Representational image (Illustration | Amit Bandre)
HYDERABAD: In the medieval ages, palm leaf manuscripts were predominantly used for documentation purposes. But over the years, most of them have withered and their knowledge rendered unredeemable. Even the ones that are still intact are remain difficult to read. To address this issue, researchers from IIIT-Hyderabad have developed a system which uses machine learning (ML), an application of artificial intelligence, to read and convert complex handwritten manuscripts into printed, editable text.
For the ML to work, Ravi Kiran Saradevabhatla of the Centre for Visual Information Technology and his team, obtained digitised manuscript from University of Pennsylvania’s Rare Book and Manuscript Library and Bhoomi, which is a collection of images sourced from multiple Oriental research institutes and libraries across India to ‘teach’ the machine.
The machine can now create boundaries around each document, identify each line among a set of ‘dense, uneven lines’, said a IIIT-Hyderabad blog post. Looking at an image of a scholar poring over a manuscript with a magnifying lens, Sarvadevabhatla got the idea of developing a ML tool to help such researchers read old and worn out manuscrips, the post stated. The tool is publicly accessible, easy-to-use, and could be beneficial for researchers involved in study of historic documents. It is called ‘Historical Intelligent Document Layout Analytics’ or HInDOLA.
