You cannot have characters at different orientations trained in the same OCR file and associated with the same character. I would recommend you learn all the characters for one orientation in one OCR file, and in a different OCR file learn the characters at the other orientation. It seems like you will know which orientation you want to read since you need to create an ROI around the letters and then in VBAI, you can use some logic and branch to a state that reads the characters horizontally using the horizontal OCR file or branches to a state that reads the characters vertically and uses the vertical OCR file. Another option would be to have different characters associated with a horisontal vs. vertical letter (i.e. use 'e' for vertical e and 'E' for horizontal e, but this could get confusing and you'd need a calculator step to convert the read text to all lower case or something to use this method.
It also looks like you might want to do some preprocessing cleanup of the image so the characters are easier to differentiate from the background.
Hope that helps,
Brad
