Unconstrained Text Detection in Manga: a New Dataset and Baseline

  • 2020-09-09 00:16:51
  • Juli├ín Del Gobbo, Rosana Matuk Herrera
  • 46

Abstract

The detection and recognition of unconstrained text is an open problem inresearch. Text in comic books has unusual styles that raise many challenges fortext detection. This work aims to binarize text in a comic genre with highlysophisticated text styles: Japanese manga. To overcome the lack of a mangadataset with text annotations at a pixel level, we create our own. To improvethe evaluation and search of an optimal model, in addition to standard metricsin binarization, we implement other special metrics. Using these resources, wedesigned and evaluated a deep network model, outperforming current methods fortext binarization in manga in most metrics.

 

Quick Read (beta)

loading the full paper ...