• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

What does higher ISO in digital give us? Is it really a higher ISO?

Asher Kelman

OPF Owner/Editor-in-Chief
Guys,

In commenting on the M9 after reading every review and examining every image I could find, I realized that one should ask oneself, "What does ISO mean?". In film, I believe it's based on how the shadows are exposed in what light. With digital, they seem to just up the gain and increase contrast so we get a usable image. however, the grey shadow areas are not really there anymore. So are we deluding ourselves when we turn up the dial? Is it just like pushing film B&W where all one is doing is increasing the contrast? In a pushed film one can print the image but there can be an absence of the lower tonalities.

So what's happening with Digital?

and why do I ask? Well in stage performances of classical music, the musicians are lit from above. This gives bright hair and shoulders but there are shadows under each facial feature especially as the faces are lit by glancing rays from above. Here's an area where true high ISO imaging would help. The shaded areas would be represented by even tonalities and not blotchy grain. Am I asking too much? Would I be better off with high ISO film?

I discovered this Wikipedia reference if that helps figure things out! Essentially we should be getting a Noise based ISO, latitude and a saturation, (bright pixel blooming limit), based speed. But how do we use this data and what do the settings on the camera really mean?

Thanks,

Asher
 
In a linear world, cranking up the gain cranks up the contrast. Fortunately, we live in a logarithmic world so that no matter the gain, a stop is still a stop.

With ideal (noiseless) physics, we could have an infinitely long palette of tones from light to dark. If we had a system that produced images with a dynamic range of say eight stops, we would need to decide which set of eight stops in the infinite palette we wanted to use. We would do that by adjusting the gain.

In fact, this is what we do except that the palette is not infinite. It is noise limited at one end and melts at the other.

In this thread, The the three images have ISOs from 400 to 6400 and have pretty similar dynamic range.
 

Doug Kerr

Well-known member
Hi, Asher,

In commenting on the M9 after reading every review and examining every image I could find, I realized that one should ask oneself, "What does ISO mean?".

The question you pose is an important , and challenging, one, and I am glad to see you (by citation of the Wikipedia article) call attention to some of the real technical principles involved.

Note in particular its mention of the recently-instituted REI (recommended exposure index) convention, which essentially says that manufacturers can assign ISO sensitivity ratings arbitrarily.

There can be two motives ascribed to exploiting this convention:

• The manufacturer can, in conformity with the standard, assign ISO sensitivity ratings such that the photographic exposure (the combination of exposure time and effective f-number) recommended by an external exposure meter calibrated in accordance with ISO standards, with its exposure index set to the manufacturer's rating, will produce what the manufacturer thinks is the "most desirable" exposure result for most uses.

• The manufacturer can, in conformity with the standard, assign ISO sensitivity ratings that will seem attractive to potential buyers.

This is discussed at some length here:

http://doug.kerr.home.att.net/pumpkin/SOS_REI.pdf

Although I know this may sound like a trviality, if we have, when moving from one camera to another a higher available situation of:

• Saturation-based ISI sensitivity
• Noise-based ISO sensitivity

with unchanged values of:

• Latitude
• Signal-to-noise ratio
• Dynamic range

then we have moved to a situation of being able to get the same performance, for some imagined scene, with a lower value of incident illuminance.

Of course, if all these things do not change (or not change) as listed, then it gets much more complicated to predict, or characterize, our new situation.

Best regards,

Doug
 

Asher Kelman

OPF Owner/Editor-in-Chief
In a linear world, cranking up the gain cranks up the contrast. Fortunately, we live in a logarithmic world so that no matter the gain, a stop is still a stop.

With ideal (noiseless) physics, we could have an infinitely long palette of tones from light to dark. If we had a system that produced images with a dynamic range of say eight stops, we would need to decide which set of eight stops in the infinite palette we wanted to use. We would do that by adjusting the gain.

In fact, this is what we do except that the palette is not infinite. It is noise limited at one end and melts at the other.

In this thread, The the three images have ISOs from 400 to 6400 and have pretty similar dynamic range.
Thanks, Winston,

In the analog world, the "gain" of pushing just increases contrast and at the most can add a 1/3 to maybe 1 real stop of sensitivity. The latter depends on the silver emulsion chemistry of each film. Even TMax 3200 is really 1000 ISO. One gains a great image but the grain becomes clumps.

In your pictures, if we identify the identical structures on the roof, uniform areas, looking at the structure at 100% would give us a good idea of what happens to the imaging a particular part of the roof at increasing ISO settings. So this is a great real world test, except that there was no tripod? so the blown up images might be blurred.

Asher
 

Asher Kelman

OPF Owner/Editor-in-Chief
Hi, Asher,



..... Note in particular its mention of the recently-instituted REI (recommended exposure index) convention, which essentially says that manufacturers can assign ISO sensitivity ratings arbitrarily....



This is discussed at some length here:

http://doug.kerr.home.att.net/pumpkin/SOS_REI.pdf

Although I know this may sound like a trviality, if we have, when moving from one camera to another a higher available situation of:

• Saturation-based ISI sensitivity
• Noise-based ISO sensitivity

with unchanged values of:

• Latitude
• Signal-to-noise ratio
• Dynamic range

then we have moved to a situation of being able to get the same performance, for some imagined scene, with a lower value of incident illuminance.

Of course, if all these things do not change (or not change) as listed, then it gets much more complicated to predict, or characterize, our new situation.

So, Doug,

I think I'll shoot the next event with increasing ISO and also with set ISO but over-exposing a tad. At some point, I might find a setting where the picture can be taken but the shadowed areas of faces are cleaner. In a way, I'm looking for film ISO, noise based, with a signal to noise ratio, that's just acceptable, I guess, of the top of my head, 10:1, although subtle changes near that I doubt I would discern.

Asher
 
I discovered this Wikipedia reference if that helps figure things out! Essentially we should be getting a Noise based ISO, latitude and a saturation, (bright pixel blooming limit), based speed. But how do we use this data and what do the settings on the camera really mean?

I think it's important to consider the following.

A doped silicon based sensor has a given native sensitivity to certain wavelenghts of light. The spectral light sensitivity, which roughly extends from 350 to 1000 nanometer, is reduced by overlaying structures which provide gates for the signals to travel, and stuctures that function as transistors. These structures and connections can be somewhat translucent but they do absorb energy that could have otherwise generated a photovoltaic energy (they also reduce the storage capacity). In addition, in a Bayer CFA structure, the light sensitive surface is covered with color filters that alone roughly reduce the effective energy that reaches the sensor to 1/3rd. So in effect we are only able to store a fraction of the energy that's falling on the sensor.

Yet there is enough energy in average daylight reflections to accumulate a charge equivalent up to some 1500 electrons per square micron, in a fraction of a second. We typically seek to maximize the exposure level (expose to the right) in order to maximize the Signal to Noise ratio, but we also want to avoid overexposing the capacity of the sensor to store the generated charge. Thus we need to opitmize the overall exposure level between maximum exposure and clipping exposure.

With an optimized exposure a single sensel can store some 10000 to 80000 electrons. The larger storage capacities are usually only possible with relatively large area sensels. Current 6.4 micron sensel pitch sensors can store roughly 60000 electrons per sensel (which is 60000/(6.4x6.4 micron)= 1465 electrons per micron). To encode 60000 individual electrons as a saturation level we need 16 bits. Much current Raw data is made available at 14-bit precision, which means that each bit of precision represents an average of 4 electrons per digital number (DN). This is accomplished by a "gain" setting at the moment of analog-to-digital (ADC) conversion.

When we underexpose, we will only need to encode smaller electron counts. If we e.g. only have a maximum of 15000 electrons (2 stops less exposure) to encode per DN, we can do that in 14 bits, but each bit (DN) will then represent 1 electron instead of an average of 4. The gain needs to be increased to do that, and its done with the ISO setting, so that a maximum charge of only 15000 electrons is output as white instead of -2 EV gray, and black as black. When the native sensitivity of the sensor is approx. ISO 100, we would have to increase the ISO to 400 to avoid 'under exposure' of white.

Because photons (1 photon generates 1 electron) arrive at somewhat random times during the exposure, the signal becomes relatively more noisy as we record fewer and fewer electrons. By boosting the weaker signal by increasing the gain, this will become more visible, and the boosting will also increase the background noise that's generated by the ADC and other electronics.

We can boost the gain or ISO to even higher values, but we will be generating more randomness in the converted signal, and larger jumps/gaps between digital numbers. So beyond a certain ISO level, quality will drop faster. The question then becomes how much we can tolerate, but increasing ISO from the basic sensitivity up always results in lower image quality (which we may need to do to get a shot with acceptable shutterspeed anyway).

So, to answer the question "What do the settings on the camera really mean?":
ISO changes the gain used by the ADC, but it also increases the noise until the Signal to Noise ratio becomes too low for our quality standards.

The easiest method to determine the highest tolerable ISO for low light exposures is by starting with a correct 'manual' (M setting) exposure at base ISO (usually ISO 100 or 200). Then underexpose a sequence of exposures, each by halving the exposure while boosting ISO x2. There will be a point where only serious postprocessing of the noise will still produce an image of adequate quality.

Another side effect of boosting the gain is that the total Dynamic Range (DR), i.e. the range of brightnesses that can be discriminated in a single exposure, is reduced. IOW while we may be able to get a white to be encoded as a (grainy/noisy) white, there will be less and less definition in the shadows, and what's visble will be relatively (much) more noisy than the brighter areas. Smaller sensel sensors will suffer faster and more than larger sensel sensors.

Cheers,
Bart
 

Asher Kelman

OPF Owner/Editor-in-Chief
Bart,

Thanks for the excellent response. Straight off this would mean that the 5DII is likely to have less noise at high iso than the new 7D unless they have, for example, made the structures overlying the sensel well less absorbing for photons or resulting electrons.

Next it would appear that merely calculating the voltages in 16BIT might decrease the jumps in values and should that give a better picture, all things being equal otherwise.

Lastly, is there anyway to expose to the right but set the camera to protect the highlights? IOW, can't the circuits be cut off before being filled?

Asher
 

Mike Shimwell

New member
Bart, a couple of thoughts:

- does boosting gain increase ADC noise if the amplifier is befopre the ADC?
- If we have an initial well capaicty of 64000 electrons and sensitivity of iso100 and a 14bit adc then we should be able to maintain dynamic range through to iso400. of course amplifier noise and shot noise (the random time of arrival of photons) degrade the signal, or alternatively viewed the signal still contains all the info required but is smaller with no reduction or a small increase in noise
- In this case then dynamic range would deteriorate above iso400, although you may still get a better signal by exposing at iso800 rather than pushing an iso400 exposure.
- I read somewhere (astrophotography site) that the 5D1 has 'unity gain', defined as 1 electron equals one adc count (12 bit) at iso 1600. This is consistent with a 64000 electron well capacity, which iirc is about correct.
- Of course, noise is the big bad guy as even 14 bit adc cameras tend to have noise that is higher than 14bits below saturation at base iso. (incidentally, I think true base iso for the 1Ds3 is slightly higher than 100 - dynamic range is higher at 200 than 100 (DXo Mark).

Your comments would be interesting.


Asher.

Almost certainly the 5D2 will have lower noise than the 7D - unless there has been a very significant jump in sensor tech. It would be interesting to compare to a 5D1 as I would expect higher per pixel noise, that may be offset by the increased number of pixels.

Moving to a 16 bit converter would give you the ability to read out all the data from a 64000 electron capacity well at base iso (though the additional bits would have no value as you increased the iso , but as noted above, noise is already above the 14bit floor so really you probably wouldn't see the information except (maybe) in stacking applications.

I don't think anyone has worked out a way to stop counting as a well fills - and even if you did, you'd need to start again when it overflowed. Fuji's double pixel approach is an attempt at achieving something like this. Effectively a high capcity/low sensitivity well combined with a low capcacity high sensitivity well. If the latter overflows then they take the signal from the bigger well to populate the higher numbers (presumably with less precision).

Mike
 

Asher Kelman

OPF Owner/Editor-in-Chief
Asher.

I don't think anyone has worked out a way to stop counting as a well fills - and even if you did, you'd need to start again when it overflowed.

I thought that about 5 years ago, Canon was involved in a consortium with HP and others at a lab derived from the work of a Professor at Stanford University. They have a CMOS sensor with every pixel addressable separately. So switching it off would be no issue. If one went further and looked at the 2cd derivative of filling v. time, then it could switch of any sensel individually that's filling too fast. So I guess Canon has had no need to introduce such technology until now when Sony and Nikon surpassed them in their latest cameras.

Fuji's double pixel approach is an attempt at achieving something like this. Effectively a high capcity/low sensitivity well combined with a low capcacity high sensitivity well. If the latter overflows then they take the signal from the bigger well to populate the higher numbers (presumably with less precision).

Mike

This Fuji option is something I should look into. I've heard that wedding photographers love the enormous dynamic range in getting the details of the brides gown and veil and still the shadowed nuances of the grooms tux, tie and silk cummerbund.

Maybe I should rent or steal one!

The other possibility for photographing musicians lit from above and shadowing their features heavily would be to use true high ISO film and expose for the shadows and develop for the highlights. Now what film and development would you suggest?

Asher
 

Mike Shimwell

New member
The other possibility for photographing musicians lit from above and shadowing their features heavily would be to use true high ISO film and expose for the shadows and develop for the highlights. Now what film and development would you suggest?

Asher

Tmax P3200 is about the fastest in 35mm - true iso 1000 with a long toe

Others like delta 3200 preferring it's tonality. May also be available in MF allowing the use of a leaf shutter body/lens combo.

Mike
 

Asher Kelman

OPF Owner/Editor-in-Chief
Tmax P3200 is about the fastest in 35mm - true iso 1000 with a long toe

Others like delta 3200 preferring it's tonality. May also be available in MF allowing the use of a leaf shutter body/lens combo.
Any advice on processing for these poorly lit and over lit conditions of a musician on stage, lights from above? What developer?

and what about color?

Asher
 
Bart,

Thanks for the excellent response. Straight off this would mean that the 5DII is likely to have less noise at high iso than the new 7D unless they have, for example, made the structures overlying the sensel well less absorbing for photons or resulting electrons.

In general, larger sensels are less noisy. However, it also depends on the quality of the rest of the electronic components used. They should have similar or better noise contribution for an overall better signal to noise performance.

Next it would appear that merely calculating the voltages in 16BIT might decrease the jumps in values and should that give a better picture, all things being equal otherwise.

Only calculating with more bits won't change the fundamental performance, it just scales the numbers. But perhaps I misunderstand what you are saying?

Lastly, is there anyway to expose to the right but set the camera to protect the highlights? IOW, can't the circuits be cut off before being filled?

Other than watching the histogram and the clipping indicators after a test shot, no. In "Live View" some cameras allow to display a Live histogram, but AFAIK that still depends on some weighted average for the entire tonemapped image.

I'm still hoping that some day there will be a true clipping indicator (based on the Raw data instead of the JPEG + tone-curve) next to the histogram display. That can be used as a guide e.g. to include, or just exclude, highlights. One could also include a clipping indicator with Live View, as long as it's fed by Raw data.

Bart
 
Bart, a couple of thoughts:

- does boosting gain increase ADC noise if the amplifier is befopre the ADC?

All amplifiers add some noise of their own. However, if a pre-amplifier has low noise, and a later amplifier has to amplify less, then the combined result may be better. It's hard to say what happens in general, because it ultimately depends on how all components work together in a specific camera body.

- If we have an initial well capaicty of 64000 electrons and sensitivity of iso100 and a 14bit adc then we should be able to maintain dynamic range through to iso400. of course amplifier noise and shot noise (the random time of arrival of photons) degrade the signal, or alternatively viewed the signal still contains all the info required but is smaller with no reduction or a small increase in noise

Yes, up to ISO 400 would in that case be relatively good, although there will be an amplification of noise, so Dynamic range will suffer a bit. Beyond ISO 400, the quality of the amplifiers becomes very important, but there may be a benefit to boost ISO to 800 (or maybe even 1600) compared to an 'exposure' boost in software at Raw conversion.

I have tested my 1Ds Mark III for noise in this thread. That also shows that while Read noise for ISO 100 and 200 is almost the same, it is still not much higher at ISO 400, but slightly more elevated at ISO 800. Higher ISOs boost the noise so much that there is little benefit to do that in camera. It is probably better to just underexpose the image, and compensate when Raw processing. The Raw processing doesn't add noise where the in camera amplifiers do.

- In this case then dynamic range would deteriorate above iso400, although you may still get a better signal by exposing at iso800 rather than pushing an iso400 exposure.

Yes one may, but the difference becomes very small. Personally I don't use anything above ISO 800, unless I shoot JPEGs (which I rarely do).

- I read somewhere (astrophotography site) that the 5D1 has 'unity gain', defined as 1 electron equals one adc count (12 bit) at iso 1600. This is consistent with a 64000 electron well capacity, which iirc is about correct.

Yes, many tests keep producing similar sensitivities. There is not much improvement over the years, although the latests Nikons seem to be a bit more sensitive (but their handling/clipping of the black point in RAW makes it a problematic choice for astrophotography).

What is needed for a boost in quality, is a higher conversion efficiency of photons (e.g. by using back projection through the silicon) and a larger storage capacity (saturation or clipping level in electrons), together with low noise circuitry. There are other tweaks possible, such as multiple read-outs with different gain, recombined after ADC quantization. I just came across a document describing just that.

Maybe Canon has something new up their sleeve, who knows (and Nikon isn't sitting still anymore either).

- Of course, noise is the big bad guy as even 14 bit adc cameras tend to have noise that is higher than 14bits below saturation at base iso. (incidentally, I think true base iso for the 1Ds3 is slightly higher than 100 - dynamic range is higher at 200 than 100 (DXo Mark).

My tests didn't show that, although there might be small differences between different copies of the camera. On my unit, ISO 100 has the highest Dynamic Range (11.37 stops), but ISO 200 (11.26 stops) is close. ISO 400 drops a bit to 10.96 stops, and ISO 800 drops further to 10.36 stops. ISO 1600 starts to dive at 9.57 stops, and ISO H is artificial.

Cheers,
Bart
 

Doug Kerr

Well-known member
Hi, Asher,

I discovered this Wikipedia reference if that helps figure things out!
That article is generally quite good.

There is a very slight lapse when it first speaks of Standard Output Sensitivity (SOS), one of the new measures of digital camera sensitivity introduced in the 2006 revision of ISO 12232. The article says:

The Standard Output Specification (SOS) technique, also new in the 2006 version of the standard, effectively specifies that the average level in the sRGB image must be 18% gray plus or minus 1/3 stop when exposed per the EI with no exposure compensation.

In fact, ISO 12232 does not prescribe what average photometric exposure will occur on a "metered" shot. It could only do that in concert with a specification of the (assumed) behavior of the automatic exposure system ("exposure meter calibration"), not part of the scope of this standard.

But it is clear that the authors of ISO 12232 "have in mind" the calibration of automatic exposure control systems prescribed by ISO 2721.

A more complete statement of the situation would be:

The Standard Output Specification (SOS) technique, also new in the 2006 version of the standard, effectively specifies that the average luminance level implied by the sRGB image must be 18% of the maximum recordable luminance level, plus or minus 1/3 stop, when the exposure is controlled by an automatic exposure control system calibrated per ISO 2721 (with no exposure compensation), with the exposure index of the metering process set to the SOS value.

Incidentally, we could (using the same language) say this about the saturation-based "ISO speed" rating of sensitivity:

The definition of the (saturation-based) ISO speed effectively specifies that the average luminance level implied by the sRGB image must be 12.8% of the maximum recordable luminance level, plus or minus 1/3 stop, when the exposure is controlled by an automatic exposure control system per ISO 2721 (with no exposure compensation), with the exposure index of the metering process set to the saturation-based ISO speed value.

It is the mathematics behind that that leads to the conclusion, reported in the Wikipedia article (and earlier confirmed in my paper on the subject), that the ISO SOS sensitivity rating will always be 0.704 times the saturation-based "ISO speed" for any given camera (in some particular "ISO speed" mode, of course).

Typically, the ISO sensitivity assessment on Canon EOS cameras has over the years been closer to the "ISO SOS" definition than the (saturation based) "ISO speed".

One thing that has been nicely fixed in the 2006 version of ISO 12232 is that in the earlier version, the parameter called Hsat, called the saturation photometric exposure for the camera, and central to the definition of sensitivity, was defined as 0.707 times the saturation photometric exposure of the camera. (Try and make all the equations fit together in the face of that!)

This was the result of a bungled treatment of the infamous "half-stop headroom" issue.

Fortunately, in ISO 12232:2006, Hsat, the saturation photometric exposure for the camera, is defined as just that.

Best regards,

Doug
 

Doug Kerr

Well-known member
I haven't spoken much about the concept of "exposure index latitude" defined by ISO 12332:2006, but this might be a good time to do it. I'm making this up as I go along.

Exposure index

First, let me clarify the precise sense in which I use the term exposure index (EI) (it is used rather inconsistently in the standard):

Exposure index
is what we tell the exposure meter is the ISO sensitivity of the film or digital sensor system.

In a traditional analog (external) exposure meter we enter the exposure index by turning a wheel, part of the exposure calculator (which is a special type of circular slide rule for solving the "exposure equation"). When using film rated by the manufacturer at ISO 400, we might set the dial to ISO 400, or we might set it to ISO 200 or ISO 800, in order to do what "exposure compensation" does on an integrated automatic exposure system.

In a digital camera, where the "exposure meter" really turns out to be (except in the "manual exposure" mode) an automatic exposure control system, basically when we set the camera to its "ISO 400" sensitivity mode, we also have set into the exposure meter "ISO 400" as the EI. We have no specific control for setting the EI to other than the nominal ISO speed we have set.

But, playing the equivalence mentioned above in the other direction, if we apply exposure compensation, this has the same result as setting the EI to something other than the "rated" ISO sensitivity in use. If we set the camera sensitivity mode to ISO 400, but set an exposure compensation of +1 EV, we have in effect "set into the meter" an EI of ISO 200.

Sensitivity ratings

Before I can discuss the definition of "latitude" I'm going to review some of the measures of "sensitivity" provided for in the standard.

ISO speed (saturation based) - Defines the sensitivity in terms of the photometric exposure required to take the sensor chain to "saturation". If exposure is set by a metering system calibrated per ISO 2721, then a part of the scene that has the same luminance as the average for the whole scene will produce an exposure result that corresponds to a luminance of 12.7% of the luminance that would produce saturation. (Sorry to say it in such a complicated way, but simpler recitations are just not complete or correct.)

ISO standard output sensitivity (SOS) The concept is identical to that for ISO speed (saturation), except that if exposure is set by a metering system calibrated per ISO 2721, then a part of the scene that has the same luminance as the average for the whole scene will produce an exposure result that corresponds to a luminance 18% of the luminance that would produce saturation.

This value of this measure is always 0.704 times the ISO speed (saturation).

Noise-based speed Snoise40 - Based on the photometric exposure at which the signal to noise ratio of the image region is 40:1 (a result thought of as the "first excellent image", where "first" means in the sense of us increasing the SNR until we first consider the image "excellent" from a noise standpoint).

Noise-based speed Snoise10 - Based on the photometric exposure at which the signal to noise ratio of the image region is 10:1 (a result thought of as the "first acceptable image", in the sense described above).

Latitude

Now with all the ingredients in hand, we can look into the ISO measure, ISO Speed Latitude.

It is stated this way:

ISO Speed Latitude yyy - zzz

where yyy is the ISO Speed (saturation), and zzz is the value of Snoise10.

The implications of this are:

• yyy is the smallest EI that one could use ("set into the exposure meter") without risking overexposure (for a scene whose average luminance is 12.8% of its maximum luminance). Looked at in another way, it is the EI that one would set to "expose fully to the right with no margin" for a scene whose average luminance is 12.8% of its maximum luminance.

• zzz is the largest EI one could use and still get an SNR of 10:1 for a scene region whose luminance is a certain fraction of the saturation luminance that I still need to reckon (I'll amend this once I've made that calculation).

Best regards,

Doug
 

Doug Kerr

Well-known member
This is an update on the latitude aspect of my previous report.

************
Latitude

Now with all the ingredients in hand, we can look into the measure "ISO Speed Latitude".

It is stated this way:

ISO Speed Latitude yyy - zzz

where yyy is the ISO Speed (saturation), and zzz is the value of Snoise10.

The implications of this are:

• yyy is the smallest EI that one could use ("set into the exposure meter") without risking overexposure (for a scene whose average luminance is 12.8% of its maximum luminance). Looked at in another way, it is the EI that one would set to "expose fully to the right with no margin" for a scene whose average luminance is 12.8% of its maximum luminance. (This is sometimes described as the "overexposure-based EI".)

• zzz is the largest EI one could use ("set into the exposure meter") and still get an SNR of 10:1 for a scene region whose luminance is the average luminance of the scene. (This is sometimes described as the "underexposure-based EI".)

[I was surprised when it came out that way! The answer was so tidy I at first thought I had accidentally done "1=1" someplace.]

Why would one intentionally set the higher EI into the meter? To get a smaller aperture, or shorter exposure time (at the expense of underexposure)

Note that with the integrated automatic exposure control in our digital cameras, we do not have the direct opportunity to, for example, set the camera sensitivity to ISO 400 but tell the meter to use an EI of 800.

But we can set the camera sensitivity to ISO 400 and set an exposure compensation of -1 stop (in which case the meter will proceed as if it thought the sensitivity was ISO 800 - that is, as if the user had "dialed in" an EI of ISO 800).

Doug
 

Asher Kelman

OPF Owner/Editor-in-Chief
How to choose ISO settings for harsh high dynamic range lighting?

Why would one intentionally set the higher EI into the meter? To get a smaller aperture, or shorter exposure time (at the expense of underexposure)

Note that with the integrated automatic exposure control in our digital cameras, we do not have the direct opportunity to, for example, set the camera sensitivity to ISO 400 but tell the meter to use an EI of 800.

But we can set the camera sensitivity to ISO 400 and set an exposure compensation of -1 stop (in which case the meter will proceed as if it thought the sensitivity was ISO 800 - that is, as if the user had "dialed in" an EI of ISO 800).

Doug

Doug, Bart and anyone else,

As you know, I regularly shoot music and dance on stages lit mostly harshly from above. Your formula fits in with my current practice. The settings I choose are ad hoc, trial and error and not calculated from measurements of average and highest illumination in the scene.

My approach to ISO, following Bart's suggestion, is as follows. To avoid over-exposure in highlights, (the violinist's dainty hands vanish!), on such very high dynamic scenes, I do set the ISO lower than that which I'd have chosen normally to cover the histogram, set to the right. I then under-expose by 0.3 to 1.5 stops. I'd like to be able to arrive at the settings following measurement. The formula above suggest a path to this. Now, Doug, with your average being 12.5% of max, how do we use this in practice, so the choice of underexposure compensation and ISO setting can be made more rational.

The maximum exposure can be fairly readily discovered. However, it seems to me that finding the average exposure across the scene of interest is more difficult. Do you suggest we take a light meter reading of the scene with the camera making such an average? Would you defocus the scene to get such an average reading.

Then, how do we go from these exposure levels to find the optimum ISO and under-exposure assuming we want the aperture no more restrictive than f3.5 to 4.5 and the speed at 1/80 -1/200 second.

Any ideas?

Asher
 

Doug Kerr

Well-known member
Hi, Asher,

You articulate the classical dilemma of photographic exposure metering, and invoke the premise of the Zone System (even though it is not usually characterized as a solution to the metering problem, but rather on much more "artsy" terms - but we photometricians know better)!

The maximum exposure can be fairly readily discovered. However, it seems to me that finding the average exposure across the scene of interest is more difficult. Do you suggest we take a light meter reading of the scene with the camera making such an average? Would you defocus the scene to get such an average reading.

Then, how do we go from these exposure levels to find the optimum ISO and under-exposure assuming we want the aperture no more restrictive than f3.5 to 4.5 and the speed at 1/80 -1/200 second.

Firstly, to go to fundamental principles, if our real desire is to use the maximum possible range of photometric exposure (in the interest of best noise performance, or to go at it another way, in interest of best exploiting the dynamic range potential of our specific system), there is no substitute for making a luminance measurement of the brightest object for which we want to capture detail.

No finessing or distressing of average scene luminance measurement will get us there. And in fact, the average luminance of the scene is not a property of any interest (except in the practical aspect that the exposure metering system may run from it).

The choice of an ISO sensitivity is then a balance between:

• The impact on noise performance
• The realities of attaining the needed exposure with "practical" f/number and shutter speed.

Suppose that our camera offers a combination of available long focal length and a small "spot" metering pattern such that we can then in fact take a luminance reading on what we feel (or know) to be the highest luminance object in the scene. Then, how can we use that to consistently make that exposure determination?

We don't know (despite the existence of various international standards for automatic exposure control systems and teh assessment of ISO sensitivity) how the metering system of our cameras are "calibrated" in the spot mode vis-à-vis the saturation photometric exposure of the sensor at some ISO sensitivity setting. (And there is a further complication that this saturation photometric exposure will not be the same for different chromaticities of the light from that scene element, owing to the differences between the three "channels" of the imaging system.)

So perhaps the best bet (if practical) would be, in a test scene, to make a series of test shots at some well-behaved ISO sensitivity using the spot metering capability on what turns out to be the highest-luminance item in the field, varying the exposure from that "recommended" by the metering system (either by switching to manual exposure control and varying the shutter speed, for example, or by working with the exposure compensation setting) and then determine which is the hottest of those exposures that delivers detail across this "brightest" object.

Then, with those results in hand, you can establish a relationship that should be usable for that camera in a range of situations. But again this depends on the premise of being able to spot meter on the harpist's [hands, forehead] or whatever the critical scene element is.

It may even be that an actual "telescopic" spot meter will be called for (at least they are quiet!).

In any case, the choice of ISO sensitivity probably devolves to using the lowest one (in the "normal" range) that allows the proper exposure to be attained with "available" aperture and "acceptable" shutter speed.

Now, all this having been said, why is it that cameras do not survey the luminance on the sensor across the entire frame - pixel by pixel - and then make an "expose to the right" determination from that? Oh, that is indeed how many non-SLR cameras do it.

A question that did not occur to me until just this second is how to Canon EOS cameras with live view meter? I need to find that out. Maybe there is something magickal available to us in that.

Well, its time for breakfast ċum intelligence briefing, and then helping Carla get us ready for the big family gathering here.
[Please excuse the typographic peculiarity above, but the forum's Bowdler filter will not permit a canonical presentation of the Latin word for with, and I can't conveniently use the "c" with an overscore so beloved of your prior life.]

Happy Boxing Day, my friend.

Best regards,

Doug
 
Last edited by a moderator:
Firstly, to go to fundamental principles, if our real desire is to use the maximum possible range of photometric exposure (in the interest of best noise performance, or to go at it another way, in interest of best exploiting the dynamic range potential of our specific system), there is no substitute for making a luminance measurement of the brightest object for which we want to capture detail.

No finessing or distressing of average scene luminance measurement will get us there.

Indeed, all average scene luminance schemes do not account for scene contrast. A low contrast scene exposed at the average scene luminance will be relatively underexposed versus an Expose-to-the-right (ETTR) goal. A high scene contrast exposed at the average scene luminance may already clip at the scene highlights and at the same time "underexpose" the deepest scene shadows. Only an accurate measurement of the highlights will allow to expose them to be near the saturation point of the affected sensels.

The choice of an ISO sensitivity is then a balance between:

• The impact on noise performance
• The realities of attaining the needed exposure with "practical" f/number and shutter speed.

Indeed, those are the 2 issues we seek to balance/optimize.

So perhaps the best bet (if practical) would be, in a test scene, to make a series of test shots at some well-behaved ISO sensitivity using the spot metering capability on what turns out to be the highest-luminance item in the field, varying the exposure from that "recommended" by the metering system (either by switching to manual exposure control and varying the shutter speed, for example, or by working with the exposure compensation setting) and then determine which is the hottest of those exposures that delivers detail across this "brightest" object.

The problem remains that reliable reflected light metering of a high contrast scene is usually difficult due to a lack of suitable subjects/objects (of adequate size). Also, the specific lighting scenario that Asher is facing (uneven lighting angles across an orchest on stage) doesn't make life easier. For an optimal exposure level the best approach would be to use manual exposure based on an incident light metering on the stage, seeking the highest level across the stage.

We can know that when we start with an average/incident/gray card exposure level, the sensel's clipping point is approx. 3 and 1/3rd stops above that (assuming light with a spectral daylight distribution) on most Canon DSLRs. So we could also use a piece of white paper, one of the brightest surfaces in the scene (with the exception of direct/specular reflections from lightsources themselves, e.g. from the varnish of instruments or from sheet music stands or other metal instruments). When measured from the piece of (blank, no notes on it) paper, we correct the exposure upwards by 3 stops, and we're there. Unfortunately, paper is't a perfect target, due to optical brighteners and sheet thickness/opacity influence, but it's readily available, and it can be calibrated for (even off-scene if one brings his/her own target along). It's probably easier to use an incident light meter when the stage light is stable a while after it's fully switched on (obviously best tested before the performance ;-) ).

However, that photometrically optimal exposure level may well be (and in Asher's practice is) a level for which a very high ISO is required (given the shutterspeed and aperture restraints). In that case it's (for high ISOs) demonstrably better (=lower noise) to underexpose by one or 2 stops and push in postprocessing, than to push the ISO in camera. This is where theory and practice need to be balanced. It also allows to retain more of the specular reflection colors, although that's not the goal, and account for variables in the scene lighting angles we may have missed. Afterall, brightness of white paper also depends on the lighting angle since its reflection in not Lambertian but semi-specular/semi-diffuse.

Now, all this having been said, why is it that cameras do not survey the luminance on the sensor across the entire frame - pixel by pixel - and then make an "expose to the right" determination from that? Oh, that is indeed how many non-SLR cameras do it.

A question that did not occur to me until just this second is how to Canon EOS cameras with live view meter? I need to find that out. Maybe there is something magickal available to us in that.

Indeed, it is still hard to understand that this capability is not utilized by the manufacturers as a basis for setting the exposure when the data is available. The Live View histogram is very small, so it can only be used as an indication, and even then it's not obvious what the manufacturer did to the tonecurve before reaching the histogram tally). An exposure setting based on Raw clipping levels/percentages during Live View, seems trivial to implement and would make life a lot easier.

Cheers,
Bart
 

Asher Kelman

OPF Owner/Editor-in-Chief
Indeed, it is still hard to understand that this capability is not utilized by the manufacturers as a basis for setting the exposure when the data is available. The Live View histogram is very small, so it can only be used as an indication, and even then it's not obvious what the manufacturer did to the tonecurve before reaching the histogram tally). An exposure setting based on Raw clipping levels/percentages during Live View, seems trivial to implement and would make life a lot easier.

Cheers,
Bart
Bart,

Is it possible that external software that control shooting of Canon cameras can use the live picture data? In that case, perhaps there's an advantage to using my laptop to fire the camera.

Asher
 
Is it possible that external software that control shooting of Canon cameras can use the live picture data? In that case, perhaps there's an advantage to using my laptop to fire the camera.

Asher,

I'm not aware of one from personal experience that really uses Raw data, but you can always try with the Canon EOS Utility and shoot tethered. It allows a somewhat better/larger (RGB) histogram preview, but it is still based on tonemapped/Gamma adjusted data.

However, that will only help in determining optimal exposure (even if empirically by test shooting). It will still not use the best ISO setting to reduce noise to a minimum in your scenario. For that you must underexpose your high ISO setting and push in Post.

Cheers,
Bart
 

Doug Kerr

Well-known member
Hi, Bart,

It will still not use the best ISO setting to reduce noise to a minimum in your scenario. For that you must underexpose your high ISO setting and push in Post.
I'm not sure I recognize why that ploy is beneficial. Could you please help me with that?

Thanks.

Best regards,

Doug
 
It will still not use the best ISO setting to reduce noise to a minimum in your scenario. For that you must underexpose your high ISO setting and push in Post.

I'm not sure I recognize why that ploy is beneficial. Could you please help me with that?

Hi Doug,

It has to do with the specific treatment of High ISO settings, say above 400 or 800. It can depend a little on camera model, so I encourage those in need of high ISO shooting, e.g. to keep a high enough shutterspeed for hand held shooting, to do an experiment themselves.

Changing the ISO on a camera effectively amplifies the electron volt charge (caused by the actual exposure) during the Analog to Digital conversion (ADC) by changing the gain. However, that not only amplifies the photon shot noise, but it also amplifies the electronic noise from the camera's circuits generated during capture and charge read-out. The higher the ISO setting, the more this pre-ADC noise is going to be amplified, but also the lower the absolute exposure level is, the noisier the signal itself will be.

To get an optimal noise performance we need to find a balance between under-exposure noise, and amplified system-noise. At base ISO, for most Canon DSLRs approx. ISO 100 and for Nikons reportedly closer to 200, each increment in the 12 or 14 bit Digital Number (DN) requires several photons (e.g. in the order of 4 photons for an EOS 1Ds Mark 3), so there is a reasonable probability that we can detect small luminance differences despite some system noise. When the ISO is boosted to 400 and our exposure is quartered, the average DN difference will be caused by a single photon (AKA unity gain), so the photon shot noise will increase a lot.

In practice this will mean that the shadow exposure levels will start to drown in the noise floor, but the higher levels of exposure will only get noisier due to the amplified camera noise. However, be cause we are boosting the gain, we also need to reduce the highlight exposure level to avoid clipping (but that's what we want, recording reduced exposure levels), which will increase shot noise.

So far the theory, let's test how this works out in practice. I can do this only for the meaningful range of ISO settings of my own camera, but I expect Asher and other EOS 5D Mark 2 owners, or similar Nikon models, to better my results by approx. a stop, maybe two stops.

Here's the result of a test I did by determining the standard deviation of the 6 grayscale patches of a MacBeth Mini Colorchecker, by Raw converting in Capture One, and using Photoshop for the statistical readout of exactly the same cropped pixels between conversions. I made sure that the different exposures gave almost the exact same RGB output levels for the corresponding patches after Raw conversion. The push was performed in postprocessing only, and I made sure they resulted in similar RGB values, so I didn't rely on exposure slider values.

Noise standard deviation of a 50x49 pixel area of each patch.
HighISO+Push.png

As can be seen, the ISO 400 has lower noise than the ISO 800 group, and the ISO 800 group has lower noise than the ISO 1600 group. Within the ISO 800 group, there is little difference between the underexposed + pushed setting, but the pushed settings will have more highlight clipping latitude (a stop headroom). Within the ISO 1600 group there is also little difference, although the ISO 800 pushed 1 stop is slightly better than the rest, and again it has 1 stop overexposure headroom. Even ISO 400 pushed 2 stops is a bit better than the ISO 1600 gain setting. The differences within each group are actually very difficult to see, but they are measurable.

IMHO the 1Ds3 doesn't perform very well above ISO 1600 (for my type of use), so I didn't test higher ISO settings, but I expect the 5D2 to show similar performance at ISO 3200.

Cheers,
Bart
 

Cem_Usakligil

Well-known member
Hi Bart,

Thanks a lot for this very good explanation, you are such a valuable resource and very generous in sharing it with all of us. :).

...IMHO the 1Ds3 doesn't perform very well above ISO 1600 (for my type of use), so I didn't test higher ISO settings, but I expect the 5D2 to show similar performance at ISO 3200.
As we have discussed this matter in the past, I had done the tests for my 5DII and I can confirm your assumptions. Basically, if I need to shoot @ 3200 ISO, I actually shoot @ 1600 ISO +1EV instead. This gives me less noise and better control of the clipping of any highlights as an added bonus.

On a related note, I think the so called Low-light ISO figure of the DxO Labs can be used as a departure point when one does not have the time to do the tests. This low-light ISO is about:
The Low-Light ISO metric indicates the highest ISO sensitivity to which your camera can be set while maintaining a high quality, low-noise image (based on a Signal-to-Noise-Ratio [SNR] of 30dB, a dynamic range of 9EVs and a color depth of 18bits). As cameras improve, the highest ISO setting to produce 30dB, 9EVs, 18bits images will continuously increase, making this scale open. The Low-Light ISO metric is of primary importance in photojournalism, sports and action photography.
The DxO figure for the 5DII is 1815 and for the 1DsIII it is 1663. So it means that anything above ISO 1600 can better be pushed in the post processing than setting the ISO of the camera to 3200.
 
Hi Bart,

Thanks a lot for this very good explanation, you are such a valuable resource and very generous in sharing it with all of us. :).

Hi Cem, thanks.

As we have discussed this matter in the past, I had done the tests for my 5DII and I can confirm your assumptions. Basically, if I need to shoot @ 3200 ISO, I actually shoot @ 1600 ISO +1EV instead. This gives me less noise and better control of the clipping of any highlights as an added bonus.

Thanks for confirming that.

On a related note, I think the so called Low-light ISO figure of the DxO Labs can be used as a departure point when one does not have the time to do the tests.

Yes, given their criteria of what's good enough, that's helpful. They do however suggest using the approximate ISO setting mentioned, but didn't compare to a pushed-in-post version at a lower setting, so this index suggests the maximum usable ISO, not necessarily the best quality at that effective sensitivity. Setting to a next lower ISO and pushing in postprocessing may well produce (marginally) better results.

Cheers,
Bart
 

Doug Kerr

Well-known member
Hi, Bart,

Thanks you so much for that detailed discussion. It will take me a while to "get my head around" all of it.

I need to be certain of one matter as I do that. With regard to your noise analysis chart, comparing, for example, the "ISO 400" and "ISO 400 +1EV push" cases, was the actual photographic exposure (f/number and shutter speed):

a. The same in both cases (perhaps manually set)

b. Metered in both cases with no EC (zero EC)

or

c. Metered in both cases but with a non-zero EC in the second case, and if so, how much (I would assume 1 Ev) and in which direction.

My guess (based on my understanding of the "push" practice) is that the "ISO 400" case was shot metered, while the "ISO 400 +1 EV push" case was shot metered with an EC of -1 Ev (or in any event, the photographic exposure for the second case was 1 Ev less than the first case).

Thanks so much.

Best regards,

Doug
 
Hi, Bart,

Thanks you so much for that detailed discussion. It will take me a while to "get my head around" all of it.

I need to be certain of one matter as I do that. With regard to your noise analysis chart, comparing, for example, the "ISO 400" and "ISO 400 +1EV push" cases, was the actual photographic exposure (f/number and shutter speed):

[...]

c. Metered in both cases but with a non-zero EC in the second case, and if so, how much (I would assume 1 Ev) and in which direction.

It's c. The ISO setting is the ISO used on the camera, and is thus steering the gain setting used at ADC. The ISO 400 +1EV push was also set to ISO 400, but relatively under-exposed by using a shorter shutterspeed (half the exposure time of the ISO 400 proper). However, the under- exposure was corrected/pushed during Raw conversion to reach 'exactly' the same output RGB values.

My guess (based on my understanding of the "push" practice) is that the "ISO 400" case was shot metered, while the "ISO 400 +1 EV push" case was shot metered with an EC of -1 Ev (or in any event, the photographic exposure for the second case was 1 Ev less than the first case).

Correct, effectively exposed at -1 EV compared to the 'correct' exposure for that ISO, and subsequently 'pushed' in Raw conversion post-processing, thus mimicking a higher ISO.

This only works well starting a bit above unity gain levels, because the loss of shadow detail due to underexposure is not a real issue anymore since the lowest levels are mostly amplified noise with a swamped signal anyway. Under normal (at or below unity gain) ISO settings one get the best S/N ratios by exposing to the right.

Cheers,
Bart
 

Doug Kerr

Well-known member
Hiu, Bart,

Thanks so much. I'll work my way through your explanation. (I don't mean to imply that it is unclear; I just need to follow the trail.)

Thanks agin so much for your continuing assistance in this and so many other areas.

Best regards,

Doug
 

Doug Kerr

Well-known member
Yesterday I said (or maybe Asher said; its so hard to tell what with all the free editing I get):

Now, all this having been said, why is it that cameras do not survey the luminance on the sensor across the entire frame - pixel by pixel - and then make an "expose to the right" determination from that? Oh, that is indeed how many non-SLR cameras do it.

A question that did not occur to me until just this second is how to [do] Canon EOS cameras with live view meter? I need to find that out. Maybe there is something magickal available to us in that.

I have since looked a little into the metering in Live View mode for the Canon EOS 40D.

The metering in that mode is done from the sensor output (there would of course seem to be no other choice) - the Canon White Paper on the machine confirms that. It also describes the metering in that case as "evaluative". But is it hard to believe that this is the same as regular evaluative metering.

The manual describes the metering in that case as "focus-frame linked". "Focus frame" refers to the rectangle that appears in the center of the screen in Live View mode. It can be moved around with the joystick. Its role seems to be to show the initial scope of the "x5 magnified" view we get by pressing the "+" button (typically to aid manual focusing)

It does not seem that it actually has anything to do with focusing. For example, if we enable and then utilize the "AF when in Live View mode", pressing the AF button takes us out of Live View mode, the camera does regular AF (using whatever AF point selection may have been made), and returns us to Live View. The location of the "focus frame" seems to have no effect (and in fact the manual, somewhat clumsily, warns us of that).

But indeed it does have an influence on metering. One can see that by keeping the camera aimed consistently (perhaps on a tripod) and then moving the "focus frame" about with the joystick. As we move it to parts of the scene with differing luminance, we can see the exposure determination change.

If we make the setting in which the Live View display actually reflects expected exposure result (that is, if we have a lower-than-metered exposure set, perhaps in M mode, the image in Live View will be "dark"; without that setting, the system applies sort of an "automatic brightness control" to the Live View view), we can have a live histogram (in our choice of "brightness" or RGB). I of course have no idea how this is derived.

One thing I do notice is that it is very easy, while in "P" mode, for example, to aim the camera so that a lot of the histogram falls off the right-hand end of the scale.

In fact, if I capture the shot in that situation, then the histogram seen in play of the image seems pretty consistent with the one during Live View. (I have to do some more work to find out if that is exactly so or only approximately so.)

What I haven't yet determined is whether Live View shots are more susceptible to blown highlights than regular shots (as a result of the differences between their two metering approaches).

Based on my observations quoted at the head of this note, we might have hoped that the Live View metering might be more able to avoid blown highlights, since it has the potential of looking at every pixel. But somehow I fear that I mind find the contrary.

Well, I'm off to do man's work (delivering the service bulletins and hymn transcriptions to church - we won't actually be attending as Carla is completely exhausted from mounting yesterday's family Christmas dinner, with about a week's preparation).

Then I can return to God's work (physics, optics, mathematics, and images of nature's beauty.

Best regards,

Doug
 

Doug Kerr

Well-known member
Hi, Bart,

Well, I'm working my way thorough all this. I am having a little problem with your chart. I have assumed that the six "grayscale" patches on the mini CC that you mention are intended to be the patches often called 24 through 19 (from darkest to lightest).

But I am surprised that the noise values for "patch 6" on your table are all smaller than for "patch 5".

Normally, we would expect the noise values (for any given exposure setup and ISO setting) to rise monotonically with luminance, owing to the shot noise varying as sqrt(N).

All I can think of is that your exposure was so high that the lightest patch always saturated (which of course would cause a decrease in digital noise), but that doesn't seem likely given the relationship between the various entries in the "ISO 1600" group.

What am I missing?

One more question while I've got you up. Is the "native ISO" the one at which the photodetectors are saturated from an electronic standpoint (the so-called "well full" level) at the same photometric exposure at which the DN tops out?

This doesn't really get in the way of my trying to grasp all this, but is just a curiosity.

Thanks.

Best regards,

Doug
 
Top