Text-line-up: Don't Worry About the Caret

Chandranath Adak, Bidyut B. Chaudhuri, Chin-Teng Lin, Michael Blumenstein

Abstract

In a freestyle handwritten text-line, sometimes words are inserted using a caret symbol ( $$^\wedge $$ ) for corrections/annotations. Such insertions create fluctuations in the reading sequence of words. In this paper, we aim to line-up the words of a text-line, so that it can assist the OCR engine. Previous text-line segmentation techniques in the literature have scarcely addressed this issue. Here, the task undertaken is formulated as a path planning problem, and a novel multi-agent hierarchical reinforcement learning-based architecture solution is proposed. As a matter of fact, no linguistic knowledge is used here. Experimentation of the proposed solution architecture has been conducted on English and Bengali offline handwriting, which yielded some interesting results.
BibTeX
@inproceedings{adak2021text,
  author = {Chandranath Adak and Bidyut B. Chaudhuri and Chin-Teng Lin and Michael Blumenstein},
  title = {{Text-line-up: Don't Worry About the Caret}},
  booktitle = {ICDAR},
  year = {2021},
  doi = {10.1007/978-3-030-86334-0\_14},
}

Loading related content...

Figure