This paper will introduce a new project aimed at investigating and detecting change in printed documents.
Printing (separately ‘invented’ in China, Korea and Germany at various times) has repeatedly been characterized as the most transformative of all technologies prior to the computer. The term has been applied to such disparate processes as the letterpress and engraving; press lithography (now employed in semiconductor fabrication); photography; and 3D additive manufacturing. All of these processes (despite their differences), are dedicated to the attempted production of what William Ivins called ‘exactly repeatable pictorial statements’, a notion shortened by Elizabeth Eisenstein to ‘fixity’. However, the actual results have often been far from exactly repeatable, due to factors such as manual errors in typesetting; artistic revisions; and damage to type and printing surfaces. It is a standing task among book and art historians to compare multiple copies of ‘the same’ book or printed artwork to each other in order to detect variant states and to determine the order of printing. This work is painstaking and difficult: it was revolutionised in the 20th Century by the invention of optical machines such as the Hinman Collator, followed by stereoscopic and photographic devices and the invention of digital compositing and image rectification techniques, which are employed in tools such as the Traherne Digital Collator, developed by the proposer’s research group to assist scholarly editors.
A new collaboration between the Universities of Manchester and Oxford, funded by the Schmidt Science Foundation, aims to go beyond using digital techniques to visualise differences in printed images (whether those are images of words or pictures), and to detect, classify, localise and describe or transcribe those differences. Our goal is to develop software to assist researchers in perceiving and understanding differences; in distinguishing between types of difference; and in localising and describing regions of difference. While large vision-language models have seen great improvements in single-image classification, the current generation are less capable in classifying differences between multiple images. We believe that this project will stress-test the state of the art and advance it by finetuning, possiblt including with data-augmentation and rules-based approaches. We will additionally create a ground-truthed dataset of sets of historical printed images which will be the basis for training our model and benchmarking future models: this data will include early-modern English printed texts such as Shakespeare; 19th century American novels; Japanese woodblock prints; and European engravings, musical scores and scientific illustrations.
The paper will first set out this research domain and outline the current state-of-the art in change detection. It will describe our approach, including research in cognitive mapping of how experts perceive and process difference, which inform the user-interface design of our application. It will aim to elicit comments from the international ADHO audience as to our approach and to its applicability to materials that we have not so far considered.
References
Kato T, Tokunaga S. Gutenberg Meets Digitization: The Path of a Digital Ambassador. In: Horobin S, Mooney LR, eds. Middle English Texts in Transition, Manuscript Culture in the British Isles. Boydell & Brewer; 2014:297-305.
Ivins, William M (1953) Prints and Visual Communication. London. Routledge
Eisenstein, Elizabeth L (1980) The printing press as an agent of change. Vol. 1. Cambridge University Press,
Tinios, Ellis, 2015. ‘Hokusai and his Blockcutters.’ Print Quarterly 32. 2, pp. 186-91