What are noise and grain?

[jellysandwich]

I've been working with anime mostly, so I
haven't had to worry about these. But now
I'm doing some TV shows (24 specifically),
and it seems kind of, well, grainy...

Pictures would be especially helpful.

-jellysandwich

[Soulhunter]

Well, grain (celluloid granularity) looks like this! (http://img249.echo.cx/img249/3271/720psample0018yb.jpg)

But noise could be anything, depends on the definition / context...

Think everything that "distorts" the image could be classified as noise!

[E-Male]

grain is best spotted and most disturbing when it's includes alot of 'changing' (for the lack of a better word)
that can't be noticed on a screenshot

IMHO if you have the space (liek on a dvd9) for older movies you should keep the grain, as it is part of it somehow (as long as it isn't too much, which is a matter of taste)

but for newer tv series there should be be no trouble with grain, i daubt it's shot on film
but there are so many other kinds of noise

[trevlac]

I've read that grain only shows in the midtones. If you look at Soulhunter's picture, you can see that the white and black are free of the "problem". The effect is due to the size of actual grains of silver (or whatever it is) that are exposed on the film.

Here's a link that describes video noise:
http://broadcastengineering.com/mag/broadcasting_analog_video_noise/index.html

It breaks it down into random and Coherent. Coherent is basically a pattern introduced by some 'outside' process. Like a nearby channel. This pattern can usually be seen on a spectrum analyzer.

The article talks about a few kinds of "Random". White noise has an even power distribution throughout the frequency range which I believe means it is most likely to show in the detail areas of the image because an image has more power in the low frequencies vs the high frequencies so there would be more noise vs image in the high frequencies for white noise.

I could be off a bit about the white noise.

----------------
Hard to say what people mean when they say noise :)

[Wilbert]

The article talks about a few kinds of "Random". White noise has an even power distribution throughout the frequency range which I believe means it is most likely to show in the detail areas of the image because an image has more power in the low frequencies vs the high frequencies so there would be more noise vs image in the high frequencies for white noise.

I could be off a bit about the white noise.

Seems correct. As you note, the powerspectrum of white noise is flat. Since the noise is independent of the signal, the total powerspectrum is just the sum of the powerspectra of the signal and of the noise.

[trevlac]

the total powerspectrum is just the sum of the powerspectra of the signal and of the noise.

Just to pontificate a bit more ... :) Sum means noise is naturally an additive process. So noise reduction must be subtractive. Take a blur for example. It basically removes the high frequency bumps from the image 'wave'. 'Averaging multiple captures' is an interesting one. Because each capture contains the 'true' image with noise, and because the noise from capture to capture is not repetative in the same pattern, adding multiple copies of the 'true' image builds up that image in the mix. Then you basically divide by the number of images to subtract the noise.


I'm guessing the various noise reducers use all these tricks ... where they can. :)

[jellysandwich]

Ahso... So that's what they are.

Thanks!

-jellysandwich