Scrabble plate OCR

[Colas Decron]

Hello everyone!

I'm implementing, on Android, some digital Scrabble tile recognition, and I got some weird recognition behaviour.
After a few transformations, the images I got are these ones:

           ..............................

and the recognition sometimes recognize a O as a I, or a D as a I.

The settings I use are:

setPageSegMode(TessBaseAPI.PageSegMode.PSM_SINGLE_CHAR);
setVariable(TessBaseAPI.VAR_CHAR_WHITELIST, "ABCDEFGHIJKLMNOPQRSTUVWXYZ");
setVariable("load_system_dawg", "0"); // But it seems not to change anything
setVariable("load_freq_dawg", "0"); // But it seems not to change anything

I would find pretty weird that with as much as obvious and digital (= perfect) characters (O is obviously not a I !!), there aren't any Tesseract settings or base images transform I can make to improve recognition.

Do you have some ideas to improve this system?
I was thinking of training Tesseract on this font, but it seems to be a huge work!

Thanks a lot!! :)