i think someone maybe can do it... recognition software especially for CCTV does not only depends on one still picture but it uses combination of several still pictures of certain image (plate number) to reconstruc the "missing picture". Because CCTV system capture app. 1-5 picture per second.
I have read in National Discovery magazine which disscuss on this related topic. But there is still problem with the software, such missconstruc I to L, 3 to 8, C to O and so on...