Originally posted by: MaximRecoil
It is impossible to end up with more information than you started with (i.e. your source).

That's not true -- but it's also not what you mean. Of course it is possible to end up with more information than you started with. That's pretty much the whole idea. What I know you're saying is that it's impossible to end up with more useful or accurate information than you started with, and you do have a point there. I still disagree, but only in specific instances. I would agree with you that it is very easy to end up with several million more pixels of junk. However, it doesn't have to be so.

Imagine a very low-sampling frequency rendering of a sine wave. Something that looks like it came from an Atari 2600 game, for example. Say that's the original data. Now, imagine that you take a look at that rendering. It would be pretty clear by looking at it that it's supposed to represent a sine wave, even though it's blocky as all hell. From the 'original data' -- i.e., the blocky rendering -- you can still immediately see the frequency and the amplitude. Good enough, but is that the end of the line, quality-wise?

Now, let's say you wanted to make a better version of that sine wave, but you start from the blocky version. It doesn't take that much thought to see it's possible (and maybe even relatively easy) to draw a much smoother curve that fits the data. Same frequency, same amplitude -- better resolution. In fact, you could probably downsample and get the same exact original blocky curve back -- or you could downsample less and still get something smoother and more pleasing than the original. Clearly, in this case, you have ended up with a result that has more information than you started with, and that information is useful and accurate.

I will be the first to admit that a full-frame, 24-bit colour DVD image is not a sine wave. It's zillions of times more complex (1 zillion = 10^a lot), and to teach a computer how to recognize where and how to make changes to improve things isn't easy. Still, the principle is the same as in my simple example. It is possible to make improvements to the original data that will give a result that is better than what you started with. It takes a watchful eye and lots of clock cycles, but it can be done. I think to say otherwise (absolutely, no less) is overly simplistic... and incorrect.

That's my take, anyway.