Russian scientists trained AI to read fragile scrolls without opening them

MOSCOW, June 3Russian scientists have trained artificial intelligence to read fragile scrolls and books by pressing one button, without opening them, this will help preserve valuable ancient sources and reduce the risk of damage, the company that developed this technology, Smart Engines (SE), reported.
This is the first fully automatic virtual scroll unrolling system that does not require human intervention

“Scientists from the Russian company Smart Engines and the Federal Research Center “Informatics and Control” of the Russian Academy of Sciences (Federal Research Center “Informatics and Control” RAS — ed.) trained artificial intelligence to extract the contents of fragile scrolls without unfolding them. Now you can get the restored text from a manuscript by pressing one button. The development will allow you to study texts of paper, birch bark and silver scrolls and printed books that have suffered due to natural aging, exposure to humidity or fires, and are stored in special conditions,” the report says.

As the company said, Russian researchers used a non-destructive X-ray tomography method and computer vision algorithms: an object that cannot be unfolded is placed in a tomograph. Then a digital copy of the document is reconstructed, on which all further manipulations are performed.
SE emphasized that the global scientific community has been trying to solve the problem of analyzing old scrolls using tomography for more than twenty years, offering different approaches with machine learning. At the same time, no one has been able to solve the issue of complete automation of the process until today.
To train the new neural network, a data corpus created by Smart Engines scientists was used, including six samples of documents folded in different ways. They were marked with letters and numbers of different sizes and diagrams with various graphic elements.
< /span>
“We hope that our development will allow us to take a step towards the research and preservation of cultural heritage and will open up new opportunities for historians, archaeologists and other specialists in the humanities. We are now actively looking for partners to conduct further research,” said the CEO of Smart Engines, Doctor of Technical Sciences Vladimir Arlazarov.

The results of the work on this project are expected to be presented by Russian scientists in August 2024 at The International Conference on Document Analysis and Recognition (ICDAR) – the leading international scientific conference in the field of document analysis and recognition (Athens, Greece).
SE was founded in 2016 by scientists with 25 years of experience in the field of character recognition, images and document processing. The company's software products such as Smart ID Engine, Smart Code Engine and Smart Document Engine are used by 10 of 13 systemically important credit institutions in Russia – Alfa-Bank, VTB, Gazprombank, MKB, Otkritie Bank, Promsvyazbank, Raiffeisenbank, Rosbank, Sovcombank and Tinkoff. SE solutions are also used to sell tickets at Russian Railways ticket offices, at automatic passport control at Sheremetyevo, when registering self-employed individuals in the Federal Tax Service My Tax application, and in other projects.