I don't know if it is an issue with google vision API, but the returned characters are in western subset only while correct locale is detected