base or 'native' ISO test (7D)

Discussion in 'Canon EOS' started by gooseberry, May 13, 2011.

  1. Since I dislike murky shadows, whenever possible, I'd ETTR @ ISO 100. Now I have just learnt about sensors' base or "native" ISO and that it may not be the lowest possible setting. The problem was that different people were claiming different 'base ISO' values; the most convincing ISO noise test I found inspired me to run my own...
    ...so I left the lens cap on, switched the lights off, and shot four sets of 4 RAW images, exposing for 1s @ ISO 100, 125, 160, 200.
    Then I applied the most extreme settings in ACR to bring out the noise in otherwise pitch-black images...
    • Exposure @ +4.0
    • Fill Light @ 100
    • Brightness @ +150
    • left all other settings at their default values (including Contrast @ +25)
    ...averaged each set in PS (loaded as layers, converted to a smart object, and set stack mode to 'mean') and applied 'Strong Contrast' curve plus maxed out Brightness / Contrast adjustments (150 and 100 respectively). Finally, made 700x700 crops from the centre @ 100% magnification.
    I realize such post processing is unrealistic so when comparing real life pictures the differences would be far less pronounced, if at all noticeable, but that wasn't my point; I was interested solely in finding out for myself what the difference would be and whether in fact 160 would yield cleaner shadows than 100. That's as "scientific" as I'd like to get. Results attached FYI.
    Think file sizes may also be quite informative as a proxy for the amount of information:
    • ISO 100 => 884KB
    • ISO 125 => 944KB
    • ISO 160 => 800KB
    • ISO 200 => 1,012KB
    • as a reference, same size and compression (10) of pure black => 48KB
    From now on, 160 is going to be my default ISO setting (additional bonus being that some extra speed can't hurt :)
    00Yita-358261584.jpg
     
  2. Very interesting effort. Congrats for some very intuitive thinking and process design. Clearly, for this sensor,ISO 160 is optimal of the ones you selected. I wonder what it is for other sensors. This would be a very useful bit of information for manufacturers to supply.
     
  3. Intermediate ISO's are not true ISO selections. 125 is a pushed 100. 160 is a pulled 200. It's no surprise that 160 is clean in this test. But what about a test for highlight clipping?
     
  4. Rob,
    I thought all ISO's, other than the "native" ISO was pushed (or pulled).
     
  5. Don't confuse a change in gain with a push or pull in software. They are not the same.
     
  6. I've read that only each doubling of ISO is an actual hardware level ISO, ie:
    100, 200, 400, 800, 1600 and so on.
    And all the others are adjusted from the above. Basically echoing Rob, I know. Anyway, I try to stick to the above, basically because it's a fine enough adjustment for me. Just to keep my shutter speeds up I almost invariably use ISO 400, fwiw.
    At least in part due to my deeply ingrained Tri-X mindset, I guess ;)
     
  7. I don't know how this plays out on your camera, but on mine any real world difference in actual final photographs due
    to choosing whatever one might regard as "base ISO" versus something a stop or fraction of a stop away varies
    between trivial and insignificant.

    On my 5D2 I cannot consistently tell the difference between ISO 100 and 200 shots, and I use either as needed
    without further concern.

    Dan
     
  8. Rob Bernhard wrote:
    Intermediate ISO's are not true ISO selections. 125 is a pushed 100. 160 is a pulled 200. It's no surprise that 160 is clean in this test. But what about a test for highlight clipping?
    I've read about pushing/pulling, but from the base ISO, which can be any value (subject to chip's design), not necessarily the 'round' ISOs. If 200 was the "true" ISO +1EV, and if pulling it by -1/3 EV yields such a dramatic improvement giving results superior to the "true" ISO of 100, why create 125 as 100 +1/3 EV instead of 200 -2/3 EV -- it should be better than 160 -- and maybe even 'engineer' 100 as 200 -1EV?
    I'm just speculating here based on my uneducated 'common sense' because I'm struggling to explain the results I got based on this theory. Apparently, ISO 50 on EOS 1Ds MkII is pulled ISO 100, yet it provides worse S/N; according to Bob Andersson of Camera Labs, more surprising, perhaps, is that using an ISO number below the native ISO number also degrades the image. The reason is that you are effectively telling the camera to blow the highlights as the brightest elements of the picture will get to full-well before the exposure is over. It gets worse because the amplifiers are now set to a lower than optimal gain meaning that, as I understand it, shadow detail can also be lost into the noise threshold of the electronics.

    Found an academic paper on noise by Emil Martinec of the University of Chicago that explains some science behind the whole thing, yet couldn't find anything on "native" ISO (but haven't read it properly).

    Some more Internet search uncovered The Truth About Digital ISO DPP article (which, funny enough, refers to the same video and suggest similar test to the one I came up with ;) No definite answers are provided, but they clearly believe that both pushing AND pulling come at the expense of IQ: To achieve higher ISOs, gain is applied and this increases image noise. [...] ISOs below the camera’s native one also are manipulated, and this, too, results in reduced image quality (generally in terms of reduced dynamic range and even increased noise in dark areas).
    Mendel Leisk wrote:
    I've read that only each doubling of ISO is an actual hardware level ISO, ie: 100, 200, 400, 800, 1600 and so on.
    In their other article, Digital ISO Speeds, Mike Stensvold writes... A digital image sensor has a native ISO. When you adjust the ISO setting, the sensor itself isn't changing; rather, the amplification is changing ...so I understand that there is only one native ISO, whatever the number happens to be -- no change to the hardware takes place when we adjust ISO by whatever increment.
    He then goes on to say... Image sensors have an innate “native” sensitivity, generally in the ISO 100 to 200 range. When you set a higher ISO speed, amplifiers in the image sensor's circuitry increase the gain before sending the image data to the A/D converter to be digitized. The sensor's sensitivity doesn't actually increase; the camera is just amplifying the data it produces. In the process, image noise is also increased, making the image “grainier”—sort of like what happens when you “push” film speed [...] If you set a lower ISO speed than the sensor's native sensitivity, the camera's image processor adjusts the image data after the A/D converter converts it to digital form. In the process, the dynamic range is reduced. So it's best to shoot at the sensor's native ISO whenever possible ...but he doesn't tell us how to determine what this native ISO is beyond shoot the same scene at different ISO settings and see which results you prefer.
     
  9. Don't confuse a change in gain with a push or pull in software. They are not the same.
    What Rob said.
    Intermediate ISOs I find rather silly.
    On the 7D, ISO 200 is also *super* clean.
     
  10. Ken Papai wrote:
    Don't confuse a change in gain with a push or pull in software. They are not the same.

    What Rob said.
    Intermediate ISOs I find rather silly.
    On the 7D, ISO 200 is also *super* clean.​
    Deep shadows aren't as clean as I would have liked them to be (which is not say that switching from ISO 100 to 160 is going to improve things much, but guess it's worth trying.)
    I don't even understand Rob's cautioning, so it's safe to assume I confuse the two, but that's not what I was trying to clarify for myself. All I wanted to do is to avoid intermediate ISOs -- which you find rather silly yourself -- presuming that the base ISO will produce better results, even if in most cased the difference is not noticeable. Thus the question to answer was how to find out what the base ISO of my camera is; manufacturers don't disclose this info in tech spec, so I devised my own methodology, which kept me entertained for an hour or so :) and gave me the results that, rightly or wrongly, appear conclusive to me.

    @ Rob: testing for highlight clipping sounds far more difficult to conduct. Besides, don't want to put DxO out of business :p
     
  11. Intermediate ISO's in-camera are not achieved by adjusting the gain on amplifier. The only time the gain is adjusted is when moving whole ISO stops. 125 is the same amplifier gain as 100 and moving the exposure slider in lightroom by +1/3. 160 is the same amplifier gain as 200 and moving the exposure slider in lightroom by -1/3.
    By exposing to the right to achieve ISO 160 you can run the risk of blowing the highlights sooner. By exposing to the left to achieve ISO 125 you run the risk of increasing noise in the shadows. How that translates to your real-world photography is up to you to test.
    I reserve the right to be wrong about all of the above, b.t.w.
    The base ISO of your camera is 100. There's no trick to figuring that out.
     
  12. Anyone who is resorting to this type of detailed analysis of a camera in a bid to improve the quality of their images needs to sit in a cold dark room for a week to re-format their brain.
     
  13. Jamie Robertson wrote:
    Anyone who is resorting to this type of detailed analysis of a camera in a bid to improve the quality of their images needs to sit in a cold dark room for a week to re-format their brain.
    ...or maybe has higher standards and/or shoots in more demanding lighting conditions?

    Are you impressed by the "non-smoothness" in the deep shadows in the detail below (?) because I certainly am not! The RAW capture was ETTR @ ISO 100 and processed rather carefully -- e.g., 16-bit per channel depth with no aggressive adjustments -- so if there's anything else I can do short of buying an FF body to reduce the noise in dark tones, I want to give it a go. There's only one way to find out, and it is to put it to test.

    Thanks for your comment anyway.
    00YjBs-358461584.jpg
     
  14. Here's the whole frame; it looks good in this tiny size, but should one wish to print it large, shadows would start looking rather ugly.
    00YjBy-358465584.jpg
     
  15. Are you impressed by the "non-smoothness" in the deep shadows in the detail below (?)​
    To be 100% honest, yes I would be perfectly happy with that, especially on a photo with such a high variation in lighting. I think it's perfectly fine, especially considering it came from an APS-C camera. I have the 5D2 and if this shot was taken with my camera I'd still be happy with it.
     
  16. Well, the question is: can it be improved (with existing equipment) or is it as good as it gets? -- and that's what I want to find out. Mind you, it's completely risk-free and $-free, so why not? -- the worse that can happen is that I'll come to the conclusion that pictures taken at ISO 160 aren't any better in the noise department, and if I notice any reduction in dynamic range, I'll revert back to ISO 100.

    Here's another example under far less challenging lighting conditions yet with equally pronounced noise in shadows.
    00YjCh-358473584.jpg
     
  17. And again the whole frame.
    00YjCk-358475584.jpg
     
  18. Tomek, I think your test is fatally flawed unfortunately.
    When you set your camera to an ISO, it is to inform the lightmeter how to expose the scene. You've supplied the light meter with so few photons, that it can not work at all. Your results might be just random. You could just be making observations on the length of exposure. Or even the temperature of the sensor at the time of exposure.
    Now there might be some change to a RAW image by changing the ISO setting, but you'd need to use manual exposure and use the same shutter speed for each exposure, preferably a reasonably short exposure, but one that the camera can repeat reliably. You'll also need to do this in a temperature controlled environment, and would be best done by remote to avoid influencing the temperature of the chip. Leave some "rest" time between exposures so as not to heat up the chip through continuous use. I think you get the idea here.
    There are yet greater flaws to your test. Firstly, what is base ISO anyway? One must define it to chase it. For me, that would be the the lowest ISO setting that records the full dynamic range that the chip can see. At a certain point, lowering the ISO just clips the bright tones that the chip can see. I suspect that is why on some cameras, one must use the menu for special access to ISO 50, rather than 100 for example.
    Then maximum dynamic range is itself murky. One person's dynamic range is another's noisy image. So there must be a subjective "noise threshold" applied to make this definition.
    Now, one would think that changing the ISO setting when recording RAW images should not change the recored image at all, except for the metadata recorded with the image to be used as a guide to the RAW conversion software. But this doesn't seem to be the case always and some ISO setting involve voltage boosts to the analog signal off the chip and, at a certain point, in camera noise reduction takes place as well. So these are the effects you're probably trying to get at.
    And I'll say here that at the low ISO that you test, you won't get into noise reduction, but I can't know that for sure though can I?
    Anyways, if you're testing for the application of analog gain to the image due to the setting of the ISO, you've got one big problem: You've supplied no signal to the chip. No light at all. No photons, No voltage generated by the chip. At voltage =0, when you multiply that voltage by increasing the ISO, the result is still 0! Any change you are observing is random. You'll need to expose the chip with some photons to observe this effect, if it exists!
    What you really want to test for is: At a given light level, and scene contrast, which ISO setting mixed with which exposure gives the best combination of dynamic range recording while showing the cleanest, noise free, shadows. This is quite a complicated test, and one that you can not possibly do with the lens cap on!
     
  19. Cheers for your post, Bruce! Suspect you're right on a number of accounts, but let me try to "defend" my position nonetheless (defend in quotation marks because I'm engaging in a debate merely for the enjoyment of a constructive discussion.)
    What you really want to test for is: At a given light level, and scene contrast, which ISO setting mixed with which exposure gives the best combination of dynamic range recording while showing the cleanest, noise free, shadows. This is quite a complicated test, and one that you can not possibly do with the lens cap on!​
    Ideally, yes, but I had to be realistic -- there's no way I could pull that one off.
    Firstly, what is base ISO anyway? One must define it to chase it.​
    For my purpose, I implicitly "defined" it as the ISO that gives the least amount of noise, conveniently ignoring other aspects of sensor's performance; had to keep it simple. My presumption was that this would be the least altered capture, "the purest RAW" if you will, but there is an evidence to suggest that this supposition might have been wrong.
    Now there might be some change to a RAW image by changing the ISO setting, but you'd need to use manual exposure and use the same shutter speed for each exposure, preferably a reasonably short exposure, but one that the camera can repeat reliably.​
    Exposure was set manually @ 1 second (the aperture was the same too, but guess that's irrelevant), which is reasonably short whilst guaranteeing that frame-to-frame relative variations are negligible, so believe this criterion was satisfied.
    You'll also need to do this in a temperature controlled environment, and would be best done by remote to avoid influencing the temperature of the chip. Leave some "rest" time between exposures so as not to heat up the chip through continuous use.​
    The camera was sitting in my room for days, and think we can safely assume that the indoor temperature was constant over a minute it took me to shoot 16 frames. There wasn't much rest time -- enough to change the ISO setting -- but figured the sensor wouldn't heat up *THAT* fast! (I'm aware of thermal noise, but understand this come into play during exposures that last minutes. Besides, I was increasing the ISO, so 160 was 'handicapped' vis-à-vis both 100 and 125.) All in all, I'd say conditions were sufficiently stable for the results to be meaningful.
    Now, one would think that changing the ISO setting when recording RAW images should not change the recored image at all, except for the metadata recorded with the image to be used as a guide to the RAW conversion software. But this doesn't seem to be the case always and some ISO setting involve voltage boosts to the analog signal off the chip and, at a certain point, in camera noise reduction takes place as well. So these are the effects you're probably trying to get at.​
    From what I learned since yesterday:
    • what the sensor itself registers indeed does not change as one changes ISO setting (it's intuitive as photodiodes on a CMOS chip are fixed hardware)
    • full f-stop changes in ISO are achieved by gain on amplifier (so it's electronic on a sensor's signal level)
    • +/- 1/3 changes are achieved in software ("artificial" pushing/pulling)
    • in-camera noise reduction kicks in at much higher ISOs and/or longer exposure times, so I didn't even bother disabling it
    What I wanted to get at was...
    You've supplied the light meter with so few photons, that it can not work at all. You've supplied no signal to the chip. No light at all. No photons, No voltage generated by the chip. At voltage =0, when you multiply that voltage by increasing the ISO, the result is still 0!​
    ...exactly that -- what the chip will do, or camera record, when it should record nothing at all. Mind you, I'm interested in noise that shows up in deep shadows where S/N ratio is very low, so reckon that was a good proxy.
    You'll need to expose the chip with some photons to observe this effect, if it exists!​
    On the contrary, believe that providing any S>0 would mask the N, which is why noise is invisible in well exposed areas (in fact it is uniformly distributed across the entire frame), so my methodology isolated the issue I wanted to investigate.
    Any change you are observing is random.​
    It's random noise for sure, but the point is that it's different for different ISOs, and that it changes in a counter-intuitive manner (at least if one's intuition is derived from film days, where, as a rule, more sensitive = grainier ;) I could re-run this test, but, as it turned out, a number of other people did more or less the same, and while our methodologies differ, the results did not, suggesting there is a consistent pattern that is stronger than variations between our experiments. For example, in addition to Tony Lorentzen whose video test inspired me to do my still picture 'equivalent', see Martin Beek's of Marvels Film Production. Also John Sheehy's findings on other Canon models paint the same story -- when empirical findings corroborate, conclusions are rather sound. Of course, I can only speculate as to why we repeatedly observe this behaviour (insiders who are truly in the know can't tell us as apparently it's classified information -- scroll down to Peter Troeger's last comment) and what the possible "side effects" such as highlight clipping are, but numerous demonstrations that 7D has the lowest noise at ISO 160 are convincing to me. Again, whether it matters in practice is a different story.
     
  20. Just to continue... :)
    I think what you've amplified in your "test" is the noise floor that is always present in every exposure. When you set the ISO to 160, the RAW converter software pushed down the noise as adjusted to an expected over exposure of 1/3 stop. So you've just crushed the noise floor. Which is something one always wants to do in a finished photograph, crush the noise floor to black for a clean image.
    The noise floor is present on every camera and should not be considered part of the usable image data. And should not be of concern in your photography.
    Instead of worrying about this kind of testing, which just amplifies the noise floor, just over expose your images a little. You'll loose some highlight detail, but you'll have cleaner shadows after bringing down the image to your preferred range. Pretty simple really.
    By the way I liked your photographic examples above :)
     
  21. Appreciate your follow-up, Bruce!
    I think what you've amplified in your "test" is the noise floor that is always present in every exposure.​
    You're spot on!
    However, is there any other kind of sensor noise than what you're referring to as the noise floor? I might be oversimplifying here, but I would have thought that, under given conditions, noise is pretty much constant. Some of these conditions we can change (ISO comes to mind ;) some are beyond our control and we just have to live with them.
    The noise floor is present on every camera and should not be considered part of the usable image data. And should not be of concern in your photography.​
    True, it's always there, but not always the same amount of it: some sensors/cameras are better than others (obviously) and, more importantly, as we've seen, a given camera generates different amount of it at different ISOs under otherwise comparable conditions. And yes again, it's an undesirable component of the image data for which we have no use, but the problem is that it renders unusable some of the desirable image data (read: weak signal produced by deep shadows)
    when N ≈ S, and therefore minimising N when I cannot increase S is very much a concern of mine.
    When you set the ISO to 160, the RAW converter software pushed down the noise as adjusted to an expected over exposure of 1/3 stop. So you've just crushed the noise floor. Which is something one always wants to do in a finished photograph, crush the noise floor to black for a clean image.​
    My understanding is that this "pulling" happens inside the camera, somewhere along the way between amplifying signal received from a sensor and transformation of that signal into a RAW file, which is then written onto a memory card. RAW converter only interprets that file according to instructions I give it (and its own internal ways of handling the RAW development.)
    Yeah, I know I can suppress the noise by adjusting black point, whether in ACR during RAW development or later via levels or curves in PS, and I often am forced to do just that, but that comes at the expense of killing shadow detail. Calling shadows treated that way "clean" is thus somewhat misleading; they are blocked shadows, devoid of detail.
    Instead of worrying about this kind of testing, which just amplifies the noise floor, just over expose your images a little. You'll loose some highlight detail, but you'll have cleaner shadows after bringing down the image to your preferred range.​
    In most cases, sacrificing [relevant] highlight detail is not a trade-off I'm willing to make; of the two evils, blown-out highlights are usually a worse crime than blocked shadows. I'm doing my best exposing to the right though. In order to increase the dynamic range, sometimes I resort to multi-RAW processing (which is when bringing out shadow detail invariably amplifies the inherent noise too). Also, when situation permits, I may take a couple of bracketed shots, one exposed for highlights and one for shadows, and merge them (here's an example of this approach), but that's not always practical.
     
  22. "Anyone who is resorting to this type of detailed analysis of a camera in a bid to improve the quality of their images
    needs to sit in a cold dark room for a week to re-format their brain."

    While I can be (almost) as interested as others in the technical details of how digital capture works, this comment is
    essentially correct. Unless one is an engineer working in the field of camera sensor design, this subject has virtually no
    significant application to actual photography.

    In real photography, even in very critical work that leads to beautiful large prints, the odds that choosing, say, ISO 125
    over ISO 100 would make any visible difference in the noise (or much of anything else) in a print are essentially zero.
    While you may be able to barely detect some difference in test images of uniform subjects that are pushed several
    stops and subjected to gross manipulations in post before being laboriously compared side-by-side at 100% or greater
    magnification, this has essentially no application beyond academic interest.

    There is, I suppose, nothing wrong with that per se. However, it must be said - perhaps repeatedly - to help new
    photographers who read these forums from getting sucked down the drain of obsessive pixel peeping rather than
    getting on with the business and/or pleasure of making photographs:

    This stuff is essentially irrelevant to you. It has essentially no effect your photographs. Let it go.

    Take care,

    Dan
     
  23. Yeah, I know I can suppress the noise by adjusting black point, whether in ACR during RAW development or later via levels or curves in PS, and I often am forced to do just that, but that comes at the expense of killing shadow detail. Calling shadows treated that way "clean" is thus somewhat misleading; they are blocked shadows, devoid of detail.
    Instead of worrying about this kind of testing, which just amplifies the noise floor, just over expose your images a little. You'll loose some highlight detail, but you'll have cleaner shadows after bringing down the image to your preferred range.​
    In most cases, sacrificing [relevant] highlight detail is not a trade-off I'm willing to make; of the two evils, blown-out highlights are usually a worse crime than blocked shadows. I'm doing my best exposing to the right though. In order to increase the dynamic range, sometimes I resort to multi-RAW processing (which is when bringing out shadow detail invariably amplifies the inherent noise too). Also, when situation permits, I may take a couple of bracketed shots, one exposed for highlights and one for shadows, and merge them (here's an example of this approach), but that's not always practical.​
    I"m afraid you're misinterpreting, and then amplifying this misinterpretation. The noise floor is always there. It always needs to be crushed in the camera, in the RAW software, or in the image editing software. You just don't want to crush usable signal also by going too far. So really the only thing you can really do is shoot with the most exposure (signal) that maintains the maximum dynamic range of the camera. Then you will have the largest signal to noise ratio and thus the cleanest image the camera will produce. I think this is almost always the lowest ISO that the you can set on the camera without using the "extended" range accessible via the camera menu. You can always extend the dynamic range of the camera by giving less exposure, but you'll be amplifying the noise (less signal, more noise) a little more so it's always a trade off.
    You are correct that multiple captures that bracket exposures will give you more signal to noise ratio with more dynamic range after blending in software. This will help. A test that is all noise and zero signal tells you absolutely nothing unfortunately.
     
  24. I've heard of this ISO test too. They also say that ISO 640 is identical to ISO 100. I recently picked up a 7D and can't seem to get my ISO below 200. It goes from 200 to Auto. How does one get to 100, 160, 320, etc..?
     
  25. [[It goes from 200 to Auto. How does one get to 100, 160, 320, etc..?]]
    To get ISO 100 you need to turn off Highlight Tone Priority.
     

Share This Page