It’s amazing how poorly this is understood. I’m also rather surprised at some peoples expectations when it comes to noise in shadow areas of video images.
First of all, all video camera sensors produce noise. There will always be noise to some degree and this is typically most visible in the darker parts of the image because if your actual image brightness is 5% and your noise is 5% the noise is as big as the desired signal. In the highlights the same noise is still there, but when the brightness is 80% and the noise is 5% the noise is much, much less obvious.
ISO: What is ISO with a video camera? On it’s own it’s actually a fairly meaningless term. Why? Well because a camera manufacturer can declare more or less any ISO they choose as the cameras sensitivity. There is no set standard. It’s up to the camera manufacturer to pick an ISO number that gives a reasonably bright image with an acceptable amount of noise. But what is acceptable noise? Again there is no standard, so ISO ratings should be ignored unless you also know what the signal to noise ratio is at that ISO. For decades video camera sensitivity was rated in db. The sensitivity is measured at 0db in terms of the aperture needed to correctly expose a 90% white card at 2000 lux. This is very precise and easily repeatable. The signal to noise ratio is then also measured at the unity (0db) gain point and from this you can actually get a really good understanding of how that camera will perform, not just sensitivity, but more importantly how much noise at the nominal native sensitivity.
But now, because it’s fashionable and makes us sound like film camera operators it’s all about ISO. But ISO on it’s own doesn’t really tell us anything useful. Take a Sony FS7 or F5. In standard gamma at 0db the ISO rating is 800 ISO. But when you switch to S-Log it becomes 2000 ISO (but you are still at 0db). Have you ever noticed that the image doesn’t get brighter even though you are increasing the ISO? The ISO is increased because what is actually happens is that you gain the ability to record a little over 1 stop further into the shadows as you are now using more of the sensors low range (which is normally well below the black level chosen for 709) with the side effect of also seeing a little more than twice as much more noise (1 stop = 6db = double). The camera isn’t actually becoming any noisier, but because your using a lower sensor range you will see more noise in the shadows, noise that in normal gammas goes unseen. It’s really important that you understand this as it explains why S-log looks very noisy in the deepest shadows compared to standard gammas.
Native sensitivity… again this is open to a bit of wriggle room by the camera manufacturer. With a camera shooting log, generally it is a case of mapping the entire sensor capture range from black to white to the zero to 100% recording range. Normally this is done using as little gain as possible as gain adds noise. But as noise reduction processes get better, including on sensor noise reduction, camera manufacturers have some space to move the mapping of the sensor to the recording up and down a bit. Sadly or us, high ISO’s sell cameras. So camera manufacturer’s like to have cameras with high ISO’s because people look at the ISO rating, but ignore the signal to noise figure. The end result is cameras with high ISO’s (because it sounds cool) but with less than optimum signal to noise ratios. It would probably be better for all of us if we started paying much more attention to the signal to noise ratios of cameras, not just the ISO. That may help prevent manufacturers from bring out cameras with ridiculously high native ISO’s that are noisy and frankly far from what we need, which is a good low noise base sensitivity.
The next issue is that people appear to expect to be able to magically pull something out of nothing. If you have areas of deep shadow in your image you can’t magically pull out details and textures from those areas without significantly increasing the noise in those parts of the picture. You can’t do it and you shouldn’t be trying to do it. If you have an 8 bit camera the noise in the shadows will be really coarse, you try to stretch those levels, even by a tiny bit, it’s going to get ugly fast (the same with 12 bit linear raw too). What’s the answer…. LIGHT IT PROPERLY OR EXPOSE IT BRIGHTER.
We appear to have lost the ability to light or expose properly. If you want detail in your shadows either expose them brighter or throw some light in there, then take the levels down in post. Remember it’s all about contrast ratios. Faces are normally 1.5 stops above middle grey and 3.5 stops above our dark shadow range. So if you want a lot of textures in your deep shadows expose the entire scene brighter, not just the foreground but the background and shadows too. If you expose faces at +4.5 above black. Mid grey will still be -1.5 stops below those skin tones and your shadows will still be 3.5 stops below your faces. the contrast ratio remains the same if you increase the overall light level, so now everything will be 1 stop brighter. Then take the levels down by 1 stop in post and bingo, you noise levels are cut in half and your shadows look so much better and might actually now contain some useable picture information.
To follow on from this I recommend reading this: https://www.xdcam-user.com/2015/03/what-is-exposure/
Hi Alister,
The problem is confounding information on different forums. The camera manufacturers state a particular camera has a native iso, which has been calculated for the most appropriate exposure for best DR without clipping and crushing, according to the man’f, yet trial and error studies contradict this on numerous and frequent occasions.
What then is your take on when to use native iso and when to use the plethora of options.
Thanks,
Drew.
The native ISO is normally where you get the darkest thing the sensor can see mapped to black and the brightest mapped to 100%, normally without the use of additional gain as gain increases noise. But as noise reduction processes improve manufacturers are starting to add gain to the native ISO to achieve a higher base ISO setting. If correctly mapped you can add gain to the native ISO without loosing DR, but you do take a noise hit. The issue is compounded by the fact that if you then try to subtract gain from this 0-100% signal to reduce noise you will then loose dynamic range.
But the DR of cameras currently being touted, despite having ranges from 10 to 15, will all end up as 5 plus highlights, at least until 2020 is adopted. Is it then that a brick wall is hit with the maximum usable iris? Is it always a given that reducing gain / iso will allow one to shoot at a wider iris for a ‘correct’ exposure, but at the expense of additional noise in the shadows, even if ETTR?
Drew.
DR is not just about highlights. Its about RANGE from dark to light. Most of the newer cameras are extending the dynamic range both in the highlights and shadows. In addition using log eliminates the typical highlight roll off used in conventional cameras. By getting rid of the roll off you retain much more information in brighter parts of the shot and this allows post to extract much more information from the highlights and thus make decisions about roll off in post that simply can’t be done in camera. Reducing gain/ISO below the sensors native sensitivity often results in a loss of dynamic range.
Thanks again for another informative read Alister. Do you know the sensitivity of the FS7 in dB?…the correctly exposed 90% white card at 2000 lux? It’s not stated in the manufacturer’s specs. A figure would be useful to compare with other cameras (who’s manufacturer’s do quote sensitivity in dBs). Merry Christmas
Sony states the PDW850 as follows: Sensitivity (2000 lx, 89.9% reflectance) F11 @ 59.94i, F12 @ 50i (typical)
Sorry, Sony do state of th FS7: Sensitivity (2000 lx, 89.9% reflectance) T14 under the following conditions: Video Gamma, 3840 x 2160; 23.98p mode 3200K. Does that mean it’s 2 stops more sensitive?
Yes, but you must also consider the noise figure at the same time. A camera can be twice as sensitive but also twice as noisy, so in practical terms the useable low light performance would be the same.
The PDW-850 is 62 dB while the FS7 is 57 dB so the FS7 is 2 stops more sensitive (F12 to T14 is about 1 stop and 50i to 24p is one more stop) but almost 1 stop noisier so in practical terms the FS7 gives a 1 stop low light improvement over the 850, not a 2 stop difference which would appear to be the case if you look sensitivity on it’s own.
Late to the party on this one. Thanks for the indispensable and generous information as always. Alistair do you happen to know the Base/native iso of the original Sony A7s Mk1 when recording video not in the SLOG mode? Say…PP6 Cine4 for instance? The place where the sensor lives without manufacturers negative or added gain if in fact it does? Thanks Wilby.
It’s about 1600 ISO.
Thank you sir. Extremely helpful as always. Wilby.