Of course you wouldn't have to read physical images every time, you could read each one once and store the binary data in a database.

I suggested images plural, because that would allow complex words to be made using basic building blocks (letters and numbers). It would probably be possible to merge them into a single file, but that sounds a little too complex for such a basic problem.

Having each letter/number is different fonts and sizes would also give OCR programs a harder time of things. Having 3 or 4 versions of each alphanumeric character, chosen at random, would also confuse things further.