学方向的研究兴趣,这也是其早期学术背景的重要组成部分。
研究主线:用数字技术挖掘中世纪早期手稿的物质性,不只关注文字内容,更聚焦纸张、刻印、笔迹等肉眼难以察觉的实体特征。
研究对象:主攻英国利奇菲尔德大教堂藏的公元 8 世纪《圣查德福音书(St Chad Gospels)》,同时覆盖公元 600-850年不列颠群岛产出的所有岛屿手稿(Insular manuscripts)。
研究技术:运用多光谱成像、反射变换成像和高精度 3D 捕获技术开展工作;其中 RTI 可识别无墨干刻的隐形文字,3D 扫描精度达 0.05 毫米,还会通过对比不同年代的数字化图像追踪手稿老化情况。
学术成果:出版专著《Digitizing Medieval Manuscripts: The St Chad Gospels, Materiality, Recoveries, and Representation in 2D & 3D》,并配套发布问答拓展文章《Collecting Light: Q&A with Bill Endres》;搭建学术网站Manuscripts of Lichfield Cathedral,免费开放《圣查德福音书》和《威克利夫圣经(Wycliffe Bible)》的知识共享资源;开发了《圣查德福音书》VR 体验测试版。
学术著作
专著
Digitizing Medieval Manuscripts: The St Chad Gospels, Materiality, Recoveries, and Representation in 2D & 3D, Arc Humanities Press, 2019.
期刊论文
“Digitization as Scholarly Intervention and Interpretive Act: A Case for 3D Capture in Studying the Agency of Materiality,” Manuscript Studies, 2025.
“The St. Chad Gospels: Ligatures and the Division of Hands,” Manuscripta, 2015.
“Imaging Sacred Artifacts: Ethics and the Digitizing of Lichfield Cathedral's St Chad Gospels,” Journal of Religion, Media & Digital Culture, 2014.
书籍章节与会议论文
(与Jenna Gorlewicz合著) “Haptics and Digital Codicology: Reimagining Tactile Experience in the Study of Manuscripts,” 收录于 Approaches to Digital Codicology: Interdisciplinarity and Intersections, 2025.
“A Manuscript Wanders into VR: Oh My!,” 收录于Insular Art at the Crossroads: Ninth International Insular Art Conference 2022, 2025.
“Oh Lord, make haste to help me: Prayer and Imagery Atop Canon Table II in the Book of Kells,” 收录于An Insular Odyssey: Manuscript Culture in Early Christian Ireland and Beyond, 2017.
“A Literacy of Building: Making in the Digital Humanities,” 收录于Making Humanities Matters, Debates in the Digital Humanities Series, 2018.
“More than Meets the Eye: Going 3D with an Early Medieval Manuscript,” 收录于Proceedings of the Digital Humanities Congress 2012, 2014.
三
出版信息
作者:Bill Endres
出版社:Arc Humanities Press
出版时间:2019年(电子书)/2023年(实体书)
丛书系列:Medieval Media and Culture
ISBN:9781802701227
四
书籍目录
导论
视觉奇观的时代——物质性的数字化与光之解密
第一章
复原:从多光谱成像到替代色彩空间
第二章
反射变换成像:表面细节的增强观察
第三章
原本不可知之物:历史照片的数字化与比较
第四章
神圣器物:开放获取、权力、伦理与互惠
第五章
知识空间的危机?——虚拟现实之展望
![]()
五
章节简介
导论
视觉奇观的时代——物质性的数字化与光之解密
The Age of Visual Wonder: Digitizing Materiality and Unriddling Light
导论为全书建立理论框架和问题意识。作者从手稿随时间变化的物质状态谈起,指出羊皮纸、墨水、颜料、磨损、擦除、水损和褪色都使手稿成为有待重新辨读的历史对象。以 St. Chad Gospels 为核心案例,导论说明该手稿不仅保存了福音书文本,也保存了关于制作地点、书写者、艺术家、颜料层、边注、旅行经历和损毁过程的多重线索。作者的基本关切是,数字化不能停留在图像复制和在线开放层面,还应进一步进入手稿物质性分析。通过将视觉信息转化为数值,研究者可以运用数学处理、光谱成像和三维呈现来发现肉眼难以辨认的内容。导论同时批判二维图像的局限,强调手稿页面并非平面图像,而是具有厚度、弯曲、反光、纹理和损耗轨迹的三维物体。作者由此提出全书的主轴,即以数字成像技术恢复和保存手稿物质性,并在技术、伦理和知识空间三个层面重新理解中世纪手稿。
第一章
复原:从多光谱成像到替代色彩空间
Recovery: From Multispectral Imaging to Alternative Colour Spaces
第一章集中讨论受损文本和图像的数字恢复方法。作者首先回顾早期使用化学试剂恢复褪色文字的传统,指出这类做法常常对手稿造成不可逆损害。随后,作者转向以光学性质为基础的非侵入式恢复路径,重点说明多光谱成像如何通过紫外、可见光和红外等不同波段捕捉页面反射信息。以 St. Chad Gospels 中水损文字、擦除文字和颜料残迹为例,作者展示了不同材料对不同波长光线的吸收和反射差异。铁胆墨、羊皮纸、碳基墨水和颜料层在光谱中的表现并不相同,恢复效果必须根据具体物质条件判断。本章的一个重要贡献是将高级成像与可操作的后期处理结合起来。作者讨论 ImageJ、替代色彩空间、数学运算、假彩色和去相关拉伸等方法,说明普通彩色图像甚至手机照片也可能通过适当处理产生有价值的恢复结果。此章的核心方向是建立一种可推广的恢复流程,使中世纪研究者能够理解图像、光线和材料之间的关系,并据此选择合适的数字方法。
第二章
反射变换成像:表面细节的增强观察
Reflectance Transformation Imaging: An Enhanced View of Surface Details
第二章讨论反射变换成像在手稿表面细节研究中的作用。与第一章重在恢复受损文字不同,本章关注页面表面起伏、压痕、刻写、墨迹隆起和颜料层状态。作者特别强调干笔书写的重要性。干笔书写通常用尖笔在页面上刻出痕迹,没有墨水,因此在普通光照下极难观察。RTI 通过从多个方向记录光线照射下的表面反应,使研究者能够在数字界面中改变虚拟光源方向,从而凸显细微凹凸。作者采用 highlight RTI 方法,以手持闪光灯模拟半球光源装置,这使该技术更易于实施,也降低了设备门槛。本章不仅说明 RTI 可用于辨认难读的干笔文字,还展示其对格线、压印、墨水和颜料隆起、页面局部损伤等问题的分析价值。其研究意义在于把手稿表面从被动背景转化为可读信息来源。页面的微小物理变化可以揭示书写过程、校订方式、使用痕迹和保存状态,因此 RTI 为古文献学、书籍史和保护研究提供了更细密的观察手段。
第三章
原本不可知之物:历史照片的数字化与比较
The Otherwise Unknowable: Digitizing and Comparing Historical Photographs
第三章把历史照片纳入数字手稿研究,讨论早期摄影资料在追踪手稿老化和保护史方面的价值。作者指出,新的数字图像并不必然比旧照片包含更多历史信息,因为手稿在过去一百多年中可能已经失去颜料、文字痕迹或页面形态。早期照片记录了手稿在某一时刻、某种光照条件和某种摄影技术下呈现出的状态,因此可以成为分析长期变化的证据。本章以 St. Chad Gospels 的 1887 年照片、photostat 副本、彩色幻灯片、黑白照片和 2010 年多光谱成像项目为材料,说明如何将历史照片数字化、对齐、叠放和调节透明度,以观察颜料脱落、页面压平、边缘变化和过去修复处理的影响。作者也提醒读者,摄影本身具有技术限制,不同相机、镜头、光源和处理方式都会影响图像结果。因此,比较历史照片既能产生关于手稿老化的新知识,也能训练研究者批判性地看待图像证据。本章的研究方向是把摄影档案转化为时间序列材料,使数字化成为研究手稿生命史和保存状态的工具。
第四章
神圣器物:开放获取、权力、伦理与互惠
Sacred Artefacts: Open Access, Power, Ethics, and Reciprocity
第四章转向数字化的人文和伦理维度,讨论开放获取、权力关系、版权、社区利益和互惠原则。作者认为,对于仍然服务于特定宗教共同体的手稿,研究者不能只把它看作文物资源或文化遗产。St. Chad Gospels 与 Lichfield Cathedral 的宗教生活、地方记忆和共同体身份相连,因此数字化工作本身也是一种社会关系。作者借用民族志中的互惠原则,强调研究者在追求开放获取和学术成果时,需要同时考虑保存机构、宗教共同体和公众使用者的需求与担忧。本章讨论 Creative Commons 授权、数字图像传播、开放数据的实际门槛、社区参与和长期信任关系等问题。作者指出,开放获取并不自动等于公平使用,因为高分辨率图像、多光谱数据和专业软件仍可能构成技术壁垒。真正有效的数字化应提供不同层次的访问方式,使普通观众、地方共同体和专业学者都能获得相应收益。本章的重点在于把数字人文从技术执行推进到责任实践,说明图像开放必须与伦理判断、社区协商和知识回馈结合起来。
第五章
知识空间的危机?——虚拟现实之展望
A Crisis in Knowledge-Space? A Look Toward Virtual Reality
第五章讨论三维成像、虚拟现实和知识空间问题,是全书从图像处理走向媒介理论的一章。作者认为,二维照片虽然极大扩大了手稿访问范围,却容易削弱研究者对手稿三维属性的理解。手稿页面有弯曲、厚度、纹理、颜料隆起和装订约束,阅读手稿也包含身体姿态、距离、光线和触觉想象等经验。虚拟现实和三维渲染可以在一定程度上恢复这种空间感,使手稿重新作为可被环绕、接近和观察的物质对象出现。作者提出知识空间这一概念,用来理解不同媒介如何组织知识的生产、保存和传播。他梳理了从希腊辩证法、中世纪手稿、印刷书籍到 Web 2.0 的历史变化,说明每一种媒介环境都会塑造特定的认知方式。虚拟现实因此被视为新的知识空间,既提供沉浸式研究和教学的可能,也带来技术门槛、真实性、解释控制和感官替代等问题。本章的研究方向是探索数字手稿的未来形态,尤其是三维技术如何帮助研究者重新思考物质性、临场感和知识传递。
六
章节选读
第一章(节选)
水损与磨损文本的修复
Recovery of Water-D amaged and Worn Text
如前所述,第 90 页在修复水损文本时面临诸多挑战。为评估相关方法与成像技术,我将相同方法应用于两张 RGB 图像,一张是大英图书馆于 2003 年通过高分辨率彩色相机拍摄,另一张是 Bill Endres 于 2010 年通过单色相机拍摄。我将示例限定在前两行的开头部分,因为这部分能有效呈现各类损伤情况。不过将两张图像并列对比后,能发现一处细微但关键的初始差异,2010 年图像中的羊皮纸看起来更偏灰色。对两张图像的颜色直方图进行分析后,这一观察结果得到了证实。按比例来看,2010 年的图像包含更多灰色,这是由红、绿、蓝频率的更大重叠所导致的。这一差异会对修复工作产生影响。
As mentioned, page 90 presents a range of challenges for recovering water-damaged text. To assess methods and imaging, I apply the same methods to two RGB images: one taken by a high-resolution colour camera by the British Library in 2003, and the other by a monochrome camera by Bill Endres in 2010. I limit examples to the beginning of the first two lines because they provide an efficient range of damage. Placing the images side-by-side, however, reveals a subtle but important beginning difference: the parchment in the 2010 image appears greyer. Examining colour histograms for each image affirms this observation. Proportionally, the 2010 image contains more grey, produced by a larger overlap of red, green, and blue frequencies. This difference affects recovery.
数码摄影并非一门精确的科学。不同的数码图像,其像素值存在差异,这会导致直方图的形状发生变化。对于 2010 年的图像,红、绿、蓝的绘图值范围更宽,进而影响了频率的重叠。直方图的谷值与峰值形状也略有不同。出现这类差异的原因在于,即便是高端相机,其传感器的灵敏度以及校准软件也存在区别。滤镜同样会造成差异。彩色数码相机配备拜耳滤镜(Bayer filter),它会阻挡所有频率,只保留每个光电探测器所需的红、绿、蓝频率。这些滤镜的灵敏度存在波动,因此需要不同的校准软件来合并频率,生成符合人眼视觉的彩色图像。不过对于 MSI 而言,当使用 LED 照明时,滤镜并非必要,这就消除了一个重要的变量。因此,不同相机拍摄的图像,所记录的反射光数值也不同。这不仅会导致直方图存在差异,更重要的是,它会使得数码修复的效果出现波动。
Digital photography is not an exact science. For different digital images, the values for pixels vary. This causes the shapes of histograms to vary. For the 2010 image, the graphed values for red, green, and blue are wider, affecting overlap. Valleys and peaks show slightly different shapes. Such differences occur because even in high-end cameras, the sensitivity of sensors and their calibrated software differ. But filters also cause variations. A colour digital camera has a Bayer filter blocking all frequencies except the red, green, or blue desired for each photodetector. The sensitivity of these filters fluctuates, requiring differently calibrated software to merge frequencies and produce a colour image that is realistic to the human eye. For MSI, when LED lighting is used, however, filters are unnecessary. This eliminates a significant variable. Therefore, different cameras produce images that record different values for reflected light. This causes histograms to differ, but importantly, it causes the success of digital recovery to vary.
对于第 90 页,使用 MSI 针对非可见频率拍摄的图像,最多只能实现有限的修复。紫外线光能够带来最佳效果,它能提升铁胆墨水(iron gall ink)书写的字母残留的可见度,这种墨水中的铁会吸收紫外线,而周围的羊皮纸则会反射更多紫外线。此外,由于羊皮纸反射的紫外线比渗透到透印文本中的更多,透印现象会逐渐消退。不过,水损导致变暗的羊皮纸会削弱这一积极效果。变暗的羊皮纸会吸收更多紫外线,因此对比度的提升十分微弱。虽然修复效果有限,但这些结果显示出像素值存在比例差异。这说明将可见与非可见频率的图像进行除法运算,可以增强墨水的痕迹,显现出受损文本的形态,这对于目前仍充满谜团的第一行文本而言意义重大。
For page 90, images taken with MSI for nonvisible frequencies produce, at best, minor recoveries. Ultraviolet light generates the best results. It increases the visibility of remnants of letters written in iron gall ink, the iron absorbing it while surrounding parchment reflects more of it. Furthermore, because parchment reflects more ultraviolet light than penetrates to the bleed-through text, bleed-through fades. However, parchment darkened by water-damage diminishes this positive gain. Darkened parchment absorbs more ultraviolet light. Contrast, therefore, increases only faintly. While recoveries are marginal, they indicate proportional differences for values of pixels. This suggests that dividing images of visible and nonvisible frequencies can enhance traces of ink and reveal patterns for damaged text, significant for the first line, still riddled in mystery.
在确认该墨水并非碳基墨水后,我们发现红外频率无法提供额外信息。例如,850 nm 的图像会让受损文本的残留痕迹消失,其更长的波长会穿透这些痕迹,反而将修复工作引向错误的方向。如前所述,铁胆墨水并不会像碳基墨水那样吸收红外光,因此这类频率最初无法提升对比度。不过,红外频率能够捕捉到比受损墨水更多的透印信息,这就带来了新的机会。通过数学运算,它们提供了另一种生成对比度的可能方式。
Confirming that ink is not carbon-based, infrared frequencies provide no additional information. For example, the image for 850 nm makes remnants of damaged text disappear. Its longer wavelengths pass through them, moving recovery in the wrong direction. As mentioned, iron gall ink does not absorb infrared light as carbon-based inks do; consequently, they are initially unbeneficial for increasing contrast. However, the ability of infrared frequencies to capture more bleed-through than damaged ink presents an opportunity. Through mathematical operations, they provide another possible way to generate contrast.
为尝试进行修复,并为数学运算筛选合适的频率,我将 RGB 多光谱图像分别除以其 638 nm 红、535 nm 绿和 465 nm 蓝通道。这一操作立刻得到了结果。这三个通道都大幅降低了透印带来的干扰,不过除以 465 nm 后,第一行的受损文本显现得最为清晰,即 tori suo。Latin Vulgate 证实了这一修复结果。第 90 页的内容是对 Matthew 20:8 的收尾,这则寓言讲述了工人们在不同时间开始工作,最终却获得相同报酬的故事。武加大译本中对应的预期词汇是 procuratori suo。第 89 页的结尾是 procura,因此 tori 补全了这个单词。suo 在三个结果中都清晰可见,在除以 465 nm 的 RGB 图像中最为明显。通过进一步的除法运算,465 nm 展现出了优化修复效果的潜力,也可用于修复其他存在水损文本的页面。不过 638 nm 和 535 nm 的结果同样有潜力,它们在不同的组合中可能会发挥作用。因此,MSI 拍摄的图像为修复工作提供了大量的可能性。不过,将 RGB 多光谱图像除以非可见频率的尝试并未成功。运算结果显示,850 nm 的像素值与 RGB 图像的像素值比例相似,完成除法后,受损文本的痕迹与透印现象都没有明显变化。在光谱的另一端,除以 365 nm 紫外线的结果同样令人失望,透印的干扰依然过强。不过受损文本显现为明亮的红棕色,这说明找到合适的配对进行除法运算或许能得到不错的结果。
To attempt recovery and identify further frequencies for mathematical operations, I divided the RGB multispectral image by its 638 nm (red), 535 nm (green), and 465 nm (blue) channels. This produced immediate results. All three dramatically decrease the interference caused by bleed-through; however, dividing by 465 nm makes the damaged text on the first line emerge most clearly: tori suo. The Latin Vulgate affirms this recovery. Page 90 begins by concluding Matthew 20:8, a parable about labourers starting their work at fluctuating times but all receiving identical pay. The expected Vulgate words are procuratori suo (his steward). Page 89 ends with procura; therefore, tori completes this word. The suo is evident in all three results, most clearly in the RGB image divided by 465 nm. Through further division, 465 nm shows promise for refining recoveries or producing them for other pages with water-damaged text. However, results from 638 nm and 535 nm also show promise. They could prove effective in different combinations. Therefore, images from MSI provide an excess of possibilities for producing recoveries. Dividing the RGB multispectral image by nonvisible frequencies, however, proved unsuccessful. Division demonstrates that 850 nm has proportionally similar values for pixels as the RGB image. Once divided, traces of damaged text and bleed-through remain relatively the same. At the other end of the spectrum, dividing by 365 nm (ultraviolet) produces likewise disappointing results. Bleed-through still interferes too strongly. However, the damaged text emerges as a bright reddish brown; therefore, finding the appropriate division partner might produce good results.
对于所有修复后的文本,生成伪彩色可以提升对比度。ImageJ 提供了许多颜色表(Look-Up Tables,luts)来实现这一功能。对于受损文本,伪彩色能够提升除以 465 nm、535 nm 和 638 nm 后的图像清晰度,对紫外线图像也有一定的改善作用。如前所述,应用 luts 需要进行实验,同一个 lut 并非总能为每一次修复都提升清晰度。
For any recovered text, generating false colour can increase contrast. ImageJ provides a number of luts for doing so. For the damaged text, false colour enhances clarity for dividing by 465 nm, 535 nm, and 638 nm, and some improvement for ultraviolet. As mentioned, applying luts requires experimentation. The same lut does not always add clarity to each instance of recovery.
无论是 e-Codices 还是 Digital Walters,随着可用的高分辨率手稿图像数量不断增加,中世纪研究者获得了前所未有的机会来深度修复文本内容。这也使得将 RGB 图像拆分为红、绿、蓝三个通道成为了一种极为重要的方法。拆分 RGB 图像本身通常无法直接完成修复,它往往只是第一步。例如,当我拆分大英图书馆拍摄的照片时,得到的三个通道并没有立刻呈现出结果。但将它们与彩色图像进行除法运算后就有了变化。三个通道的运算结果中,透印现象都有所减轻,其中除以红色和蓝色通道的两个结果中,出现了修复效果,分别是 tori 中的 to 和 suo 中的 su。生成伪彩色进一步优化了这些结果,在除以蓝色通道的结果中,我还修复出了 tori 中的 ri。
Whether e-Codices or the Digital Walters, because of the large and growing number of available high-resolution images of manuscripts, medievalists have unprecedented opportunities for recovering profound levels of content. This makes splitting RGB images into their red, green, and blue channels a highly significant method. Splitting RGB images tends not to generate recoveries in their own right. It tends to be the first step. For example, when I split the photograph taken by the British Library, the resulting three channels did not immediately provide results. However, dividing them into the colour image does. In all three, bleed-through lessens; in two, dividing by red and blue, recoveries occur: to of tori and su of suo. Generating false colour enhances these results. For dividing by blue, it leads to recovering ri of tori.
必要时,这些修复结果可以指导后续的数学运算。对于第 90 页的开头部分,虽然目前的结果已经可靠,但如果其他页面的字母更难修复,对单个频率进行除法运算或许能得到更精细的修复效果,并提供特定的处理方法。因此,在已有良好结果的基础上,比如除以 465 nm 的结果,进行下一步操作是合理的。由于 638 nm 的结果是第二好的,我也使用了这个频率,它与 465 nm 的差距足够大。对于铁胆墨水,我更倾向于用较高波长除以较低波长,这会让羊皮纸比修复后的文本更暗,我发现明亮的修复痕迹更容易辨认。不过作为测试,我也将它们反转了,看看更亮的羊皮纸是否能让结果更清晰。
When necessary, results from these recoveries can guide further mathematical operations. For the beginning of page 90, while the results are secure, dividing individual frequencies might produce refined recoveries and provide specific approaches if letters on other pages prove more resistant. Building off good results, such as dividing by 465 nm, therefore, provides a next logical step. Since 638 nm produced the second-best results, I also use this frequency. It provides a nice distance from 465 nm. For iron gall ink, I prefer to divide higher wavelengths by lower ones. This causes parchment to be darker than recoveries. I find bright recoveries more discernible. As a test, however, I reverse them: in case lighter parchment clarifies results.
对频率进行除法运算后,得到的结果通常是页面偏黑,可能只有微弱的可辨认文字行痕迹。对于第 90 页,当用 638 nm 除以 465 nm 时就是这种情况。不过调整直方图后,文本就显现出来了。此时像素值的范围大约在 0 到 3.4804 之间,当将这些值拉伸到完整的 0 到 255 范围后,文本就显现出来了。查看前两行的开头部分,可以发现修复效果十分显著。
Dividing frequencies normally produces results in which pages appear blackish, with perhaps faint hints of discernable lines of script. For page 90, this is the case when dividing 638 nm by 465 nm. Adjusting the histogram, however, reveals the text. In this case, values for pixels range from about 0 to 3.4804. When these values are stretched to cover the full range (0 to 255), the text emerges. A view of the beginning of the first two lines reveals strong recovery.
进一步的实验持续带来了良好的结果,将可见频率互相进行除法运算,都能得到修复效果。例如,用 535 nm 除以 505 nm,以及用 450 nm 除以 592 nm,都能清晰呈现受损文本。在第二个例子中,为了展示修复后的文本呈现为更暗的状态,我用较高频率作为除数。不过我还是更偏好更亮的文本。但羊皮纸的瑕疵或变色可能会让其中一种方式更有效。
Further experimenting continued to produce good results. Dividing visible frequencies into each other all generated recoveries. For example, dividing 535 nm by 505 nm and 450 nm by 592 nm provide good views of the damaged text. In the second example, to demonstrate recovered text appearing darker, I divide by the higher frequency. Again, lighter text is my preference. However, blemishes or discolouration of parchment can cause one or the other to be more beneficial.
紫外线频率的效果依然没有达到预期。将其他各个频率分别除以 365 nm 后,只有一个图像得到了不错的修复效果,即用 850 nm 除以 365 nm。不过这个图像的清晰度,还是比不上通过可见频率除法得到的结果。修复结果可能难以预测,有时手稿的材质与状态会以无法预期的方式影响反射光,导致我们无法获得预期的效果。因此,指导原则虽然重要,但也仅仅是指导而已。面对众多的选择,这些原则能够为我们的方法提供指引,使其系统化。但数码修复本身依然需要不断尝试与实验。
The ultraviolet frequency continued in its trend of not being as beneficial as expected. Dividing each of the other frequencies by 365 nm generates only one image that produces good recoveries: 850 nm by 365 nm. This image, however, provides less clarity than those derived from dividing visible frequencies. Results can be unpredictable. Sometimes, the materiality and condition of a manuscript affect reflected light in ways that do not provide the expected leverage. Therefore, guiding principles are important, but they are just that, guidance. For an otherwise overwhelming number of options, they direct and systematize the approach. However, digital recovery requires play and experimentation.
对于通过除法进行的修复,彩色相机拍摄的图像也提供了更多的机会。不过如前所述,拆分这类图像只能得到三个可用于除法运算的图像,每个通道对应一个,即红、绿、蓝。因此,可能的运算方式仅限于红除以绿、红除以蓝、绿除以蓝,如果需要也可以反转。拆分大英图书馆拍摄的照片后,绿色通道除以蓝色通道的结果得到了稳定的修复效果。应用不同的 luts 生成伪彩色,进一步优化了这些结果。另一种生成伪彩色并完成修复的方式,是合并与 RGB 图像预期不同的频率。例如,可以将绿、蓝和紫外线频率分别对应红、绿、蓝通道,合并生成伪彩色图像。
For recovery by division, images produced by a colour camera also provide further opportunities. However, as mentioned, splitting them produces only three images for division, one for each channel: red, green, and blue. Therefore, the possibilities are limited to dividing red by green, red by blue, and green by blue (and the reverse, if preferred). Splitting the photograph taken by the British Library generated solid recovery for the green-channel divided by blue. Applying different luts generates false colour, enhancing these results. Another way to generate false colour and recovery is to merge different frequencies than those expected for an RGB image. For example, a false colour image can be generated by combining green, blue, and ultraviolet frequencies for the red, green, and blue channels.
不过对于第 90 页,以各种组合合并频率都没有得到有益的结果。但依然存在其他的可能性,除法运算后的频率结果也可以进行合并。例如,将红外频率作为红通道,将一个除法运算后的频率作为绿通道,即 592 nm 除以 505 nm,再将紫外线作为蓝通道,合并后得到了极佳的结果。紫外线频率将更多受损文本的痕迹带入 RGB 图像中,而红外频率则淡化了透印现象。不过对于绿通道而言,用较高频率除以较低频率这一步十分关键,这会让修复后的文本比周围区域更亮,生成所需的对比度。如果将除数与被除数反转,铁胆墨水的痕迹就会变得更暗,不过在我看来,这会让它们的细节更难辨认。
For page 90, however, merging frequencies in various combinations did not generate beneficial results. Nonetheless, further possibilities exist: the results from divided frequencies can also be merged. For example, merging an infrared frequency for the red-channel, a divided frequency for the green (592 nm divided by 505 nm), and the ultraviolet for the blue produces excellent results. The ultraviolet frequency brings more traces of damaged text into the RGB image while infrared fades the bleed-through. However, for the green-channel, dividing the higher frequency by the lower frequency is significant. This causes recovered text to be lighter than its surroundings, generating needed contrast. When the divisor and dividend is reversed, the traces of iron gall ink become darker; however, to my eye, this makes their nuances more difficult to see.
最后,合并通道并不局限于只有一个通道包含除法运算后的频率,两个甚至全部三个通道都可以包含这类结果。在之前的合并图像中,蓝通道效果最好的除法运算结果是 535 nm 除以 465 nm。为了找到这个合适的除法组合,我最初在蓝通道的大致范围内测试结果,之后再逐步扩展。为了给红通道也加入一个除法运算后的结果,我发现 700 nm 除以 625 nm 是个不错的选择。再次说明,优化修复效果需要不断实验。将原始 RGB 图像分别除以三个通道,能让我们初步了解透印与受损文本是如何反射光线的。基于这些结果,我们可以开展有针对性的实验,提升得到有效结果的概率。
Finally, merging channels is not limited to a sole channel containing a divided frequency. Two or all three channels can contain them. In the prior merged image, one of the best divisions for the blue-channel turned out to be 535 nm divided by 465 nm. To find this division, I initially tested results within the general range of the blue-channel and then expanded them. To include a third divided result for the red-channel, I found that 700 nm divided by 625 nm was a good choice. Again, enhancing recoveries requires experimentation. Dividing the original RGB image by its three channels provides initial understanding about how bleed-through and damaged text reflects light. From these results, informed experimentation can proceed, increasing the chance of generating revealing results.
以上内容来自书籍官网与学者主页
https://www.arc-humanities.org/9781802701227/digitizing-medieval-manuscripts/
https://www.ou.edu/cas/english/people/bill-endres
经数字人文资讯小编翻译整理而成
如需转载,请后台私信联系
编辑 丨魏翔
校对 丨洪冰凤
排版 丨罗斯鹏
![]()