mirror of
https://github.com/JaidedAI/EasyOCR.git
synced 2026-11-20 12:07:07 +08:00
The original Japanese model only used a long list of Kanji which proved to be suboptimal to recognise particular Japanese constructs. I added a script that starting from JMdict and JMnedict, creates a wordlist using Hiragana, Katakana and Kanji variants. It then creates a character file from this wordlist. I also added a punctuation file with the characters that made sense from the Japanese punctuation article on Wikipedia [1]. [1]: https://en.wikipedia.org/wiki/List_of_Japanese_typographic_symbols