Skip to content

Python application to identify speech bubbles and read text from comic book pages (French adaptation)

License

Notifications You must be signed in to change notification settings

jprobichaud/comic-book-reader

 
 

Repository files navigation

Comic Book Page Reader

Summary

Application python pour extraire les bulles et le texte de celles-ci avant de générer des fichiers audios correspondants.

Installation

C'est un peu primitif pour l'instant.

  1. Installer conda (miniconda suffit : https://docs.conda.io/en/latest/miniconda.html)
  2. Créer l'environment : conda env create -f conda_environment.yml
  3. Activer l'environment conda : conda activate ComicReaderFr

Example:

python from-folder.py repertoire_de_sortie  fichier_image.jpg

Le script va analyzer le fichier fichier_image.jpg et mettre les fichiers audios, le fichier html, les différentes images et le texte extract (un fichier par bulle détectée) dans le répertoire, avec comme base de nom fichier_image

Il y a quelques fichiers dans le répertoire examples

Technologies

Python, Pytorch,

Libraries

doctr.io for OCR (Optical Character Recognition), OpenCV snakers4/silero-models for TTS generation

Developer Notes

References

Many thanks to: https://github.com/damishshah/comic-book-reader !

Dubray, David & Laubrock, Jochen. (2019). Deep CNN-based Speech Balloon Detection and Segmentation for Comic Books. https://arxiv.org/abs/1902.08137.

I wanted to give a shoutout to the research team from Cornell for their research in this area. I reached out to them when I was considering a Nueral Net approach to this problem and they helped answer some questions that I had. You can read their excellent research paper at the above link and check out their model here: https://github.com/DRDRD18/balloons

About

Python application to identify speech bubbles and read text from comic book pages (French adaptation)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 60.8%
  • JavaScript 20.2%
  • CSS 11.5%
  • HTML 6.4%
  • Dockerfile 1.1%