Images with noise!!!

[Dennis Joseph]

Originally Posted by Dominik Seibold

Try detail-off. Btw, the crispening-parameter of the detail-settings controls the amount of noise-reduction, but I always feel most satisfied with detail-off (no noise introduced by unneeded sharpening and so no noise-reduction needed. The result is a very clear and natural picture).
Also use cine-gammas instead of std-gammas. std-gammas leave a lot of headroom for aggressive knee-settings, so they are more noisy because of their (digital) extra-gain. I recommend cine1 as a general-purpose-gamma.
Also be aware of the fact, that -3db reduces the dynamic range. So just use it if the scene is not contrasty (when maximum exposure without clipping leads to an overexposed picture). It's a bad idea to use -3db and then underexpose to avoid clipping.
Btw, there are two kinds of noise: Noise introduced by the analog gain-circuits and noise introduced by the a/d-converters. The analog gain-noise looks more pleasing (gaussian noise vs. quantization noise). So always prefer analog gain to digital gain (furthermore digital gain of course always amplifies analog noise. In fact analog gain increases the s/n-ratio because digital noise gets relatively lower).

I use cine gammaes only. 0 db, iris full open and still some noise. Am I missing something?

[Dominik Seibold]

Originally Posted by Leonard Levy

Dominik,
Could you explain this idea in more detail (no pun intended) please. I've never heard this before and don't understand what you are saying. On the contrary, most of the experienced engineers I know actually tend to be suspicious of cine gamma type curves.

Well, I also never heard this, but that's what I'm seeing when switching between the std- and cine-gammas. std-gammas are much brighter than cine-gammas. Also you can observe, that their-clipping-to-white-point (with knee turned off) is much lower than the clipping-points of the cines (and furthermore that the std-gammas share all the same lower clipping-point and the cine-gammas share all the same higher clipping-point). But where does the extra-brightness come from? Magic? Because I don't believe in magic, it must come from some kind of amplification. And amplifying a signal also means amplifying its noise. And that's not purely hypothetical, but you can actually see the extra-noise. Why more noise? Because of the higher brightness of the stds you have to decrease the exposure-settings, so the s/n-ratio (signal-to-noise ratio) drops. Also I observed that the auto-exposure doesn't care about your gamma-settings and that it seems to be calibrated to work right with the std-gammas. Auto-exposure does underexpose with cine-gammas.
So does it make any sense to lower the clipping-point? Yes, because if you want to be able to compress the signal with a knee-function, there must be some headroom and lowering the clipping-point gives you that headroom, because that lower clipping-point is just a virtual one - the real clipping-point of the sensors hasn't changed. But with cine-gammas you don't need any extra-headroom, because they are fixed, because they don't have a knee-function.
So it's all about the question, if you need that ultimate-compression which the std-gammas with their knee-function offer. I don't.

Originally Posted by Leonard Levy

I hope you don't take offense at the question, but what is your background? - i.e how long have you been doing video, engineering background etc. I just want a context for what your saying since its unusual to me.

That video-stuff is just an old hobby of mine. But I'm a (25 year old) student studying electrical engineering and computer science (at a university in germany/erlangen), so I have some technical background knowledge.

[Seun Osewa]

I use cine gammaes only. 0 db, iris full open and still some noise.

Try -3db. Note also that 6-bit dithered laptop monitors make images appear noisier than they actually are. Use CRT monitors or TV screens to judge noise.

[Leonard Levy]

Dominik,
I asked the background question because I suspected you might not have a wealth of experience on the video engineering side.
I'm no expert in this and someone like Adam Wilt will probably tear me apart, but I think you are fundamentally miss-characterizing the situation.

So here's my bad version - The standard gammas without knee added are the closest to the way the chip and electronics process a video image. No extra noise or extra sensitivity added. The knee in standard gammas is an extra processing step common on all video cameras to soften the chip's inherent tendency to get increasingly sensitive to light and thius get better pictures where bright areas exist.

Most video engineers prefer standard gammas unless there are highlight problems because they can achieve an exposure with the widest range of video information.

Cine gammas are a relatively recent development that add radical knee circuitry (much more knee) and lower gamma to increase the exposure range but they may distort the grey scale, reduce low light sensitivity and slightly underexpose - thus perhaps limiting the amount of video information and to me that sounds like it could perhaps increasing noise.

What the trade-offs are is pretty complicated and may come down to individual choice and experience. You may be right to find that cine gamma for your shooting reduces noise. Someone else might find the opposite. So far I prefer standard 3 with a preset knee but i may change my mind yet.

Lenny Levy

[Dominik Seibold]

Originally Posted by Leonard Levy

The standard gammas without knee added are the closest to the way the chip and electronics process a video image.

So if you have a 24bit sound-card for recording an analog signal, but your cd-player just has 16bit, you would set the recording-level to -96db, because it's closest to the cd-player? No, you would it set to 0db and use the extra-information for dithering/noise-shaping-algorithms for getting a better s/n-ratio.

Originally Posted by Leonard Levy

No extra noise or extra sensitivity added.

It's not about adding or subtracting sensitivity, it's about throwing away or using an existing dynamic range of the chips. It's about loosing or using bits the chips can output.

Originally Posted by Leonard Levy

Most video engineers prefer standard gammas unless there are highlight problems because they can achieve an exposure with the widest range of video information.

With std+knee you can get an ultra wide dynamic range if you need it, but in most cases you don't need it and pay the bill with having less information in the more important higher-contrast-areas resulting in more noise.

Originally Posted by Leonard Levy

Cine gammas are a relatively recent development that add radical knee circuitry (much more knee) and lower gamma to increase the exposure range but they may distort the grey scale, reduce low light sensitivity and slightly underexpose - thus perhaps limiting the amount of video information and to me that sounds like it could perhaps increasing noise.

cine-gammas don't have knee, but a nice looking soft roll-off (you could call it soft-knee, but it's not comparable in terms of "more" or "less" knee). Yes, they have a lower gamma than std+knee, but that's good, because std+knee is way too aggressive. With std(+knee) in most cases you have to underexpose, so you loose video-information.
Here you can compare all gamma-curves available:
http://www.dominik.ws/gammas-native.png (100%-version - 30MB!)
http://www.dominik.ws/gammas-small.png (25%-version - 2.4MB)
The exposure-settings were always the same. More or less brightness and clipping is just the result of the gamma-curves.

[Leonard Levy]

Dominik,

I have no intention of arguing video engineering terminology with you. Based on 25 years as a cameraman and numerous discussions about gamma with very good technicians I suspect I you are misusing terms here but i may be wrong, and I don't question your subjective preference for cine gammas.

lenny levy

[Dominik Seibold]

You're right, I'm not a professional video-engineer, so I'm not thinking like one, but I know the basic concepts of analog and digital signal-processing, which recur in all kind special applications. Of course it's a matter of taste, what you want, but it's not a matter of taste what's the best way to get there. And using std without knee is really not to exploit the capabilities of these chips. If you want a similar look like std3-no-knee but a bit smoother and with less noise, use cine1.