How do you Determine Optimal ISO?

[Michael Pancier]

Just got my 5d Mark II.

While the camera can take great pics up to ISO 3200 (handy when the situation calls for it) --- how does one determine the optimal ISO for everyday shooting such as landscape, macro, etc. (non action shots)

On my 5d, I usually shot at ISO 200 cause I thought I read somewhere that was the native/optimal ISO for the camera.

I hear the 5d Mark II's native ISO is 6400. I'm not going to shoot at 6400 in daylight ....

so.....

do I just set it at auto ISO? (I usually use Auto when I handhold), but on a tripod when I'm shooting a landscape or controlled macro shot - am I better off at ISO 100 or ISO 200?

Just curious what the brain trust suggests......

[Alfred Forns]

Mike I can't imagine ISO 6400 being the naive resolution?

I used the lowest ISO I can get away with while keeping the necessary shutter speed. I don't hesitate cranking the ISO particularly while using the D3.

The auto ISO I do use in spot situations but its not on by default. Noise wise I don't think you will see much difference between 100-200 or 400 with your camera. Might want to try on the little point and shoot to test, huge difference going from 80 to 200 !!!

[Bob Blanchard]

I'm with Alfred on this. I use the lowest I can get away with to do the job. Technology has made leaps and bounds in this area, but ISO still equals noise (just not as much as it used to).

[Doug Brown]

I agree with Al. For me it's all about shutter speed, since I hand hold the big glass. But I'm constantly watching my ISO and reducing it to the lowest number I can get away with. It's unusual for me to drop below 400, but I mostly shoot avian flight. For landscape on a tripod, I'd go with 100 or 200.

[Grady Weed]

I have been told I push the outer limits by many photographers who look over my images. Some have asked my why I go so low on ISO and not gain on the shutter speed. I have always tried to get as much ISO and a balanced shutter speed as possible. Usually I set my 5D on AV, use ISO 100-400 when hand holding and almost always 100 when on tripod for landscapes. The shutter speed has for the most part not suffered or translated into a bad image sharpness wise.

I want a 5D Mark ll because of the video capabilities and the increased mega pixels. Finances hold me back for now. So, the current 5D is good enough for now. So far your test shots look good.

I have never used auto ISO. I was taught to use AV mode for most of the time. I am wanting to go fully manual to see what I can do with it. I think you will find the best results with Alfred's suggestion above. Then explore and post the results. I just can't imagine too many circumstances where anything above ISO 3200 is needed, unless you photograph for NASA on the lift offs! Then........o baby!

[Paul Wolf]

Michael:
If I interpret your question correctly, I think you are asking if it is better to go low or go native. I once read somewhere, too, that the optimal ISO on my 1D Mark III was 200. According to this article, 200 resulted in lower noise than 100. Well, I did some tests and according to my tests with my lens and the light and a host of other variables, I still found 100 to produce less noise than 200. The difference in real life shooting, however, is minimal if my exposure is correct. With my 1D, I can often shoot at ISO 400 or even 800 and still produce shots with noise levels lower than my 20D used to put out at ISO 200. Still, to be safe I still stay as low as possible.

So, as others have stated, I shoot as low as I can go that will allow me the shutter speed and f-stop I need for the situation. I don't auto-anything, so I can't comment on Av vs Tv or any other automatic setting. Also, as you may be aware, no matter what ISO you shoot, nothing beats the proper exposure for controlling noise and with today's equipment, sometimes that means shooting higher ISO than you would think just to avoid underexposing and having to compensate in post processing, which will always increase your noise.

Paul

[Michael Pancier]

My recent shots this past weekend were at ISO 200 and they were clean as a whistle as they say. I usually use in the past ISO 100 or ISO 50 if I want a long shutter speed (blur water for e.g.) ....

In the film days, I used Velvia so I shot at ISO 50 or ISO 100. Just wondering if I'm going to get a noticeable difference between 100 and 200 .... esp. on a tripod.

I just did some more online research and I guess what they are saying (again, not from official canon site) is that the native ISO is from 100-6400. (Whatever that means).

I'll play around at lower ISO's and see if I can see any difference.

[Leroy Laverman]

Native ISO is a rather confusing term and to really dig into it requires a discussion involving a lot of physics. The long and short of it really is - at what ISO setting is the S/N ratio the best. As an example take a look here http://www.luminous-landscape.com/re...-mkii-p1.shtml scroll to the bottom then move back to the top until you see the bar graph. For the 1Ds MK II ISO 100 gives the best (highest) S/N ratio. It's also interesting to note that at ISO 50 the S/N is similar to ISO 800. In the real world exposure and post processing can impact S/N as well. Most of the older cameras have a 'native' ISO of 100 but some of the newer models have improved sensor sensitivity so that the 'native' ISO is now 200.

[Judd Patterson]

For each of my cameras, I have a general starting ISO based upon a mix of native resolution and what I demand from the camera. I then also have an acceptable range where I move the ISO up and down without too much concern.

With my 5D, I typically start at ISO 160 because that is very close to what I've heard as the native resolution and with landscapes that is usually sufficient for a decent shutter speed...but I go up to ISO 800 with very little concern.

My photography style changes with the 40D, which I start at ISO 400. I seldom drop below that value because I am often looking for shutter speeds as fast as possible to reduce blurring/image shake. I will go up to 800 without much concern, but when I step beyond that I begin to worry about noisy shadows and pay particular attention to "shooting to the right".

These values are mostly a result of experience with the cameras. I can't say much for the Canon 5D Mark II, but my hope is that things are extremely usable up to 1600 or 3200 (seems to be your experience so far). I doubt the native ISO is much beyond ISO 200...if it were super high like ISO 6400 we would expect at least marginally degraded ISO 100 or 200 images compared to ISO 6400...which is just not the case from what I've seen.

[Maxis Gamez]

I think this year I got as high as ISO 800 5 times. 80% of my database are at ISO 200-400 with the 40D and 30D. Just my opinion.

[Pat Nighswander]

Shooting owls requires higher iso than I am used to wanting to use... but with the canon 30d.... this old camera.... noise starts at about iso 800

[Roger Clark]

Hello All. I just got back from a trip.

To answer the original question, there is no "native ISO" with digital cameras. The sensors in digital cameras have a Quantum Efficiency, QE, that tells us what percentage of light is converted to an electrical signal. Changing ISO does not change that. Changing ISO in a digital camera is changing amplifier gain(s) after the signal is read off the sensor.

The Signal-to-Noise ratio, S/N, is generally best at the lowest ISO the camera has. The luminous Landscape article on the 1Ds Mark II must have an error in the low iso S/N measurements. All properly tested digital cameras have so far shown S/N on digital cameras to be dominated by photon noise (except in the very darkest parts of an image or long exposure night images). That means that S/N will decrease as the square root of the increase in S/N. For example, images at ISO 400 will be twice as noisy as at ISO 100. Some high ISO jpeg output does not follow this trend because they do spatial averaging to reduce noise (you can see this effect as kinks in the dpreview noise versus ISO curves).

Digital cameras have what is called Unity Gain ISO: the ISO where the A/D converter is digitizing 1 electron for one unit increase in the output level. Think about it: you're measuring quantum level effects with a consumer digital camera!! (That's pretty cool for the physicists out there.)

For the smoothest images, highest S/N, use the lowest ISO, with one caveat. The very lowest on some cameras, like ISO 50 on my 1DII clips highlights. The lowest ISO should have been set at about 70. After learning this, I never use ISO 50 except in unusual circumstances and than am careful to lower the exposure to avoid saturation (in this case the camera will not properly let you know of the saturation).

More on sensors (technical):
http://www.clarkvision.com/imagedeta...rmance.summary

The other effect that may be of concern is dynamic range. Most digital cameras have dynamic range limited by the A/D converter rather than the sensor at low ISOs. So ISO 100 and 200 usually have about the same dynamic range on DSLRs, but the range decreases as you go up in ISO. You generally lose one stop of dynamic range for each doubling of ISO.

I am constantly changing ISO and use the lowest ISO I can get away with, but still go high (e.g. 800 or higher when needed).

Roger

[Sabyasachi Patra]

Michael,
I find that there is little difference between ISO 100 and 200 in my 1D II. I don't like the ISO 400, but am forced to use it especially in the early morning and late evenings. At times, I am forced to use the ISO 800 as without that it just becomes impossible to capture action.

I would click landscapes at ISO 100 on the tripod. Well, for most of the images I use support. I have realised that I am not like Doug Brown who can "shoot" with the accuracy of a sniper while handholding. :-)

Cheers,
Sabyasachi

[Sabyasachi Patra]

Michael,
I find that there is little difference between ISO 100 and 200 in my 1D II. I don't like the ISO 400, but am forced to use it especially in the early morning and late evenings. At times, I am forced to use the ISO 800 as without that it just becomes impossible to capture action.

I would click landscapes at ISO 100 on the tripod. Well, for most of the images I use support. I have realised that I am not like Doug Brown who can "shoot" with the accuracy of a sniper while handholding. :-)

Cheers,
Sabyasachi

[David Kennedy]

To answer the original question, there is no "native ISO" with digital cameras...For the smoothest images, highest S/N, use the lowest ISO, with one caveat. The very lowest on some cameras, like ISO 50 on my 1DII clips highlights. The lowest ISO should have been set at about 70. After learning this, I never use ISO 50 except in unusual circumstances and than am careful to lower the exposure to avoid saturation (in this case the camera will not properly let you know of the saturation)....

Roger,
What you could say is that the "extended" ISO modes shouldn't be considered in this discussion because the manufacturer's add then as an "extra," but do not consider them to be optimal. ISO 50 is noisier than ISO 100 on Canon bodies because it is part of the "extended" ISO...maybe that will change someday.

As for Michael's question, I would agree that for the 5D Mark II, the best ISO to use would be the camera's ISO 100. DxO Mark's test showed that ISO to have the best signal to noise ratio, as well as teh greatest Dynamic Range (11.16 EV) Interestingly, ISO 100 on the 5D Mk. II is really ISO 73. http://www.dxomark.com/index.php/eng...EOS-5D-Mark-II

And, like Judd, I remember reading E.J. Peiker writing that ISO 160 was the best for the 5D with respect to Dynamic Range. DxO actually suggests that ISO 100 yields its maximum dynamic range (10.77 EV): http://www.dxomark.com/index.php/eng...e/Canon/EOS-5D.

And good lord, I'm not apologizing if E.J. finds this and yells about me being "absolutely false" again. Read the data for yourself and draw your own conclusions.

[Roger Clark]

Roger,
What you could say is that the "extended" ISO modes shouldn't be considered in this discussion because the manufacturer's add then as an "extra," but do not consider them to be optimal.

Hello David,
Yes, I agree that extended ISO should not be considered in this discussion. The way I look at it is the extended ISO is there if you really need it but has reduced quality so isn't recommended.

ISO 50 is noisier than ISO 100 on Canon bodies because it is part of the "extended" ISO...maybe that will change someday.

But it's not noisier. The S/N is higher. The page you reference on the 5DII shows the S/N is higher too. Its extended because of other limitations, like clipping. For example, on the 1D mark III, ISO 50 is really the same as ISO 100, and you lose a stop at the high, is end easy to clip the highlights, and with no added S/N.

As for Michael's question, I would agree that for the 5D Mark II, the best ISO to use would be the camera's ISO 100. DxO Mark's test showed that ISO to have the best signal to noise ratio, as well as teh greatest Dynamic Range (11.16 EV) Interestingly, ISO 100 on the 5D Mk. II is really ISO 73. http://www.dxomark.com/index.php/eng...EOS-5D-Mark-II

I just got my 5DII today and will do a full sensor analysis. Calling the ISO is really this or that is technically wrong. The ISO specification has some leeway for manufacturers to implement the way they want. This was also true in film days with ASA. For more detail, see equation 3 at:
http://www.clarkvision.com/imagedeta...photons.and.qe
The constant, 12.4, varies between different manufacturers and this is the
reason we see different apparent metering levels. That is in the ISO spec.

And, like Judd, I remember reading E.J. Peiker writing that ISO 160 was the best for the 5D with respect to Dynamic Range. DxO actually suggests that ISO 100 yields its maximum dynamic range (10.77 EV): http://www.dxomark.com/index.php/eng...e/Canon/EOS-5D.

10.77 seems low as most other DSLRs do better (perhaps they have a different definition--no I checked it). I have often seen poor testing methods that results in scattered light which results in reduced measured dynamic range.

I looked at their test methods, and I suspect they have some scattered light that is limiting the dynamic range. I checked their 1D Mark III dynamic range and it is low compared to my tests at:
http://www.clarkvision.com/imagedeta...rmance.summary
I use methods equivalent to what sensor manufacturers do and is much more labor intensive than the 1 exposure tests used by others. The dxomark tests appears to use a single test chart, and scattered light from the bright spots probably compromise the dark test patches.

Roger

[Roger Clark]

10.77 seems low as most other DSLRs do better (perhaps they have a different definition--no I checked it). I have often seen poor testing methods that results in scattered light which results in reduced measured dynamic range.

I looked at their test methods, and I suspect they have some scattered light that is limiting the dynamic range. I checked their 1D Mark III dynamic range and it is low compared to my tests at:
http://www.clarkvision.com/imagedeta...rmance.summary
I use methods equivalent to what sensor manufacturers do and is much more labor intensive than the 1 exposure tests used by others. The dxomark tests appears to use a single test chart, and scattered light from the bright spots probably compromise the dark test patches.

Roger

I probably should explain the above a little more. The effects probably go back to film days. Did you shoot slide film? I did. But slide film had only about 7 or so stops of dynamic range. We seemed to do quite well with such limited range. But now everyone seems to want more and more dynamic range. But there are a couple of effects that limit achievable range. 1) High dynamic range scenes usually are often not great light, so we usually don't need high dynamic range (if we did we wouldn't have all those great images on slides). 2) The reflections and scattering in lenses reduce achievable dynamic range. So even as sensors get better, it will be more difficult to record it in actual scenes. HDR methods help this problem to some degree.

So recording high signal range is a problem when testing the dynamic range of cameras. The simple test is to use a calibrated test target. But depending on the optics, a small dark spot will be contaminated by reflections and scattering in the lens by brighter spots in the image. I've seen this effect in many tests using targets.

In the sensor industry, sensors are tested by doing a series of exposures such that one covers the entire range from saturation to no measurable signal. It may take 20 to 40 or more images to cover that range and a fair amount of time to analyze it. These are published methods in the electronics journals. It takes me about 40 hours of work to analyze one sensor. I'm working on software that will speed it up, so I can complete it faster. (I'm working on the 5D Mark II now.)

I work with spacecraft designers as part of my work and evaluate airborne and spaceborne imaging sensors as part of my job. I and other astronomers have been doing digital camera sensor analyses for a few years (Emil, a member of BPN is another such analyst). We've commonly seen DSLRs with greater than 11 stops of dynamic range, but I have yet to see one with greater than 12 stops.

Roger

[Robert O'Toole]

.....how does one determine the optimal ISO ..........

I think the answer here would be any ISO that allows you to make the image possible in regards to sharpness and DOF. Expose to the right and deal with any noise in PS.

Robert

[Joseph Martines]

This has been an interesting discussion on ISO which I really appreciate.

I read somewhere that the optimal ISO for the Canon 40D was 640. This fall I used 640 with good results shooting BIF.

Just recently I thought that perhaps I should be using a slower ISO and have started shooting 100 - 200 ISO to see what type of results I can get.

Some of this is common sense but, with the technical side being what it is we should be able to get more out of these new fangled devices called cameras.

[Bill McCrystyn]

Roger, while I think Robert perhaps explained what the best ISO would be in any given situation as most suggested, called horse sense, still leaves me in regard your noteworthy information with a couple more questions. (1) Is the noise in the S/N then coming from the amplifier (hence a quality issue) after the sensor? My D300 seems very quite. (2) Am I to understand that changing the ISO does not change the sensors sensitivity or electrical output? (3) OK, I lied, I have 3 for you. It was my understanding that we work with 8-10 stops of dynamic range on Earth and the best we could hope for (looking at my histograms) would be in the area of 5, hence all the blowouts and problems. What am I not understanding?

[Roger Clark]

Roger, while I think Robert perhaps explained what the best ISO would be in any given situation as most suggested, called horse sense, still leaves me in regard your noteworthy information with a couple more questions. (1) Is the noise in the S/N then coming from the amplifier (hence a quality issue) after the sensor? My D300 seems very quite. (2) Am I to understand that changing the ISO does not change the sensors sensitivity or electrical output? (3) OK, I lied, I have 3 for you. It was my understanding that we work with 8-10 stops of dynamic range on Earth and the best we could hope for (looking at my histograms) would be in the area of 5, hence all the blowouts and problems. What am I not understanding?

Hi Bill,

While what Robert said makes sense, it is still not clear in each situation what would be best. Although if one used the lowest ISO for a given situation that gave the exposure time desired and "expose to the right" as Robert says, it will work well in most situations. However knowing the sensor performance characteristics would help improve the results and in some situations it could be a major difference.

The better DSLRs all have very similar performance trade point, so below should apply to nearly all DSLRs.

On the sensor, one might trade 3 things for image quality (I'm assuming you got the exposure to freeze desired action, and depth of field you want):

1) maximum dynamic range,
2) maximum signal-to-noise ratio, and
3) maximum low signal detection while maintaining good dynamic range.

There are three main sources of noise that control what is best in the above conditions.
1) Read noise (noise from getting the signal off the sensor). This is VERY low, amounting to only 3 or 4 electrons.
2) Post read electronics noise, mostly in the the A/D converter electronics. Current cameras have only about 12-bits (even 14-bit system) and below that noise dominates.
3) Photon noise. This is the effectively ALL the noise you see in your images brighter than the darkest shadows from all cameras, from the Nikon D3 to the P&S with the smallest pixels. Photon noise is the fundamental physical limit. The signal-to-noise ratio (S/N) is the square root of the number of photons collected. This means the more light you collect, the better the S/N and the better the image quality. Larger pixels collect more light. The Nikon D3 has such wonderful S/N performance because it has the largest pixels of any consumer DSLR on the market, thus collecting the most light per pixel.

So the factors to consider.

1) Maximum dynamic range. If the camera A/Ds weren't so limiting, the lowest ISO might be best. Often the sensors saturate and clip signals. For example, the 5d Mark II ISO 50 is the same as ISO 100 just the exposure is halved, so you are more likely to saturate the highlights. Thus using "extended ISO range" at the low end harms dynamic range in most cases and with the added risk of blown highlights is not worth the risk, at least for me. As one moves up in ISO, from 100 to 400, on cameras with larger pixels, dynamic range is pretty constant. so use any ISO in that range. Dynamic range on most DSLRs is in the 11.5 stop range at low ISOs on a per pixel basis. On an image basis, one can coax 14 or more stops out of an image, even with 12-bit systems. E.g. see my article on exposure latitude:
http://www.clarkvision.com/imagedeta...ure_latitude-1

2) Maximize signal-to-noise ratio, Expose to the right and use the lowest ISO. Now I stop at the lowest non-extended ISO because of the problems discussed above. As you move up in ISO,the S/N drops with the square root of the increase, e.g. a 2x drop from ISO 100 to 400. This means if you have to boost shadows, the noise will be greater and more apparent especially as you increase ISO. I try and use the lowest ISO I can
and get the exposure so that the histogram is to the right as far as possible without clipping any highlights I want to keep. I try and do most wildlife and landscape work at ISO 100 or 200, but don't hesitate to increase it when needed.

3) Maximize low light detection. For example, a night shot when you are trying for an animal in the dark, perhaps with stars in the sky. Here is where you need to increase the ISO such that the noise from the A/D converter and other electronics is small compared to the read noise. For this you need to know the specifics of your camera. Most cameras are in this range at ISO 800 to 1600. There really is no need to go higher than ISO 1600 on any camera for which I have seen sensor test data. On 14-bit cameras, this point is usually ISO 800, e.g. like the 5D Mark II. On 12-bit cameras, 1600 may be slightly better. If you need faster exposure, it is actually harmful to dynamic range without any actual improvement in low light capability by increasing ISO. What I do when I need faster exposure is to use exposure compensation or switch to manual take the image then boost it in post processing. This certainly helps with stars as you saturate fewer stars and thus keep more star colors.
Of course the disadvantage of that method is that in the field you can't review the image. I may boost the ISO and take some test shots to make sure I'm getting what I want, then back off to ISO 800 or 1600.

Intermediate ISOs (other than 100, 200, 400, 800, etc) add a secondary amplifier which increases noise slightly, but not enough to be concerned about in my opinion. So use them if you desire.

I predict future generations of cameras with have good 16-bit A/D converters such that we may not need to define ISO in raw files. The entire sensor dynamic range will be well digitized and you can choose ISO in post processing much like you can now choose white balance. This is quite reasonable as scientific sensors have had 16-bit A/Ds for years and never need a gain (ISO) change. The problem is that high speed 16-bit A/Ds are expensive and take more power, but electronics are improving.

Does this answer your questions?
Roger

[Bill McCrystyn]

Thanks as always Roger. This answered many questions. Your image exposure lattitude article gave me the dynamic ranges I was use to seeing and where the rest came from. Image sensor size always made sense but understanding QE to A/D converter noise after read explained the rest. It was nice hearing about those Nikon sensor sizes. Like anything so often, you usually get what you pay for. Nice to know it isn't just the name. I have one last question about the A/D converters. In specific with my D300. If I am (I know you dissaprove) shooting 8 bit 36mb file image Tif's, do I get the advantage in that mode, of the 12/14 bit A/D system with my D300 for the noise advantage?? Many thanks for your time. Bill

[Roger Clark]

I have one last question about the A/D converters. In specific with my D300. If I am (I know you dissaprove) shooting 8 bit 36mb file image Tif's, do I get the advantage in that mode, of the 12/14 bit A/D system with my D300 for the noise advantage?? Many thanks for your time. Bill

Bill,
The images are gamma encoded, meaning it is not a linear conversion (this is true whether 8, 12, or 14 bit conversion) But with 8-bit you have less precision and that increases noise. Increased noise also reduces dynamic range. If you shot raw and converted to 16-bit tiffs, you should see less noise in shadows. This can be important in my experience when boosting shadows to show shadow detail.

Roger

[Bill McCrystyn]

Thanks Roger, I will consider it. Bill

[Emil Martinec]

To answer a variety of statements above:

1. For many DSLR's -- 1D3, 1Ds3, 40D probably 50D but I haven't tested it, all at ISO 100; D300 at "Lo" ISO -- the sensor saturates before the maximum raw level is attained at the indicated lowest "real" ISO (ISO 50 on the Mk 3's is actually ISO 100 pulled one stop, not a separate hardware amplification). Consequently the dynamic range can be higher (and is on all the above listed Canons) at ISO 200; there is more headroom, and the read noise doesn't drop enough between ISO 100 and 200 to compensate.
2. Consequently I would disagree slightly with Roger and say that a camera does have a "native" ISO -- it is the amplification at which sensor saturation occurs at the highest quantized raw level and not before. I agree with him that strictly speaking there is no defined ISO for RAW data, the ISO specification refers to converted images; nevertheless one can apply the analogous criterion to RAW data; this is for instance what DxOmark does in their ISO sensitivity test.
3. On Canon cameras, "intermediate" ISO's (the 125-250-500-1000 etc series, and the 160-320-640-1240 etc series) are a total waste of time, as I have explained in detail elsewhere http://www.naturescapes.net/phpBB3/v...p?f=2&t=139621. The upshot is that, for 1 series cameras and the 5D (don't know how these ISO's are implemented in the 5D2), the only effect of using the intermediate ISO's is a loss of 1/3 to 2/3 stop of highlight headroom; and for the xxD cameras these ISO's are implemented in software, again resulting in a loss of highlight headroom. Joseph's comment that "ISO 640 is best for the 40D" is one I have seen elsewhere on the net; the analysis on which is based is flawed, and the preceding link contains the correct version (spread across several posts in the thread) that leads to the conclusion that intermediate ISO's are deprecated on Canons (here again I differ somewhat with Roger). On Nikons, these ISO's are properly implemented, so there is no reason not to use them.
4. Depending on the camera, extended ISO's at the low end are often the lowest "standard" ISO, pulled by a stop. This improves S/N at the expense of lost highlight headroom, as in any ETTR situation. Canon Mk 3's at ISO 50 are an example of this.
5. Extended ISO's on the high end are usually the highest "standard" ISO, digitally amplified in software. This does nothing for RAW shooters except loss of highlight headroom, and so such ISO's should be avoided.
6. The ISO at which "the S/N ratio is best" is rather subtle. Here is the S/N for the 1D3 at fixed exposure (Tv/Av) in stops, at various ISO:
   (Note for experts: these curves were theoretically generated from measurements of read noise and photon shot noise; pixel response non-uniformity has not been incorporated.) Each curve ends at the point where the RAW data saturates, which is one stop lower in exposure for each stop increase in ISO. The upper end of the curves is controlled by photon noise, which is part of the signal and so independent of the ISO setting of the camera; thus all the curves merge when there is enough light to get away from the read noise of the camera electronics. The lower end of the curves is where the read noise becomes important, and there one sees a differentiation, as higher ISO actually has less noise relative to signal (otherwise there would be no reason to have ISO for RAW shooting, rather than pushing lower ISO). Note the diminishing returns with increased ISO; the gain from ISO 800 to 1600 is minimal, and frankly I don't think it's worth the loss of highlight headroom. The ISO 3200 curve (that ISO is "real" or hardware implemented on the 1D3) lies on top of the ISO 1600 curve, apart from ending one stop earlier on the highlight end; one does equally well staying at ISO 1600 and pushing in post-processing, with the added advantage of an extra stop of highlight headroom.
7. The way to read the above graph is that S/N rises with signal level, so it pays to expose to the right where the S/N is higher. Typically one needs to lower the ISO to gain enough headroom to push the histogram to the right without clipping. The optimal ISO for a given exposure (Tv/Av) is the highest one that preserves enough headroom not to clip highlights, though as the curves show, the benefit actually comes in the shadows in the form of increased S/N. On the other hand, lowering the ISO and increasing the exposure allows to use more of the right side of the graph with its higher S/N, and so if slower shutter speeds or wider apertures are allowed by the shooting conditions this always beats using a shorter exposure at higher ISO.

[Axel Hildebrandt]

Thanks, Emil! Very interesting read. In practice, would you suggest to use ISO 200, 400, 800 and 1600 on Canon cameras only? This is what I remember if the main objectice is to keep the noise as low as possible.

[Emil Martinec]

Thanks, Emil! Very interesting read. In practice, would you suggest to use ISO 200, 400, 800 and 1600 on Canon cameras only? This is what I remember if the main objectice is to keep the noise as low as possible.

Yes that's what I'd recommend. And 1600 is a close call over 800 and pushing in post. As for ISO 100, the S/N curve actually flattens out due to pixel response non-uniformity (which I didn't include, but which is included in similar plots on the DxOmark website), and so the gain in S/N at the high end is less than shown in my plot. That, plus the lower DR, mean I never use ISO 100 unless I want an intentionally blurred subject (eg waterfalls).

[Michael Pancier]

Yes that's what I'd recommend. And 1600 is a close call over 800 and pushing in post. As for ISO 100, the S/N curve actually flattens out due to pixel response non-uniformity (which I didn't include, but which is included in similar plots on the DxOmark website), and so the gain in S/N at the high end is less than shown in my plot. That, plus the lower DR, mean I never use ISO 100 unless I want an intentionally blurred subject (eg waterfalls).

So for us non technical folks, should we then be shooting above ISO 100 such as ISO 200 to get the least possible amount of noise presuming correct exposure?

[Emil Martinec]

So for us non technical folks, should we then be shooting above ISO 100 such as ISO 200 to get the least possible amount of noise presuming correct exposure?

Well, I did say it was subtle. If you have chosen your shutter speed and aperture (say for freezing motion, or desired DOF), and you can raise the ISO while preserving all the highlights you want preserved, then the higher ISO is better. If on the other hand you have the freedom to change the exposure, then increasing the exposure is always better (more photons!), and if necessary lowering the ISO to protect the highlights you want to preserve.

It's just that with my main subject matter (birds), I only rarely find myself in a situation where I can lower the shutter speed enough to make the slight gains of ISO 100 worthwhile (especially since ISO 200 on my 1D3 is so clean). YMMV!

Just to amplify a little ;), the histogram of the scene defines a window of exposure levels (scene DR) that are relevant to recording the image. One then wants to place that window on the S/N graph so as to maximize the S/N. Since the curves rise to the right, the best way is to move the window to the right, ie increase the exposure (ETTR), lowering the ISO so that the window doesn't fall outside the range of RAW levels recorded by the camera (channel clipping). If on the other hand the placement of the window is constrained by shooting conditions, then the second best way to improve S/N is by raising the ISO so that RAW saturation moves down to the top of the window of exposure levels; this improves shadow S/N without compromising highlights. Finally, current cameras show no improvement in shadow S/N beyond ISO 1600, so there is zero advantage to using higher ISO relative to pushing an underexposed ISO 1600 image (and the disadvantage of reduced highlight headroom with the higher ISO).

[Roger Clark]

Well, I did say it was subtle. If you have chosen your shutter speed and aperture (say for freezing motion, or desired DOF), and you can raise the ISO while preserving all the highlights you want preserved, then the higher ISO is better. If on the other hand you have the freedom to change the exposure, then increasing the exposure is always better (more photons!), and if necessary lowering the ISO to protect the highlights you want to preserve.

I'm sorry Emil, but I think you are adding a lot of unnecessary confusion. I think your plot is confusing to photographers, even though it is correct. For example, in the above above paragraph changing exposure is not the only way to improve S/N. You could open the aperture too.

Michael asked:
"So for us non technical folks, should we then be shooting above ISO 100 such as ISO 200 to get the least possible amount of noise presuming correct exposure?"

There is one clean answer. Presuming correct exposure for the ISO, the lowest ISO ALWAYS gives the highest S/N in the image. That is the answer I gave on January 2, above.

Your plot implies the S/N is not changing between ISOs at the high end. It does not change if one does not change exposure. For example, who meters at ISO 1600 then changes to ISO 100 and shoots with the exposure and aperture settings from ISO 1600? Photographers do not work that way. Digital photographers "expose to the right" and that means exposure and/or aperture changes when you change ISO. So you should redo the plot to illustrate what the S/N people get in their photographs when they've exposed to the right at each ISO. The result is not subtle at all. The S/N in the image, e.g. on an 18% gray card for proper photographic exposure at each ISO decreases with the square root of the ISO increase. E.g. from ISO 100 to 400, the S/N drops a factor of 2 (square root 400/200).

It's just that with my main subject matter (birds), I only rarely find myself in a situation where I can lower the shutter speed enough to make the slight gains of ISO 100 worthwhile (especially since ISO 200 on my 1D3 is so clean). YMMV!

Yes, that may be, and that is why I said that I often shoot at ISO 100 and 200. ISO 200 is still really good on DSLRs with large pixels. But it is lower S/N.

Just to amplify a little ;), the histogram of the scene defines a window of exposure levels (scene DR) that are relevant to recording the image. One then wants to place that window on the S/N graph so as to maximize the S/N. Since the curves rise to the right, the best way is to move the window to the right, ie increase the exposure (ETTR), lowering the ISO so that the window doesn't fall outside the range of RAW levels recorded by the camera (channel clipping). If on the other hand the placement of the window is constrained by shooting conditions, then the second best way to improve S/N is by raising the ISO so that RAW saturation moves down to the top of the window of exposure levels; this improves shadow S/N without compromising highlights. Finally, current cameras show no improvement in shadow S/N beyond ISO 1600, so there is zero advantage to using higher ISO relative to pushing an underexposed ISO 1600 image (and the disadvantage of reduced highlight headroom with the higher ISO).

Personally, I think you are again making it too complex. Its a simple rule: 1) expose to the right to maximize the S/N at a given ISO, and use the lowest ISO to maximize the S/N. Both are important in order to get the cleanest images.

As for ISO 100, the S/N curve actually flattens out due to pixel response non-uniformity (which I didn't include, but which is included in similar plots on the DxOmark website), and so the gain in S/N at the high end is less than shown in my plot. That, plus the lower DR, mean I never use ISO 100 unless I want an intentionally blurred subject (eg waterfalls).

I feel this is a misleading statement. While the S/N may reach upper limits due to pixel response non-uniformity in the highlights, the S/N is so high, who cares? What matters, and what people see in images concerns what noise you SEE in images? People see noise in the mid tones to shadows. And that is exactly where ISO 100 would show improvements over ISO 200 even including pixel non-uniformity. Wayne Cosshall has a good illustration in his noise tests assembled as a chart of gray levels versus ISO. For example, see his chart for the D3:
http://www.dimagemaker.com/article.php?articleID=1220
Notice how for the brightest gray patches, noise seems to disappear even at high ISO, but noise is very prominent at mid gray and darker gray patches.

So, again, lowest ISO gives the highest S/N images.

Roger

[Emil Martinec]

I'm sorry Emil, but I think you are adding a lot of unnecessary confusion. I think your plot is confusing to photographers, even though it is correct. For example, in the above above paragraph changing exposure is not the only way to improve S/N. You could open the aperture too.

Perhaps I am misusing the term exposure, which I take to be the amount of light hitting the sensor (same as in exposing film), which comprises both aperture and shutter speed. So the horizontal axis of the plot is essentially the amount of light hitting the sensor. So what is "correct exposure" for a given ISO? It is the one that bumps the upper tail of the scene histogram against the raw saturation point.

Michael asked:
"So for us non technical folks, should we then be shooting above ISO 100 such as ISO 200 to get the least possible amount of noise presuming correct exposure?"

There is one clean answer. Presuming correct exposure for the ISO, the lowest ISO ALWAYS gives the highest S/N in the image. That is the answer I gave on January 2, above.[/quote]

Yes this is correct, up to ISO 1600. Beyond that there is essentially zero improvement in S/N, but a loss in highlight headroom.

Your plot implies the S/N is not changing between ISOs at the high end. It does not change if one does not change exposure. For example, who meters at ISO 1600 then changes to ISO 100 and shoots with the exposure and aperture settings from ISO 1600? Photographers do not work that way. Digital photographers "expose to the right" and that means exposure and/or aperture changes when you change ISO. So you should redo the plot to illustrate what the S/N people get in their photographs when they've exposed to the right at each ISO. The result is not subtle at all. The S/N in the image, e.g. on an 18% gray card for proper photographic exposure at each ISO decreases with the square root of the ISO increase. E.g. from ISO 100 to 400, the S/N drops a factor of 2 (square root 400/200).

Well duh. The horizontal axis IS exposure; increasing the exposure (in the sense of the word I have been using) is the same thing as moving to the right on the graph.

[quote
Yes, that may be, and that is why I said that I often shoot at ISO 100 and 200. ISO 200 is still really good on DSLRs with large pixels. But it is lower S/N.
[/quote]

Yes, I was mistaken in saying ISO 200 is as good.

Personally, I think you are again making it too complex. Its a simple rule: 1) expose to the right to maximize the S/N at a given ISO, and use the lowest ISO to maximize the S/N. Both are important in order to get the cleanest images.

I would say if one wants a rule of thumb, it's ETTR up to ISO 1600 (or even 800, since the gains are really marginal above that) for best S/N; beyond that ISO threshold, keep the ISO fixed, underexpose and adjust in post-processing -- the S/N will be the same, and there is more dynamic range in highlights.

[Emil Martinec]

Gack! Sorry for the misplaced quotes (that will teach me to post w/o previewing :o)

Also, that last comment should have been

I would say if one wants a rule of thumb, for best S/N it's ETTR at the lowest possible ISO up to ISO 1600 (or even 800, since the gains are really marginal between 800 and 1600). By lowest possible ISO, I mean that a lower bound on shutter speed may be set by the desire to freeze motion (subject or camera); then one moves the histogram to the right by increasing the ISO, up to 1600; if that requisite shutter speed demands an ISO above 1600, instead keep the ISO fixed at 1600, underexpose and adjust in post-processing -- the S/N will be the same as "proper" exposure at ISO above 1600, and there is more dynamic range in highlights.

And again, Roger was right and I was wrong about ISO 100; it will have better S/N than ISO 200, when the exposure is increased to take advantage of the extra headroom.

[Bill McCrystyn]

It seems that practical expediency would dictate shooting BIF with large wingspans set to f/8-f/11 using AV and ISO 400/800, then assuming one maintains an appropriate shutter speed moves the histogram D/R window to the right by using EV comp would yield the best results. On an exceptionally bright day with room left over or initially presummed, then you may take advantage of the next stop down in ISO.

[Roger Clark]

It seems that practical expediency would dictate shooting BIF with large wingspans set to f/8-f/11 using AV and ISO 400/800, then assuming one maintains an appropriate shutter speed moves the histogram D/R window to the right by using EV comp would yield the best results. On an exceptionally bright day with room left over or initially presummed, then you may take advantage of the next stop down in ISO.

Bill,
I would say for BIF, the first thing to consider is if you want to freeze the motion, show a little motion, or a lot of motion. That sets exposure time. Next is for the given focal length and distance, how much depth of field is needed. Third, given the light levels will determine which ISO to meet those requirements. I personally, will compromise depth of field a bit if the ISO needed is getting too high. For example, I would rather image at f/8 and ISO 400 than f/11 and ISO 800. It would probably elect f/5.6 and ISO 200 first and go to slower f/stops only if I needed the depth of field.

Roger

[Roger Clark]

Ok, after this exchange, I thought it interesting to go through my bird gallery:
http://www.clarkvision.com/galleries/gallery.bird
The majority of the images were done at ISO 200, next (perception; I didn't actually count) 400, then 100, then 800. (There were a few 320s). I didn't notice anything higher than 800.

Roger

[Bill McCrystyn]

Ok, I suppose this will be some what subjective and more eyes may weigh in but -

To my thinking more resolution trumps shutter speed both as a visul requirement and practical necessity.

Looking over your images which I enjoyed, I think I found two good examples of my point. From a shutter speed standpoint these were two of your slower shots. A great deal of them were much faster, some in the 1/4000 range and taken at a much wider (wide open) aperature.

The Brown Snake Eagle was taken with a 500 f/4 BUT shot at f/8 because of the 2X used and was allowed at ISO200. Not only is this a testament to the 2X but perhaps even more so to the resolution achieved at f/8 which in my opinon was one of your best shots for sharpness of a large wingspan raptor in your group. Not only has f/8 saved your DOF but has added resolution not available at f/4. It would also be noted that it was one of your slowlest shots at 1/640. So often the S/S required is over estimated. Sometimes with BIF small movement in the quicker moving parts of the anatomy can actually add to the image and most of the time at worst, will be over looked. On the other hand resolution/sharpness is never over looked as many of us fall into the softness trap all too often. The best shots are tack sharp.

This brings me to the second image. The Black Shouldered Kite. A much smaller raptor and much, much faster. Here you managed a 500mm and 1.4X at f/5.6 1/800 @ ISO100. While DOF was less of a concern reguarding size, 1/800 was barely adequate and still did not completly stop the action. It still regardless rendered a very pleasing image and f/5.6 was adequate. I think perhaps however that you would have been better served at ISO200 giving you f/8 maintaning 1/800 or even as much as ISO400 rendering f/8 at 1/1600, still far slower than most of your other images and providing even more resolution stlll yet.

These ovservations are not in anyway to detract from your wonderful images but only to point out that better resolution trumps the need for speed in my opinion. While they are fast moving birds, they are not speeding bullets and most usually can be captured at the lens receptical especially with the aid of the vibration reduction used today.