'OCR' and a very huge chuck of processing power.

There are a number of commercial products which do this. Is it your
intention to create something similar? Maybe your best bet is to approach
these product manufacturers and ask them if they're willing to license their
technology to you.

If, on the other hand, this is a student assignment, tell your professor
he's got a cool sense of humour. ;-)

The practical solution is you feed the images to some commercial OCR
software. However, if you want to play:

If your images are perfectly cropped, so they are all the exact same
size, your first task to enhance the contrast.

You want to tweak the contrast to get say 10% of the pixels black. You
tweak the cutoff until that happens. 10% is a magic number to be found
by experiment.

Let's for a moment presume the font is fixed pitch. You slice the
image into 6 pieces, one for each letter/digit.

Now you compare each piece with each  of the 30 possibilities,
computing a bit count difference.

That is pretty crude. It would work perfectly for screen captures
though, a related problem -- converting screen captures back into
text.

For a real license system, e.g. an automated speeding ticket system,
you have a number additional problems:

1. the image is unevenly lit. You must compensate.

2. you are looking at the image at an angle. You must compensate.

3. You have to discard out of state license plates for manual id.

4. the image may be dirty, or mud splattered.

5. the font might be variable pitch.

6. you have to find and crop the licence plate in a large confusing
image.