"Native" vs. "Non-Native" ISO Settings ... Does It Matter?

[Paul Guris]

I have seen multiple people on multiple sites stating that "native" ISOs (400, 800, etc.) are less subject to noise than "non-native" ISOs (560, 640, etc.). This recently came up on this thread (http://www.birdphotographers.net/forums/showthread.php/82037-American-Robin?p=654602). Sid Garage, the man with the awesome avatars, stated the following (quoted with permission):


Little bit I know from my previous research(non professional of course)
ISO value is related to amount of amplification performed on the signal from the photosite (basic element of our camera sensors). Lower the ISO less amplification and higher the ISO more amplification. More amplification results in more noise. IMO there is no concept of non-traditional ISO and traditional ISO. Maintaining lower ISO will yield better results. Coming from a general tech background, this is the way I understood it as well. At a recent Nikon Photography School class I attended, I asked one of the instructors the same question and he had pretty much the same answer as Sid. I view it more like a volume control. Turning a volume knob so it lines up with a number isn't intrinsically superior to stopping in between numbers. It seems that if "non-native" ISOs were inferior, then the sensor would have to be created in such a way that something electrically different occurred at full ISO stops.

Can anybody here help definitively settle this issue?

[Flavio Rose]

All the information I have comes from this article:

http://www.guillermoluijk.com/article/isos5dmkii/index.htm

It is written in Spanish; the Union Jack icon on the top left gives you an automatic translation. The bottom line conclusions are as follows (my correction of the automatic translation):

"As a rule of thumb and final summary on the use of different ISO values ​​of our bodies:


If you shoot in RAW, do it only at real ISOs and up to ISO1600, so that the remaining intermediate ISOs are all discarded unless they are real in your camera. You should use the lowest ISO which exposes the histogram to the right in RAW, but stopping at ISO1600.
If you shoot in JPEG: do it at the ISO that best suits the shot among all those available on your camera, whether real or forced. You will thus use the lowest ISO which achieves correct exposure in the JPEG."

The article says that Emil Martinec has written on the topic; he is American and writes in English so his writings may be more accessible.

I seem to remember that intermediate ISOs are "real" in Nikons but not in Canons. However, I have never done any experiments with my Canon bodies to confirm that they are not.

[John Guastella]

Have you seen this article from Digital Photo Pro? The Truth About Digital ISO (http://www.digitalphotopro.com/gear/cameras/the-truth-about-digital-iso.html).

Also, there is an interesting video (mentioned in the Digital Photo Pro article) regarding ISO setting and noise levels on the Canon 7D. Canon EOS 7D Noise Test (http://vimeo.com/10473734).

John

[Daniel Cadieux]

I've used, and still use, "non-native" ISOs on my 40D and I'll be darned if I can see any difference in noise with my naked eyes in real world situations. I'm sure if I "pixel-peep" at magnifications I'd never use otherwise I may be able to discern some slight differences...and if that is what it takes to see a "possible" difference then I won't worry about it...just use the ISO that best gives you the SS/DOF combo that you need:w3.

[Flavio Rose]

More info here (and in English):

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html#SNR-DR (http://theory.uchicago.edu/%7Eejm/pix/20d/tests/noise/noise-p2.html#SNR-DR)

I would note that some of this seems to be based on hypothesizing how the sensor is constructed, and I am skeptical of analyses based on such hypotheses. It makes more sense in my view to experiment with various settings and examine the resulting raw files, as I think Luijk has done. The issue is at any rate not a big one.

[John Chardine]

I think there's a thread on this in BPN. I remember something about "native" ISOs being processed at a lower level in the camera than the intermediate numbers. I agree with Daniel that this may make little if any difference in real terms to the image delivered to you from your camera.

[Doug Brown]

I don't hesitate to use in-between ISOs. I try to get things right in the camera, and if ISO 500 gets me there, then so be it. :S3:

[William Malacarne]

I agree with getting it right in camera...but in this case, at least with my thinking. In post processing is better. If say the camera is set at 1000 ISO what is really happening is you are shooting at 800 ISO and the camera extrapolates data to 1000 ISO. It is a global correction. Now if I shoot the picture at 800 ISO, I can make exposure compensation in a localized manor and not globally. I have more control of the output.

Bill

[Roger Clark]

More info here (and in English):

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html#SNR-DR (http://theory.uchicago.edu/%7Eejm/pix/20d/tests/noise/noise-p2.html#SNR-DR)

I would note that some of this seems to be based on hypothesizing how the sensor is constructed, and I am skeptical of analyses based on such hypotheses. It makes more sense in my view to experiment with various settings and examine the resulting raw files, as I think Luijk has done. The issue is at any rate not a big one.


Flavio,
Emil is a BPN member although I have not seen posts from him recently. He does VERY good work, and his analyses are very high quality.

The basic idea is that there is an amplifier on the sensor that sends the data to the analog to digital (A/D) converter (or ADC). The on-sensor amplifier has only one gain. The ADC has a number of gains. One some cameras, the ADC gains are factors of 2, corresponding to ISO 100, 200, 400, 800 etc. Then a second amplifier (or done digitally) amplifies the signal from the powers of 2 amplifier. For example, ISO 130 would be the gain at ISO 100 followed by a 1.3x amplification. In some cameras, if you wanted ISO 160, you might get ISO 200 gain times a 0.67 gain. In that case you lose 1/3 stop dynamic range. The second amplification step can add more noise.

But on other cameras there are more gains than just factors of 2 and maybe no second amplifier.

In either case the increased noise effects are very small (unless you are really concerned about 1/3 stop loss in dynamic range). One probably can't notice the difference in real-world images outside a testing lab.

I wouldn't worry about it, although I do tend to use powers of 2 ISO and let the shutter speed vary by the 1/3 stop ot so as needed. Although on the chance that a camera only has the powers of 2 gains, once up in the ISO 800 or above range, it might be best to stick with powers of 2 ISO to avoid 1/3 stop loss in dynamic range.

Roger

[Alan Lillich]

I've heard this in various ways, none terribly convincing. This is one case where I abandon my usual scientific bent and channel Artie Morris: "Can I see a difference in middle ISOs?" No, I can't and unless someone shows clear evidence I will happily use all ISOs in 1/3 steps. I started to use only whole native ISOs like Roger, then decided I would rather have the flexibility to get the shutter speed that I really want even if it means some (invisible to me) loss of dynamic range.

I've usually heard that middle ISOs are more of an issue with Canon than Nikon. No slam to Canon meant - I use both and use all ISOs with both. I've heard rationale somewhat like what Roger said, that Nikon's A/D amplifiers do the 1/3 steps and Canon uses a mix.

I've also heard one case of using one stop multiples of non-native ISOs. Vincent Laforet was on the Photofocus podcast #67. At 46:30 in the midst of answering a question about noise reduction for video he said he only uses whole stop multiples of 160 for Canon bodies - 160/320/640/1250/2500. For video he limits the 7D to 640, 5D-II to 1250, 1D-IV to 2500. He "swears for a fact" that 320 is cleaner than 200 and sometimes even 100, and that 1250 is cleaner than 800.

[John Chardine]

Well 2 actually:
1. Is the 7D test that is referred to above by Lorentzen valid?
2. Is it worth trying to duplicate the test at home? If so how would you do it?

Thanks.

[Alan Lillich]

I would also like to hear Roger's take on the infamous 7D test, if it is the same one I'm thinking of. That test was a guy who put on the lens cap and took shots at all ISOs. Yes, a totally dark image. Sure all the signal then is noise. But I see a whole lot more value in a more realistic test, say using good quality color charts. Who knows how the read and amplifier noise changes between a zero signal and a more typical signal - still at the low end but not absolute zero. I also don't think the dark test works at all on Nikon because they don't apply a low end bias like Canon.

[Roger Clark]

Well 2 actually:
1. Is the 7D test that is referred to above by Lorentzen valid?
2. Is it worth trying to duplicate the test at home? If so how would you do it?

Thanks.

Hi John,
Regarding question 1, I have not seen any data either way on video to know if it is correct. I suspect not based on still image raw data. But video could use a different A/D (e.g. needed to get faster conversion rates).

2) Video modes average many pixels together, then apply a tone curve and downsample to 8-bits, so the output is complex and not like a raw file from still image mode. But one could analyze pairs of video frames (subtract pairs of frames and analyze the noise) to see which is better, just like doing it with raw. What one may find though, is that below about ISO400, the output is limited by the 8-bit conversion so not much difference between them.

To test sill image mode, take pairs of images on a boring subject, like a white wall. Set the camera up on a tripod and illuminate the wall with an incandescent light, set the focus to manual, and make the wall out of focus, so there is no fine detail. Take a pair of images at each ISO, setting exposure to the right and not saturated (e.g. white wall at meter +1 stop) and not too close to the right on the histogram. This will likely show a smooth increase in S/N following a square root increase with decrease in exposure, as the system noise will be photon noise limited. That means regardless of ISO and post A/D gains, a lower ISO giving a longer exposure will produce better S/N.

Another test would be the darkest "shadows" where post amplifier gains would show. Lens cap on, no lights, and take pairs of exposures at each ISO. Subtract the images and analyze noise and plot noise versus ISO. If there is a stair step trend, there is a second amplifier involved. But my bet is that in any real world photo, including long exposure astrophotos, you couldn't tell the effect of in between factors of 2 ISOs.

The third test is image a scene with exposure an compensation series, e.g. meter 0, +1/3, +2/3, +1, etc intil the image is driven to saturation. Do that at each ISO. Do intermediate ISOs saturate 1/3 stop sooner? If so, that is evidence of the second stage ISOs being made from the output of the A/D.

There are also some cameras that make high ISOs by multiplying the data from the low ISO data, e.g. digitally. For example, ISO 3200 is made from ISO 1600 and multiplying by 2.

But in the field, if you need it, use it is my opinion. Noise in images at all levels except the deepest shadows is dominated by photon noise. To minimize the apparent noise, and get the highest S/N one needs light, which means longer exposure. So use the lowest ISO you can get away with to avoid blur (unless you want blur).

Roger

[John Chardine]

Thanks Roger. This at least gives me an idea of what is involved. Pretty technical stuff!

[Roger Clark]

Thanks Roger. This at least gives me an idea of what is involved. Pretty technical stuff!

In my opinion, the fact that detailed analyses lake that done by Emil can show the fine details down to a few electron level illustrates what an astounding level digital cameras operate at. Such performance was hard to imagine a decade or so ago, especially for uncooled sensors operating in harsh field conditions outside a lab. And cameras have only gotten better since those done in Emil's analysis.

Roger