Indiscapes: Instance Segmentation Networks for Layout Parsing of Historical Indic Manuscripts

Abstract

Historical palm-leaf manuscript and early paper documents from Indian subcontinent form an important part of the world's literary and cultural heritage. Despite their importance, large-scale annotated Indic manuscript image datasets do not exist. To address this deficiency, we introduce Indiscapes, the first ever dataset with multi-regional layout annotations for historical Indic manuscripts. To address the challenge of large diversity in scripts and presence of dense, irregular layout elements (e.g. text lines, pictures, multiple documents per image), we adapt a Fully Convolutional Deep Neural Network architecture for fully automatic, instance-level spatial layout parsing of manuscript images. We demonstrate the effectiveness of proposed architecture on images from the Indiscapes dataset. For annotation flexibility and keeping the non-technical nature of domain experts in mind, we also contribute a custom, webbased GUI annotation tool and a dashboard-style analytics portal. Overall, our contributions set the stage for enabling downstream applications such as OCR and word-spotting in historical Indic manuscripts at scale.

Combination of deep neural networks and logical rules for record segmentation in historical handwritten registers using few examples

“…These networks can handle different layouts of printed documents, but require many training examples -more than 1,000 documents in these studies. To the best of our knowledge, the only attempt at applying object detection networks on historical documents was done by Prusty et al [55]. They have trained Mask R-CNN on 120 to 350 documents to find instances of different page objects, such as text-lines and page boundaries, in historical Indic manuscripts.…”

Deep Learning for Historical Document Analysis and Recognition—A Survey

“…Extant copies of these early manuscripts written in Greek or Latin and usually dating from the 4th century to the 8th century AD, are classified according to their use of either all upper case or all lower case letters. Several researchers addressed the analysis and recognition of these documents; even considering only those published in the ICDAR 2019 proceedings we can count nine papers ([ 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 ]).…”

“…Different text lines are located in Indic historical documents in Reference [ 14 ] by using a deep model based on a Mask R-CNN with a ResNet-50 backbone. The different task of instance segmentation (that separates individual objects in the page, e.g., each text line) with respect to semantic segmentation (that aims at identifying pixels belonging to a given object type, e.g., text lines) is taken into account and discussed in the paper.…”

“…Models that comply with the above mentioned features have also been used in the field of historical documents processing and understanding, as summarized in the following. Prusty et al [ 14 ] use FCN for instance segmentation of text lines and other areas in historical documents and a similar approach is used in Reference [ 76 ]. Individual Japanese characters are segmented with FCN in Reference [ 20 ].…”

“…Applications of object detection models for historical documents range from keyword spotting in early printed works [ 21 ] by using Faster R-CNN, to the location of different text lines in Indic historical documents considering a Mask R-CNN architecture [ 14 ].…”