Lost Visions: retrieving the visual element of printed books from the nineteenth century

我们所处的时代，由于数字技术的发展，图像以前所未有的速度激增。因此，具有讽刺意味的是，我们过去的视觉文化正受到同样的技术进步的严重威胁。许多口头文本现在由机器存储和传递，数字搜索技术帮助我们理解这些文本的含义、来源和接收。然而，由于经济原因或复杂的知识产权问题，这些文本中的插图经常被省略。当包含它们时，它们通常是低质量的，并且没有理解它们所需的元数据。虽然它们的市场价值几乎不会因为这些缺点而受到影响，但它们作为美学、文学和历史信息来源的用途却受到严重损害。有一种风险是，明天的读者几乎不知道插图，尽管研究表明，插图文本的质量、意义和策略与没有插图的文本大不相同，甚至与那些没有插图的口头文本截然不同。整个类型的艺术，教育和信息产品可能有效地失去了我们。大英图书馆的数据集提供了一个独特的机会来开始解决这个问题。该数据集由68,000个数字化卷组成。2500万页)，涵盖了漫长的19世纪的“文学”、“哲学”、“通史”和“地理”，这一时期可以说是英国书籍插图最重要的时期。这些年来，与生殖技术的快速变化并行的是艺术及其接受意义的变化。艺术民主化了，书籍插图比以往任何时候都更具有收藏价值和流动性。数据集中的页面以高分辨率和OCR进行扫描，并记录其中发现的非语言对象的存在。我们将使用计算方法来识别这些非语言对象的视觉特征（即它们是否具有地图，图表，表格，图形，重要排版布局等的性质）。我们的目标是通过提供书的完整书目细节、图像的确切位置和大小，并在可能的情况下提供标题或标题以及艺术家，来增加可用的元数据。此外，我们将开发用于识别图像再利用的工具（以复制，重新绘图，重新切割和中介的形式）。结果将是一个基于web的界面，提供对带有附带元数据的图像数据库的可搜索访问。因此，这个项目将拯救大量的图像，否则所有意图和目的都将丢失。虽然拨款的规模不允许对这些插图的主题进行全面分析，但我们将尝试用众包的方法来分配标签。我们目前正在与一个数据管理SME讨论协作如何能够大规模地提高语义标记的效率并降低单位成本。这本身是一个大项目，但其最终影响要大得多。它有可能彻底改变插图的理解方式和重要性，为渴望图像的商业世界提供插图材料，并导致在大型数据库中对图像进行分类和分析的更准确的方法。为了处理项目期间出现的问题，并实施一项影响战略，其影响将远远超出项目本身，我们成立了一个由合作者和外部顾问组成的庞大小组，他们在项目本身以及在潜在用户世界中既具有专业知识又具有影响力。

Ours is an age when images proliferate at unprecedented speed thanks to digital technology. So it is ironic that our past visual culture is seriously threatened by the very same technical advances. Many verbal texts are now stored and delivered by machine, and digital search techniques help us understand the meaning, provenance and reception of these texts. However, for reasons of economy or of complex IPR issues, the illustrations in these texts are frequently omitted. When included, they are often of low quality and without the metadata which are needed for understanding them. While their market value hardly suffers because of these shortcomings, their use as sources of aesthetic, literary and historical information is seriously compromised. There is a risk that tomorrow's readers will be almost unaware of illustration, even though research has shown that illustrated texts have qualities, meanings and strategies which are very different from those of texts that are not illustrated, and even strikingly different from those of the same verbal texts stripped of their images. Whole genres of artistic, educational and informative products may effectively be lost to us.A unique opportunity to start addressing this issue is provided by a dataset held by the British Library. This dataset consists of 68,000 digitised volumes (approx. 25 million pages) covering 'Literature', 'Philosophy', 'General History' and 'Geography' from the Long Nineteenth Century, a period that is arguably the most important in British book illustration. In these years, rapid changes in reproductive techniques were paralleled by changes in the meanings of art and its reception. Art was democratised and book illustrations became more widely collectable and mobile than ever before. The pages in the dataset are scanned at high resolution and OCR'd and the presence of non-verbal objects found in them is noted. We will use computational methods to identify the visual characteristics of these non-verbal objects (i.e. whether they partake of the nature of maps, diagrams, tables, graphs, significant typographic layout etc.). We aim to add to the available metadata by giving full bibliographical details of the book, the exact location and size of the image, and, where possible, a caption or title, and an artist. In addition, we will develop tools for identifying the re-use of images (in the form of reproductions, re-drawings, recuttings and transmediations). The result will be a web-based interface that provides searchable access to a database of images with accompanying metadata. As such, this project will rescue large numbers of images which would otherwise to all intents and purposes be lost. While the size of the grant does not allow for a full analysis of the subject matter of these illustrations, we shall trial crowd-sourcing methods for assigning tags. We are currently in discussion with a data management SME on how collaboration could increase the effectiveness and decrease the unit cost of semantic tagging on a large scale. This is a project big in itself, but far bigger in its eventual impact. It has the potential to revolutionise how illustration is understood and the importance accorded to it, to supply an image-hungry commercial world with illustrative material, and to lead to ever more accurate ways of classifying and analysing images in large databases.To deal with issues which arise during the project and to implement an impact strategy which will be felt well beyond it, we have set up a substantial panel of collaborators and external advisers who have expertise in the project itself and both expertise and influence in the world of potential users.