My problem is similar to what is described
here. I generate perfect, hi-res text using ImageMagick's
convert command line tool, and then give the result as an input to
tesseract, but what I get is very bad quality. Lowercase "w" become uppercase, uppercase "X" become lowercase "h" etc. I've tested a few fonts - including OCR-A - used different color spaces, configured tesseract to ignore language dictionaries etc., I can't get to a settings that assures me a seamless conversion. However, I haven't used any training yet.
What am I missing? Is it about training? In your experience, have you found anything that assures no error while keeping the text human readable and using a non-copyrighted font?