Pages: [1] 2   Go Down

Author Topic: what ISO and CPIA say about the "ISO" setting on a camera, and other measures  (Read 7092 times)

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

There is much confusion about what the ISO setting on a camera does and how its numerical values should be calibrated. The full answer is in ISO standard 12232 (2006 revision) and the Japanese Camera and Imaging Products Association [CIPA] standard DC-004. The latter can be found at that link, but unfortunately the former costs US$123, here so I have not seen all of it.

The CIPA requires that its members (all Japanese cameras makers) calibrate the exposure index settings (the "ISO" dial or menu setting) on their cameras using one of two measures, "Standard Output Sensitivity" or "Recommended Exposure Index". Both are based on brightness levels seen in end-products: getting "appropriate" levels in a camera's default JPEG output. Neither the ISO not CPIA standards care in the slightest how the intermediate raw levels are used in order to achieved this result.

ISO 12232 also defines several other measures of sensor performance, related to highlight and noise handling. Note that these measure the sensor, not the camera as a whole, and in particular with a CCD (and some CMOS sensors), these measurements are on analog output signals in voltages or charges and are totally unaffected by any subsequent amplifier gain or ADC conversion. Thus they are independent of the exposure index ("ISO") setting on a CCD camera.  In particular, for a CCD, the saturation-based measure is the so-called base-ISO speed, which is about 100 for current CCDs with micro lenses and about 50 for those without. Things get more complicated with CMOS sensor, some of which apply different levels of gain to the signal while this signal is still on the sensor chip, depending on the camera's settings.

The ASA/ISO measure of film speed is based on adequate shadow handling, looking at handling of illumination levels four stops below the mid-tone level. The closest that ISO 12232 comes to matching this is in its two SNR-based measures, Snoise10 and Snoise40. (These are rarely discussed in "public" by camera makers of forum participants, but are important to technical users like astronomers, so are discussed into technical literature on sensors.) Roughly, these noise-based measures are given by the exposure index (level of sensor exposure) at which the signal-to noise ratio on the image of an 18% gray card is 10:1 and 40:1 respectively. For modern CCDs in MF cameras and backs, Snoise10 seems to be in the range 800-3200, limited largely by sensor dark noise, while Snoise40 is far lower and is limited by photon shot noise, since a SNR of 40:1 requires a signal of at least 1600 photons counted.

However, the measure that most people talk about around here (and often misunderstand because of its misrepresentation by DXO) is the sensor saturation based one "Ssat", which is what used to be referred to as base-ISO speed: this is a measure of the maximum sensor exposure level, meaning minimum exposure index that gives adequate highlight headroom: it is the sensor exposure level that sends an 18% gray card image to 12.7% of the full well capacity of a CCD (things got more complicated with some CMOS sensors.)

EDIT: I found parts of the ISO12232 standard, from which I quote (emphasis and bracketed comments are mine) it clearly state that the saturation based measure relates to overexposure latitude not "ISO speed".
Quote
6 Determination of ISO speed
6.1 General
… The maximum exposure level is the exposure level where typical picture highlights will be clipped as a result of saturating the image sensor signal capacity or reaching the camera signal processing maximum signal level. The minimum exposure level depends on the amount of noise that can be tolerated in the image. These situations lead to two different types of speed values, saturation signal-based values and noise-based values. The ISO speed is preferably determined using a noise-based method. [Snoise40] The saturation-based value is preferably used to indicate the camera's overexposure speed latitude.  [Ssat] A second noise-based value is preferably used to indicate the camera's underexposure speed latitude. [Snoise10] ...
The next bit is rather technical, but the gist is that:
"ISO speed" should be reported as the value of Snoise40 (suitably rounded)
“ISO Speed Latitude" should be reported as the range from Ssat to Snoise10 (suitably rounded)
as indicated by my bracketed notes above.
Quote
6.4 Method of reporting

The ISO speed of a DSC shall be denoted ... “ISO xxx” … the reported number “xxx” shall be the value from the third column of Table 1 from the same row as the Snoise40 value (in the second column of Table 1) determined in 6.2. The ISO speed latitude shall be denoted “ISO Speed Latitude yyy - zzz D” ... The reported number “yyy” shall be the value from the third column of Table 1 from the same row as the Ssat value (in the first column of Table 1) determined in 6.1. The reported number “zzz” shall be the value from the third column of Table 1 from the same row as the Snoise10 value determined in 6.2.

TL;DR
- The saturation-based measure Ssat of sensor performance is in no way intended by the ISO as a measure of what the exposure index ("ISO") setting on a camera does, and in fact contradicts the CPIA requirements for calibration of that setting.
- For CCDs (and some CMOS sensors), Ssat gives exactly the same "base ISO speed" numerical value regardless of the exposure index ("ISO") setting.
- The saturation-based Ssat measure is totally unrelated to the ISO film speed; film speed is instead more closely related to the ISO's SNR10 or SNR40 measures, which typically given far higher values than Ssat.
- The ISO prefers that "ISO speed" of a digital still camera be determined and reported using the noise-based method Snoise40 … but I have never seen this done outside of technical publications!
« Last Edit: March 02, 2014, 11:25:47 am by BJL »
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387

The CIPA requires that its members (all Japanese cameras makers) calibrate the exposure index settings (the "ISO" dial or menu setting) on their cameras using one of two measures, "Standard Output Sensitivity" or "Recommended Exposure Index". Both are based on brightness levels seen in end-products: getting "appropriate" levels in a camera's default JPEG output. Neither the ISO not CPIA standards care in the slightest how the intermediate raw levels are used in order to achieved this result.

However, the measure that most people talk about around here (and often misunderstand because of its misrepresentation by DXO) is the sensor saturation based one "Ssat", which is what used to be referred to as base-ISO speed: this is a measure of the maximum sensor exposure level, meaning minimum exposure index that gives adequate highlight headroom: it is the sensor exposure level that sends an 18% gray card image to 12.7% of the full well capacity of a CCD (things got more complicated with some COS sensors.)

Wikipedia has a good section on ISO speed as applied to digital cameras. Norman Koren also gives insights in his Imatest documentation.

Since most participants in this forum use raw, neither the SOS or REI standards are relevant, since they refer to JPEG output. However, the REI standard is actually a non-standard, since the rating can be assigned arbitrarily. For raw files, the original ISO-12232 standard (1997) is more applicable, and this is what DXO uses. I see nothing wrong with the DXO determinations. Japanese manufacturers can use the 1997 standard to set their REI rating, and this is what Nikon and other manufacturers do.

If one exposes according to the built in light meter, both the calibration of the meter and the ISO rating of the sensor are involved. If the manufacturers want to allow a certain degree of highlight protection, they could adjust either the calibration of the meter or that of the sensor. In practice, they do the latter, since the former would confound the use of an external meter.

One can test the calibration of the meter and sensor combination by exposing a uniformly illuminated surface according to the meter reading and looking at the raw value in Rawdigger or another program that works with raw files. The saturation according to the 1997 saturation standard should be 12.7%. This is close to what I get with my Nikon dSLRs and I think Canon gives similar results.

I think that the 1997 standard is what most of us raw shooters should use. For ETTR exposure, one would give an additional 0.5 EV of exposure.

Bill
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

Since most participants in this forum use raw, neither the SOS or REI standards are relevant, since they refer to JPEG output.
Neverthless, SOS (and its mysterious cousin REI) are the only standards that a (Japanese) camera maker is allowed to use for in-camera light meter calibration (and this calibration is mostly what the EI setting of a digital camera is) and so this pair is the de facto global standard, so it its absurd to criticize a camera maker for following that standard.  Frankly, it is also absurd to expect camera makers to not calibrate their light-metering and auto exposure based on JPEG usage, since that is the most common way that cameras are used.)

For raw files, the original ISO-12232 standard (1997) is more applicable, and this is what DXO uses. I see nothing wrong with the DXO determinations.
Try reading again what ISO 12332 says (and does not say) again: the Ssat that DXO approximates is not defined in ISO 12232 (1997 or 2006 version) as the correct or even recommended way to calibrate the EI setting; ISO 12232 describes Ssat as measuring the lower limit of the ISO speed latitude of the sensor --- NOT the "ISO speed", and absolutely not a recommendation for in-camera light meter calibration.

It amazes me how a measure of what EI gives a minimum level of highlight protection, based on a somewhat arbitrary placement of mid-tones at 12.7% of maximum, get misinterpreted as an ideal level of highlight protection or of level placement.  In particular, in low-light situations (exposure index well above base-ISO speed), where there is little or no risk of sensor blooming (overfilling some photosites), it seems an obviously poor approach to amplify so far that amplifier clipping of highlights is risked, and so pushing midtones to within this minimum recommended distance below maximum raw level is a poor strategy. Why would I want my camera's light meter recommending that?
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311

so it its absurd to criticize a camera maker for following that standard. 
the point is that for comparisions if you use raw you can't take the same ISOs from different cameras because the legit standard exactly allows manufacturers to be flexible...
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto

Hmm

How do I set my P45 for JPEG?

Best regards
Erik

Neverthless, SOS (and its mysterious cousin REI) are the only standards that a (Japanese) camera maker is allowed to use for in-camera light meter calibration (and this calibration is mostly what the EI setting of a digital camera is) and so this pair is the de facto global standard, so it its absurd to criticize a camera maker for following that standard.  Frankly, it is also absurd to expect camera makers to not calibrate their light-metering and auto exposure based on JPEG usage, since that is the most common way that cameras are used.)
Try reading again what ISO 12332 says (and does not say) again: the Ssat that DXO approximates is not defined in ISO 12232 (1997 or 2006 version) as the correct or even recommended way to calibrate the EI setting; ISO 12232 describes Ssat as measuring the lower limit of the ISO speed latitude of the sensor --- NOT the "ISO speed", and absolutely not a recommendation for in-camera light meter calibration.

It amazes me how a measure of what EI gives a minimum level of highlight protection, based on a somewhat arbitrary placement of mid-tones at 12.7% of maximum, get misinterpreted as an ideal level of highlight protection or of level placement.  In particular, in low-light situations (exposure index well above base-ISO speed), where there is little or no risk of sensor blooming (overfilling some photosites), it seems an obviously poor approach to amplify so far that amplifier clipping of highlights is risked, and so pushing midtones to within this minimum recommended distance below maximum raw level is a poor strategy. Why would I want my camera's light meter recommending that?
Logged
Erik Kaffehr
 

jerome_m

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 670

Quote from: BJL link=topic=87439.msg711840#msg711840 date=1392928376It amazes me how a measure of what EI gives a [u
minimum[/u] level of highlight protection, based on a somewhat arbitrary placement of mid-tones at 12.7% of maximum, get misinterpreted as an ideal level of highlight protection or of level placement.  In particular, in low-light situations (exposure index well above base-ISO speed), where there is little or no risk of sensor blooming (overfilling some photosites), it seems an obviously poor approach to amplify so far that amplifier clipping of highlights is risked, and so pushing midtones to within this minimum recommended distance below maximum raw level is a poor strategy. Why would I want my camera's light meter recommending that?

One could indeed insist that in low light the meter recommends the camera to let more photons in. In practice, that would mean that I use a slower shutter speed (because the lens aperture is already max and there are only that many photons around since it is dark). So, for example, in a dark Jazz club where I take pictures, my camera would recommend f/2.8 and 1/15s when I use iso 1600 and still f/2.8 and 1/15s when I use iso 3200. That would be confusing to the photographer and I need 1/30s in all cases to freeze movements.

Alternatively, what you are suggesting could mean the following: in the same Jazz club, I should use manual mode on the slowest speed I require, so I dial f/2.8 and 1/30s. I don't run the risk to saturate anything with photons, because there aren't that many around. I don't need a meter and the iso dial is not relevant. I just need the adjust the exposure digitally when converting the raws. That would work indeed.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

the point is that for comparisions if you use raw you can't take the same ISOs from different cameras because the legit standard exactly allows manufacturers to be flexible...
Agree "REI" allows much flexibility (to allow fancy pattern metering methods maybe), so I would prefer SOS.

But if the goal it comparable raw files, the answer is simply: choose the same exposure level with the different cameras by setting aperture and shutter speed manually.  All the stuff about calibrating the EI settings with SOS or REI or whatever is just about how auto-exposure works, and how the photosite outputs are then descried in numerical raw levels.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

How do I set my P45 for JPEG?
By default raw conversions in your favorite raw software: the SOS and REI calibration works just as well then, since their goal is simply that if you are in auto-exposure mode, default JPEG conversions come out at an appropriate brightness level. One camera with a 16-bit raw might put a mid-tone at level 4000/65536 (four stops of headroom) with default raw conversion knowing to map that to JPEG level 118, while another with a 14-bit ADC might put that mid-tone at level 2000/16384 (three stops of headroom) with default raw conversion again knowing to map that to JPEG level 118.  So long as you goal is to look at JPEGs rather than at multi-color raw histograms, both methods work fine.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

Alternatively, what you are suggesting could mean the following: in the same Jazz club, I should use manual mode on the slowest speed I require, so I dial f/2.8 and 1/30s. I don't run the risk to saturate anything with photons, because there aren't that many around. I don't need a meter and the iso dial is not relevant. I just need the adjust the exposure digitally when converting the raws. That would work indeed.
Indeed, that is what I mean: "ETTR for minimizing noise" is nonsensical in that case; you give the sensor as much exposure as your shutter speed and aperture limits allow, and still do not get close to the right end of the "sensor photon count" histogram. The main functions of the ISO setting then are to (a) indicate the amount of gain needed afar the sign leaves the photosites (analog or digital bit shifting or both)  to give appropriate brightness levels in the final JPEG output, and (b) to ensure than enough analog gain is used to avoid losing signal into the noise floor of subsequent processing. With current CCDs and 16-bit ADCs, even base-ISO gain is probably enough to do this, or else at most one or two stops of gain above that, so more analog amplification is useless, and bit shifting is a better approach.
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311

Agree "REI" allows much flexibility (to allow fancy pattern metering methods maybe), so I would prefer SOS.

But if the goal it comparable raw files, the answer is simply: choose the same exposure level with the different cameras by setting aperture and shutter speed manually

indeed, but then you still need to make an agreement about the gain... are we just using the best gain (and different) for each camera or we try to come up with some equivalency in gain settings (to be applied post exposure)... that nominal ISO can't be used as such equivalency and so some people are crying about "fake" ISOs (even they are all proper and according to the legit standard).
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387

Neverthless, SOS (and its mysterious cousin REI) are the only standards that a (Japanese) camera maker is allowed to use for in-camera light meter calibration (and this calibration is mostly what the EI setting of a digital camera is) and so this pair is the de facto global standard, so it its absurd to criticize a camera maker for following that standard.  Frankly, it is also absurd to expect camera makers to not calibrate their light-metering and auto exposure based on JPEG usage, since that is the most common way that cameras are used.)

Try reading again what ISO 12332 says (and does not say) again: the Ssat that DXO approximates is not defined in ISO 12232 (1997 or 2006 version) as the correct or even recommended way to calibrate the EI setting; ISO 12232 describes Ssat as measuring the lower limit of the ISO speed latitude of the sensor --- NOT the "ISO speed", and absolutely not a recommendation for in-camera light meter calibration.

Since neither of us have access to the official ISO 12332 standard (1997 or 2006 version) I can't re-read ISO 12332, but did you read the two links I cited regarding Ssat? Both references are consistent with what I stated regarding the saturation standard. I did not state that ISO 12232 applied to light meter calibration--ISO 2721 does that. Doug Kerr discusses exposure meter calibration and how it ties into sensor calibration via ISO 12232 and he interprets ISO 12232 in the same manner as I have.

I did not criticise camera makers for using the REI standard, but merely noted that the standard is arbitrary and allows the makers to use ISO 12232 1996 saturation standard if they so desire.

It amazes me how a measure of what EI gives a minimum level of highlight protection, based on a somewhat arbitrary placement of mid-tones at 12.7% of maximum, get misinterpreted as an ideal level of highlight protection or of level placement.  In particular, in low-light situations (exposure index well above base-ISO speed), where there is little or no risk of sensor blooming (overfilling some photosites), it seems an obviously poor approach to amplify so far that amplifier clipping of highlights is risked, and so pushing midtones to within this minimum recommended distance below maximum raw level is a poor strategy. Why would I want my camera's light meter recommending that?

Current dSLRs do not record ISO as metadata, but rather amplify the signal (presumably to use the full range of the ADC), and there is a risk of clipping in the ADC with increasing exposure. Light meters have no idea of the reflectance of the scene, mid-tone or otherwise--they expose a white and gray card at the same level but work in most instances since the average scene reflectance is about 12.8% and they allow 0.5 EV of highlight headroom. If one desires to place the highlights just short of clipping, one can take a reading from the highlights and increase exposure by about 2.5 EV.

Regards,

Bill
« Last Edit: February 20, 2014, 07:59:59 pm by bjanes »
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311

Current dSLRs do not record ISO as metadata, but rather amplify the signal (presumably to use the full range of the ADC)

dSLRs may be not, but dSLMs do (sometimes) - Panasonic GH3 for example use both analog/digital gain _AND_ "ISO by tag" in their raw files
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387

Indeed, that is what I mean: "ETTR for minimizing noise" is nonsensical in that case; you give the sensor as much exposure as your shutter speed and aperture limits allow, and still do not get close to the right end of the "sensor photon count" histogram. The main functions of the ISO setting then are to (a) indicate the amount of gain needed afar the sign leaves the photosites (analog or digital bit shifting or both)  to give appropriate brightness levels in the final JPEG output, and (b) to ensure than enough analog gain is used to avoid losing signal into the noise floor of subsequent processing. With current CCDs and 16-bit ADCs, even base-ISO gain is probably enough to do this, or else at most one or two stops of gain above that, so more analog amplification is useless, and bit shifting is a better approach.

Agreed. ETTR as judged by the camera histogram is useless with high ISO, since the histogram does not show how many photons are captured. How high the point of diminishing returns is when increasing the ISO to diminish read noise depends on the sensor. With the Canon EOS 1D M4, reduced read noise is achieved up to about IS0 800 as shown by Bill Claff's chart shown below. With the Nikon D800e, one can see a minimal improvement of about 1/3 EV up to ISO 800. The D70 (CCD sensor) shows a similar situation. This is for photographic DR, not engineering DR.

Bill

PS
Edited to include the chart.
« Last Edit: February 21, 2014, 02:58:47 pm by bjanes »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
SOS or REI calibration are independent of Ssat and raw level placement choices
« Reply #13 on: February 21, 2014, 12:06:45 pm »

Bill,

   Firstly, I realize that I explained things badly in my previous reply to you, so let me try again.  Also, I have now read the relevant parts of ISO 12232-2006: that is where I got those direct quotes from.

Maybe the simplest point is this: the JPEG-based exposure index standards "Standard Output Sensitivity" [Isos] and "Recommended Exposure Index" [Irei] are independent from the sensor performance measures Ssat, S10 and S40, which relate only to sensor output levels (and maybe raw levels), not JPEG conversions.  Two cameras could use the same sensor and the same Isos calibration of their exposure index setting, choose the same combinations of shutter speed and aperture ratio in auto-exposure modes, and deliver the same levels in JPEG output (midtones at about JPEG level 118) and yet use very different raw levels and so have different Ssat values and measures by DXO.

Several different approaches seem to have been used in different cameras:

A. Every doubling of the EI setting doubles the analog gain used and so doubles the raw level given for a photosite that received the same amount of light; raw->JPEG conversion then matches a given raw level to the same JPEG level regardless of the EI setting. [Many CMOS DSLRs do this.]

B. Doubling the EI setting only changes the light meter operation (for example doubling the shutter speed chosen in A mode) but does not change the analog gain, so the same photosite signal maps to the same raw level. The raw file is then flagged to notify raw convertors, so that they can apply "digital gain" in order to produce default JPEG conversions that are in accordance with Isos.  [Some Hasselblad backs seem to do this.]

C. A bit of both, like increasing analog gain and raw level placement up to about EI 800 or 1600 (as in A.) but then for higher EI settings, not increasing gain any more, and instead proceeding as in B. [Many CCD DSLRs did this, with no further analog gain at the special "high ISO" or "boost" settings. For example the analog gains increases stop at EI=800 on my old Olympus E-1.]

D. A variation on B. and C. where in case that the analog gain is not increased, the same effect on raw levels is achieved by  bit-shifting up the ADC output levels before writing to the raw file, avoiding the need for a "digital gain flag" in the raw file. [DXO data hints that some Phase One "IQ" series backs might do this.]


What happens the DXO versions of Ssat measurement each case?
A. The DXO measure increases in proportion to the EI setting, but differing form it by roughly (stops of highlight headroom in the raw file) - 3.
B. The DXO measure is the same at all EI settings; it is always the base ISO speed of the sensor.
C. The DXO measure increases in proportion to EI up to the "analog gain" threshhold, and then is constant above that.
D. As in A.: the DXO measure increases in proportion to the EI setting.


For a truly "ISO-less" camera, analog gain is superfluous and option B. is probably the best, and the MF backs with 16-bit ADCs but CCD sensors of well under 16 stops of total dynamic range are rather close to the "ISO-less" situation.
« Last Edit: February 24, 2014, 10:12:41 am by BJL »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

How high the point of diminishing returns is when increasing the ISO to diminish read noise depends on the sensor. With the Canon EOS 1D M4, reduced read noise is achieved up to about IS0 800 as shown by Bill Claff's cart shown below. With the Nikon D800e, one can see a minimal improvement of about 1/3 EV up to ISO 800. The D70 (CCD sensor) shows a similar situation. This is for photographic DR, not engineering DR.
Agreed!  And your three example fit nicely with my understanding of how the three different sensor types in those cameras behave:

1. With a CCD as in the D70, gain is applied off-sensor, immediately before the signal goes into the ADC, so the only noise source "downstream" of the analog gain is from the ADC itself. So once the analog gain is sufficient to raise the incoming noise level above the noise floor of the ADC, further analog gain is of little or no benefit. If the ADC is good enough that its DR is at least as great as the DR of the sensor, this should happen not very far above base-ISO speed.

2. With some Canon CMOS sensors, analog gain (over a three stop range in the one Canon technical report I read) is applied very early, during signal transfer from the photosite to a sense capacitor at the edge of the sensor, and then this amplified analog signal is transported off the sensor to an off-board ADC. Thus there is more room for "downstream noise" or signal DR limitations in the subsequent transport path, and so potentially more benefit to be had from amplifying the signal in high EI situations as much as possible without clipping it.

3. With CMOS sensors using on-chip column-parallel ADC (as in the D800, and almost all recent non-Canon CMOS sensors) analog gain is applied immediately before ADC, with both done at the same place at the end of each column of photosites. So the situation is similar to that of CCDs, and once again there should be little benefit to more than a couple of stops of analog gain increase as EI is increased.
Logged

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/

BJL,

I have browsed quickly though all this and find it interesting (wish I had more time), thanks for posting this.

But at the end I am back with the same basic question.

How can you compare the actual sensitivity of an IQ280 vs a D800 when shooting raw with ISO100 dialed in camera?

Cheers,
Bernard

peterottaway

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 91

BJL,

I have browsed quickly though all this and find it interesting (wish I had more time), thanks for posting this.

But at the end I am back with the same basic question.

How can you compare the actual sensitivity of an IQ280 vs a D800 when shooting raw with ISO100 dialed in camera?

Cheers,
Bernard
 At practical level the same way we used to do with non metering MF especially, use a hand held meter and run a series of tests.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
actual sensitivity: meaning base ISO speed, or "ISO speed" as ISO12232 defines?
« Reply #17 on: February 24, 2014, 01:50:06 pm »

Bernard, thanks for your reply.
How can you compare the actual sensitivity of an IQ280 vs a D800 when shooting raw with ISO100 dialed in camera?
Do you mean sensitivity
- in the sense of ability to detect light, as the ISO recommends with its definition of "ISO speed" as measured by Ssnr40?
or
- in the sense I would could "delicateness"; sensitivity to strong highlights as measured by blown highlights via the "base ISO speed", which is SSat when measured only with respect to sensor saturation, not subsequent amplifier clipping?

Note: for the first [Ssnr40] higher is better; for the second [Ssat, roughly speaking] lower is better!


The base ISO speed, measuring just the highlight handling capacity, is roughly the saturation based measurement Ssat done in a way that depends solely on photosite saturation, with the risk of amplifier clipping eliminated by using a low enough exposure index setting.  This seems to be what DXO reports on its (confusingly named) graph of "ISO sensitivity" for the minimum EI setting.

For the D800, this minimum Ssat is 74, seen for camera EI setting of either 50 or 100 (so setting EI=50 use the same analog gain as EI=100, and just adjusts the light metering to choose twice the exposure level in auto modes, and probably adjusts default raw-to-JPEG conversion appropriately.)

DXOmark does not yet have the IQ280 tested, but I expect it to behave much like IQ180, for which DXOmark gives a base ISO speed of 28, seen for EI settings of 35, 50, and 100. (So again the analog gain is not varied one this range of EI settings; instead just the auto-exposure choice of exposure level and the default raw-to-JPEG conversion behavior vary.)

The ratio 74/28 of base ISO speeds is more than micro-lenses alone can account for, indicating that in comparison to the IQ180, the D800 has some combination of (a) CFAs that let through a high fraction of light, further increasing its quantum efficiency (b) lower full well capacity per unit area of sensor, reducing its ability to avoid sensor saturation and blooming.


If instead you mean sensitivity in the sense of "ISO speed", more closely resembling film speed measurement by been base on shadow handling and visible noise levels, then the official procedure is something like making a sequence of shots of gray cards at various exposure levels (adjusting illumination and or exposure duration), measuring the SNR of the output for each one, and finding the exposure level at which the SNR is 40:1. The result of that might surprise many people: for both an IQ280 and a D800, this will be overwhelmingly base an adequate photon counts, unrelated to the exposure index setting: it would be roughly the exposure level at which the photon count is 1600 (giving a shot noise count of 40, which overwhelms the dark noise in those sensors.)  At the base ISO-speed described above, the photon count is 12.7% of full well capacity, so:

The ISO speed of these sensors is roughly (base ISO speed) * (full well capacity * 0.127)/1600

Does anyone know the FWC for either of these sensors?  At a guess it is about 30,000 to 40,000, for which the ISO speed would be about 2.5 to 3 times the base ISO speeds that I estimate above.
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20630
  • Andrew Rodney
    • http://www.digitaldog.net/

Indeed, that is what I mean: "ETTR for minimizing noise" is nonsensical in that case; you give the sensor as much exposure as your shutter speed and aperture limits allow, and still do not get close to the right end of the "sensor photon count" histogram.
But that's not the histogram ETTR refers to. If you go back to the original ETTR article, here on LuLa, the "To The Right" is the right of the camera histogram representing the JPEG.
ETTR should be renamed to 'correct' or 'ideal' exposure for raw. If we had such a histogram, we'd shoot to that, not the right of it.
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

But that's not the histogram ETTR refers to. If you go back to the original ETTR article ...
Yes, unfortunately "expose to the right" is used in a couple of related but different meanings.


A. The name itself refers to exposure, which refers to the amount of light received by the sensor, and so one meaning is "gather as much light as you can without overfilling any photosite and thus blowing out some highlights", which optimizes signal-to-noise ratios. The relevant histogram there is the usually unavailable one of photosite occupancy fractions.

This approach makes sense, but only when you can use a long enough exposure time to (almost) fill some photo-sites, and so when the camera's base ISO speed can be used; it is irrelevant to elevated ISO speed situations.


B. Another version of ETTR  is about maximizing the numerical level placement in the raw file, on the basis that lower overall levels are recording less fine tonal graduations. This can be approximated by looking at the JPEG histogram that cameras offer, though a raw level histogram would be ideal.

This approach makes sense up to a point, but once the increment between successive raw levels is distinctly smaller than the noise floor of the signal entering the ADC, further raising of raw levels (and thus further reduction of the size of the gaps between them) becomes irrelevant to IQ (even if it makes it easier for people who like to compare cameras by comparing their raw histograms.) With ADCs whose dynamic range (bit depth, roughly) is significantly greater than the dynamic range of the incoming signal, this goal is irrelevant: amplifying to place the signal from a full well just below maximum raw level will at the same time have raw level spacing well under the noise floor, and increasing the raw level placement improves nothing while risking loss of highlight detail, so for IQ purposes, fixed gain at this level is probably a perfectly good approach. Maybe there is some convenience to authors of raw processing software and students of raw histograms if the camera then bit-shifts the ADC output levels before recording to the raw file, but that is of no benefit to IQ.
Logged
Pages: [1] 2   Go Up