Skip to content

Latest commit

 

History

History
39 lines (28 loc) · 1.19 KB

README.md

File metadata and controls

39 lines (28 loc) · 1.19 KB

speech-bubble-extraction

Extracting the text from speech bubbles from manga

A page can be translated by running the below commamd

python easyocr_translate.py --img_path IMG_PATH --out_path OUT_PATH

Its a work in progress, the quality of cleaning and translation is very poor

Sample Output

After training the efficientdet for 3 epochs on the Manga109 dataset

Google translation with easyocr

Snow
Forest

Credits

I am using the scripts in src folder from signatrix implementation of efficientdet which is available here
I also used the manga109api to save time on parsing the manga109 dataset

To do

  • Enhancing the text containing parts of the image using morphological operations for improving the OCR's performance
  • Improve Cleaning
  • Improve text placement