We often see in camera reviews observations that "camera X seems to a higher actual sensitivity for the same ISO sensitivity setting than camera Y". It is not always clear how this has been determined.
Rarely is there any discussion of how the assigned ISO sensitivity rating for a camera compares to the value that would be assigned under the international standard for the ISO "speed" (sensitivity) of a digital camera, ISO 12232. It is as if there were no such thing as a "correct" ISO sensitivity rating.
I recently looked into the intimately related matters of the calibration of a camera's automatic exposure control system and the "actual" ISO sensitivity ratings for the various selectable settings for sensitivity. The whole story here, and some of my findings, are covered in a technical article I expect to shortly publish on my technical information site, The Pumpkin.
But I thought I would present here and now some of my basic findings, in order to solicit comments on what some of them mean, and to solicit any other information that would help me complete my analysis and refine my explanation of it.
************
** A "standard" automatic exposure camera **
Suppose we have a camera with an automatic exposure control system (a) whose "calibration" was precisely that specified by the ISO standard for such things, ISO 2721, and (b)in which the ISO sensitivity rating for each of the "ISO sensitivity" settings was precisely in accord with the ISO standard for such things, ISO 12232.
Then if we take a picture of a uniform-luminance test target (using the automatic exposure control system), the exposure on the sensor (exposure in the sense of the product of illuminance and exposure time, the phenomenon to which the sensor responds) will ideally be 12.8% of the "saturation" exposure (in that same sense) - the largest exposure for which the sensor system reports a unique digital result.
** Free-standing exposure meter calibration **
ISO 2720 prescribes the calibration for free-standing exposure meters. The defining equation involves the factor "K", called the "reflected light exposure metering constant". By choosing the value of that constant, the meter manufacturer can control the exposure (in the sense of the combination of an aperture and a shutter speed) that will be "recommended" for any given measured scene luminance and value of ISO sensitivity we have set into the meter. The standard allows a range of values for this constant. Often a value in the range 12.5-12.7 is used.
The standard for automatic exposure control systems (ISO 2721) does not use that way to express the prescribed "calibration", but one can determine that the calibration it prescribes is the same as that prescribed by ISO 2720 for a freestanding exposure meter with a K of 12.7.
** Canon EOS dSLR cameras **
It has been widely stated that Canon SLR cameras (film and digital, in fact) have automatic exposure control systems which, treated as if they were free-standing exposure meters, would have a K value of 12.5.
I cannot determine if this is so for any of my Canon SLRs since I do not have a calibrated luminance source or a calibrated photometer. But some rather "agricultural" tests seem to suggest that something close to that applies to my EOS 300D and 20D.
Canon has made available a simple field test that will allow us to determine if the overall operation of the automatic exposure system is consistent with Canon's design objective.[Thanks to member Bill Neukranz for reconstructing the description of that test.] This test does not tell us either the calibration of the automatic exposure system nor the "calibration" of the assignment of ISO sensitivity - to test either would in fact require a calibrated luminance source (or a calibrated photometer). Rather, it tests the joint effect of the two. (It is that joint effect that determines whether the overall exposure "strategy" is that upon which Canon has adopted is being fulfilled.)
** The Canon test **
I'll give a simplified description of the test, leaving out some foo-foo stuff that is probably irrelevant anyway.
We use the camera to take a picture of a uniform test target and examine the image in Adobe Photoshop, using the mode in which the image is virtually converted to a grayscale color space. (Which grayscale color space is to be used isn't specified - I am assuming that it is the grayscale color space that was the default for Photoshop up through version 4: the "1.8 gamma" grayscale color space.)
This color space has a single coordinate, which we may call "blackness", with the symbol "K". (Note that this has nothing to do with K, the reflected light exposure metering calibration constant. The symbol is probably borrowed from the "K" coordinate in the CMYK color space.)
If the camera's exposure control system is behaving as designed,the K value for the image (hopefully fairly uniform across it) should be 55%.
Under the 1.8 gamma grayscale color space, a K value of 55% represents a luminance of 23.7% of maximum luminance. If there were no disruption caused to the tonal scale by contrast processing, that would mean that the exposure on the sensor would be 23.7% of the saturation exposure.
Note that this is 1.86 times the value that would occur for a theoretical "standard" camera, as defined above.
This could result from either of the following or a combination:
1. The exposure meter aspect of the automatic exposure system has a calibration factor higher than prescribed (indirectly) by ISO 2721
and/or
2. The assignment of an ISO sensitivity to the behavior of the camera's sensor system, for any given setting, is lower than would be dictated by ISO 12232.
AS mentioned earlier, it seems unlikely that factor (1) above is a big part of this. Thus we are left with no choice but to think that Canon has elected to use a substantially conservative ISO sensitivity rating for its digital cameras.
Why might they have done this (if that is indeed what we are dealing with)?
Perhaps Canon had determined that the "exposure philosophy" called for by the interaction of the two ISO standards was too conservative - that it resulted in an unnecessarily-large "headroom" against clipping.
They could have reduced the headroom by increasing the calibration constant of the exposure metering portion of the system above that intimated by ISO 2721. But then, photographers choosing to use an external exposure meter (calibrated to ISO 2720) would get results not at all consistent with that given by the camera's automatic exposure control system. However, by using a conservative ISO sensitivity rating, the headroom would be reduced without causing that problem.
Again, who knows.
************
I would appreciate any information you folks can pile onto this, including any results of comparing the exposure metering (as such) of Canon dSLRs with that given by trustworthy freestanding exposure meters.(Note that rarely, though, do we know the precise value of K used by the manufacturers of such meters, but we can guess it is close to 12.5.)
Visit COLORRIGHT to get the colors right in your digital slr.
and note where I pointed out to him that one's chosen setting for grayscale in Photoshop's Color settings influences the "K value."
His usual recommendation for the proper K value is based on a 2.2 gamma grayscale setting.
(I don't know whether this has any bearing on your findings, but thought you might want to see it.)
In this thread it also came up that the color working space and the grayscale should be set to the same gamma for the expected results, although that may be mitigated by "untagging" the file.
and note where I pointed out to him that one's chosen setting for grayscale in Photoshop's Color settings influences the "K value."
[/ QUOTE ]
I hadn't seen that particular thread. I'll take a look pronto.
Of course you were quite correct in your observation of the signficance of the choice of grayscale color space for the evaluation. (I dwell on that matter in my article, but just jumped to an apparently-unwarranted conclusion in the "short" discussion in this thread.)
[ QUOTE ]
His usual recommendation for the proper K value is based on a 2.2 gamma grayscale setting.
[/ QUOTE ]
I hadn't known that. It certainly makes sense.
[ QUOTE ]
(I don't know whether this has any bearing on your findings, but thought you might want to see it.)
[/ QUOTE ]
Yes, it has an important impact.
On that basis, the "bogey" for the relative luminance of the recorded image (the K=55%) for a uniform luminance scene becomes 0.173, much nearer to consistency with that for the hypothetical "ISO standard" camera. The "discrepancy" is now only 0.43 stop, and I can see a good rationale for that. It would very nearly amount to a discard of the 1/2-stop "headroom" built into the ISO model, something that would be well warranted when we have available an "intelligent" metering system (such as the "evaluative" metering scheme that is the centerpiece of the Canon exposure control doctrine).
I'll discuss this further in my article, which will now take on a different slant!
[ QUOTE ]
In this thread it also came up that the color working space and the grayscale should be set to the same gamma for the expected results, although that may be mitigated by "untagging" the file.
[/ QUOTE ]
I wouldn't have expected that. I would have thought that what was needed would be for the working space to be consistent with the space in which the image was recorded. (The "virtual conversion" of the image to gray scale is just an arbitrary tool for "reading" the luminance implication of the recorded image. We could get the same thing by analysis of the sRGB R, G, and B values - it's just not so handy.)
I'll have to give that some more thought (after I have read the thread you cited!).
I looked at the thread you cited. Very interesting. You did well (and patiently) in clarifying the matter of which grayscale space was to be used for the assessment of the test image.
Thanks again for your input to my learning process in this matter.
This is of course the sort of thing that makes this forum so valuable. I'm very grateful that it will be continuing.
>>>>I would have thought that what was needed would be for the working space to be consistent with the space in which the image was recorded.<<<<<
My thinking on the subject had to do with either an arbitrarily Photoshop created file (therfore in the chosen working space) or raw images (in reality ahot in "camera xxx rgb" then converted to a chosen working space through Raw Converter xxx's tonal curve, and subject to whatever camera input profile is used by that converter.) AS you know a raw image isn't really in any working space until it has been converted. Besides different input camera profiles and tonal curve different converters use different gamma in the intermediate stages of conversion. For an example, Capture One uses a 1.8 gamma engine while most other raw converters use something much closer to 2.2 gamma; hence CO users have had a hard time getting expected results from Coloreyes 20/20 camera profiling package, which assumes that the supplied target image has been rendered through a 2.2 gamma converter.
If talking about in-camera jpegs, then it would be appropriate to say`"the space in which the image was recorded," although subject to the chosen contrast parameter!
So I would certainly keep these things in mind before chisling any hard and fast rules in stone!
Thanks for taking on this task. I am lookiing forward to the result!
[ QUOTE ]
If talking about in-camera jpegs, then it would be appropriate to say`"the space in which the image was recorded," although subject to the chosen contrast parameter!
[/ QUOTE ]
Indeed, and the whole discussion here pertains to observing an image generated by the camera. Perhaps I was not sufficiently scrupulous in my language - I need to go back and look at what I said.
However, what I said would be equally applicable to a situation in which I arbitrarly generated a grayscale test wedge image in, say, Picture Publisher 8, with 21 patches intended to represent relative luminance values from 0.0 to 1.0 in steps of 0.05. I would then have that written as a JPEG file in the sRGB color space. (And the image is not tagged with the color space - that editor will not do that.)
Now, I open that in Photoshop, wanting to see what K value, under say, a simulated gamma 1.8 grayscale conversion (which is what actually happens when we probe the image with the Info panel's probe, with one side set to "gray scale" and "Gray gamma 1.8" in effect as the Phtoshop grayscale working space) would be given for each of those patches.
The result will come out right if the working space (under which my test image is actually interpreted by Photoshop) is set to sRGB, which of course has quite a different gamma than the "virtual" working space (Gray gamma 1.8) under which I want to see what the K value would be for each patch. In effect, the Info process simulates conversion of the image from the current working space to the space nominated for "info" display. If the current working space is set to the space in which the image file was generated, this will work out fine.
If I change the grayscale working space to Gray gamma 2.2, then of course the K value given by the Info box in its simulated grayscale mode will be different than when the grayscale working space was gamma 1.8. But the K value given with Gray gamma 1.8 in effect (for the Info process), and the K value given with gamma 2.2 in effect, will both imply precisely the same luminance (interpreted in each case under the color space involved).
For example, if I examine the patch in the test image that is intended to represent a luminance (Y) of 0.35 (it is RGB 160,160,160 in the sRGB color space, the space in which the image is written) using the Gray gamma 1.8 color space as the simulated color space for Info examination (and with sRGB in place as the Photoshop working space), the indicated value of K will be 44%. Knowing the rule for the gray gamma 1.8 color space, we find that this value of K represents a relative luminance of 0.35.
If I examine this patch using Info with the Gray gamma 2.2 color space as the simulated color space for Info examination, I will get a K value of 38%. Knowing the rule for the gray gamma 2.2 color space, we find that this value of K represents a relative luminance of - - 0.35.
So we can use any grayscale color space we want to examine our image, but before we conlude what luminmance is represented, we have to know what grayscale color space that is. And of course if we are given a "bogey" for K, as you so well pointed out in the thread you cited, we need to know which color space that is intended to be interpreted under.
[ QUOTE ]
So I would certainly keep these things in mind before chisling any hard and fast rules in stone!
Linear Mode
