>> - You *cannot* take CCTV footage, enlarge it 20x, "run an imagine
>> enhancement algorithm" and then read a car numberplate from 300 yards.
>> It doesn't work like that.
>
>    Actually it's not *that* far-fetched. When one understands the basics of
> image manipulation and computers, it's intuitive to think that image data
> that's so noisy or blurry is so destroyed that no useful information can be
> retrieved from it. However, in many cases the information isn't *completely*
> destroyed and there are very clever heuristic algorithms to try to guess
> what the information was.

Especially as in the example mentioned, car number plates, you know the 
font and size characters that are used.  Even with a very low resolution 
image (I'm talking just a few pixels for each character) you can often 
make a very good guess what the original letters were.  Given multiple 
frames, each with random noise and/or slightly different positions you 
are going to home in on one set of characters that is most likely.  That 
would probably seem impossible to most humans.

Post a reply to this message