Exposure basics, lesson two point one (& ISO)

Started Mar 19, 2013 | Discussions thread
Shop cameras & lenses ▾
Anders W Forum Pro • Posts: 21,465
Re: Exposure basics, lesson two point one (& ISO)

Ulric wrote:

Jack Hogan wrote:

He wouldn't be naive to believe that he saw more noise (he would). He would be naive in thinking that to capture the same tonal range (shadows and highlights) of the same scene with the same lens and Exposure (ss and f/n) as his friend's EM5 he would have to use the same ISO in the G3. His naivete would cost him a ton of IQ: one stop of highlight headroom and 1 stop of SNR, and he possibly would lose his National Geographic desk

I give up; that is, I don't get it. Given the same exposure, why would the g3 on iso 3200 produce more noise than the em5 on iso 3200, but the g3 on iso 1600 would not?

The short answer: Because ISO labels are arbitrary, and in the case at hand Olympus called 6400 what Panasonic calls 3200, so if you set the G3 at 3200 you need to set the EM5 at 6400 to capture the same image information all else being equal. Pehaps an example would help?

You are an intermediate photographer and you have been given an assignement: shoot a game indoors with the same 50mm lens and two cameras, an EM5 and a G3. The desk tells you that anything other than a dof of f/6.3 and the blur stopping power of 1/800s will be accepted, and you need to get the highest IQ possible without blowing the highlights of the brightest object in the scene, the white helmets of the players. Easy, you say. You switch both cameras to full manual mode in order to set the mandated shutter speed and aperture on both - Exposure has therefore been fixed equally on both and will remain set there for the entire game. Then the issue of ISO comes up.

Quick like a fox you remember various DPR threads where this was discussed ad nauseam. You therfore take a sample shot at a guesstimate ISO of 800 on the G3 and see by looking at the hystogram* that the white helmets are more than 1 stop below clipping. You therefore dial in 1600 ISO take another sample shot, verify that the helmets are now all the way to the right of the hystogram just before clipping, and you are all set. You can now fire away, forgetting completely about Exposure and ISO and metering potentially getting fooled by lights or reflections and concentrate solely on getting the shot. You get some great shots.

At half time you move the lens to the EM5 and go through the same set up procedure. You verify that ss and f/n are the same mandated values of f/6.3 and 1/800s (therefore Exposure is the same as on the EM5) and you go through the exact same ISO setting procedure as for the G3. This time though, in order to ensure that the white helmets end up just before clipping you need to dial in a different ISO value. No sweat, the sample images from the two cameras look about the same in terms of noise and eDR as set up (the EM5 a tad better as DxO's graphs suggested) so you fire away forgetting that Exposure and ISO even existed. You get some even geater shots.

Same scene, same light, same lens, same Exposure (ss and f/n), same tonal range (image information) captured. Why are the ISOs different? You then remember DPR threads that mentioned that in-camera ISO labels are not consistent from camera to camera or from manufacturer to manufacturer. So what one calls 1600 the other can call 800 or 3200, the standard allows them to do it. Does it matter? No, as long as you remember that in the same situation when the G3 calls for ISO 1600 that is instead called 3200 on the EM5.

Let's assume, as in the above example, that the G3 and the E-M5 are shot at the same exposure (same light, same lens, same f-stop, same shutter speed). Then consider the following questions and answers.

Question: If the cameras are set to the same camera ISO, say 1600, will the G3 clip the highlights at a lower light level than the E-M5?

Answer: Yes. It follows directly from the DxO "measured ISOs" of 1481 and 782, respectively, that the G3 will start clipping about one EV before the E-M5 does.

Question: Is it, in this particular scenario, important to ETTR for optimal image quality, i.e., make sure that the brightest hightlights of the scene (in this case the white helmets) are brought right up to the clipping point of both cameras, which in this case could be accomplished by keeping the G3 at ISO 1600 and moving the E-M5 up to ISO 3200?

Answer: No. The implication of keeping both cameras at ISO 1600 is that the G3 will express the light levels on an ADU-level scale from 0 to about 4000 whereas the E-M5 will express them on a ADU-level scale from 0 to about 2000. The loss of numeric precision (amount of quantization error) by using 0 to 2000 instead of 0 to 4000 is in this case of little importance since the noise levels at the chosen exposure (i.e., an exposure such that it won't clip the highlights at ISO 1600 on the G3) is too high for that to matter. The read noise (as measured in electrons) is constant on both cameras from about ISO 1600 on so there is no advantage from that point of view to increase ISO on the E-M5 any further. If anything, the E-M5 would enjoy a slight benefit by keeping ISO down to 1600 because it reserves one EV worth of extra highlight headroom, e.g., for specular highlights on the white helmets.

Question: When, then, is it important to ETTR, if, as in this scenario, ETTR is accomplished by raising the ISO rather than by increasing the exposure?

Answer: When raising the ISO brings substantial benefits in terms of reduced read noise (as measured in electrons). Suppose that instead of the choice between ISO 1600 and ISO 3200 on the E-M5, we were considering the choice between ISO 200 and ISO 400. In this case, it would bring substantial benefits in terms of shadow noise to ETTR at ISO 400 rather than to remain at ISO 200 and stay one EV shy of the clipping point. The reason is that read noise is nearly cut in half by going from ISO 200 to ISO 400 on the E-M5.

Just thought that I would point out that the Dynamic Range is maximum at base ISO setting (with the E-M5, and in general). Yes, if one is shooting hand-held without mechanical stabilization then it is desirable to be able to use a higher Shutter Speed. That is where the relatively small decrease in DR (-0.2 stops at ISO=400 compared to ISO=200) might well be worth the DR trade-off involved.

However, that (Shutter Speed advantage) seems like something of a separate (but potentially irrelevant) issue to me, to be differentiated from matters strictly surrounding overall SNR. The kind of test-shots being discussed are presumably shot from a tripod (where Shutter Speed is not an issue).

(Assuming that the E-M5 is as linear as you believe that it is at base ISO, and that is not an issue, etc.), then the Read Noise will be lowest relative to the (we assume effectively fully utilized, allegedly thanks to the E-M5's preview "blinkies"? ) Full Well Capacity at base ISO=200 (not at ISO=400).

For the same Scene Luminance and lens-system T-Number, the Photon Shot Noise will be lower at base ISO=200 than at ISO=400 (by the square-root of two) because the Shutter Time is twice as long. Thus it is base ISO=200 overall SNR which would be the highest. Does that make sense to you?

Certainly. Remember that my entire post is written, as indicated at its top, under the assumption that exposure is kept constant.

For the same Scene Luminance, then the camera's F-Number would have to change to compare, right ? So, the implicit operating assumption is that a higher F-Number is better in the application ?

No, in the example described by Hogan, the f-number doesn't change. The choice he outlines is between shooting the E-M5 at ISO 1600 and not reach the ETTR criterion or shooting it at ISO 3200 and reaching it.

If exposure can be increased without negative side-effects, it is of course better to stay at ISO 200 and increase exposure to reach the ETTR criterion than to keep exposure unchanged and go to ISO 400. However, if exposure cannot be increased, it is better to go to ISO 400 and ETTR rather than to remain at ISO 200 and stay one EV short of ETTR and that was the point I was making here.

How would it be (assuming that the sensor is linear, and that the entire Full Well Capacity is effectively utilized) that losing 0.2 stops of DR (by switching from ISO=200 to ISO=400) is better. What matters is the ratio of the Full Well Capacity divided by the Read Noise. Make sense ?

If exposure is kept constant and you are one EV short of the ETTR criterion at ISO 200, you don't lose 0.2 stops by going to ISO 400. Instead, you gain 0.8 stops. The full DR at ISO 200 is only realized if the sensor is saturated. Such is not the case in the scenario (originally outlined by Hogan) that I am talking about.

Question: How, then, can we know whether read noise is reduced or not by increasing ISO at a certain point on a particular camera?

Answer: By checking out the read-noise figures reported by Sensorgen (bobn2) here


Click on a specific camera model to see its read-noise values for different ISOs. Note that small differences may be due to measurement error and that the values become inreasingly unreliable at higher ISOs due to problems of measurement. In general, you should look for the point at which the read-noise values no longer tend to fall significantly when doubling the ISO.

If results for a certain camera are not (yet) available at Sensorgen but reported by DxO, you can find the essential information via the DxO DR curve. If the loss of DR is less than one EV when you double the ISO, read noise is still falling. When the loss reaches one EV when you double the ISO, it is no longer falling.

While Sensorgen and DxOMark both report the results for the main ISOs (100, 200, 400 and so on), they don't report them for the intermediate ISOs (except for the base ISO). As a rule, that doesn't matter much but exceptions exist. With the E-M5, for example, it is advisable to skip the intermediate ISOs between 200 and 400 since these don't bring the same benefit in terms of read-noise reduction as ISO 400. See here for details:


 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +28 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow