I have a large set of difficult words to recognize through ᴏᴄʀ.
The image are all of lowercase latin alphabet without diatrics, but despite using languagehints, I end up with results like 学会学论發及 in a lot of cases.
So how to force Google cloud vision to use a specific set of letters and not just hint ?
Currently, there is no way to “force” vision API to use a specific set of characters. Even if that was the case, the ML model could still not be able to predict the correct result.
It may be worth opening a public issue through this form, detailing your use case and the difficulties you are encountering. Perhaps a product enhancement can come from that.