Compare a correctly blurred copy of the image to your output.  The test passes if they are identical, otherwise it fails.

I don't believe John's method is good.  If a bug causes the output to be all black instead of a blurred image, the OCR test would pass while a simple byte-by-byte comparison would fail.