Why expose ETTR and how to do it?

Started 3 months ago | Discussions
Erik Kaffehr
Erik Kaffehr Veteran Member • Posts: 5,180
Why expose ETTR and how to do it?
3

There are some more or less bombastic discussions about ETTR. There are often explanations that may be authoritative but not necessarily correct,

At the same time there is a simple explanation, that also happens to be correct. If we illuminate a set of pixels with constant light, their response will vary. The reason is that light arrives in quantums and different numbers of quantums will hit the pixels.

Lets take a ColorChecker and illuminate it with reasonably constant light, take an image of that color checker and open the raw file in RawDigger.

Here you can see the raw data in the raw file. There is a selection placed over the neutral patches and a histogram is shown on the left side showing the distributions of the pixel responses. It would be reasonable that each field would be represented by a single spike. But, what we see are kind of 'bell curves' the brightest patch is very narrow the darker patches are getting broader with a lower central spike. This is the effect of photon arrival statistics.

But, that was the bright side of the image. Movie goers know that there is also a dark side. The image shown here has a very dark side:

Here i have also selected the six neutral patches. The spikes in the histogram are much broader.

Let's choose the fourth brightest neutral patch:

Looking at say the G-channel (green pixels), we can se that the signal varies quite a lot from 7284 to 8067. The right column shows the standard of deviation (stdev). Stdev tells us how much the signal varies. The average value is 7693.5 and sigma is 137.0. We can use the ratio of average divided by stdev as a measure of signal quality. It is called Signal to Noise Ratio. For a clean signal SNR would be high. In this case it is about 56.

Now, let us check the dark side:

Here the average green channel is 169.5 and the standard deviation is 24.4, So, signal to noise ratio is 169 / 24, that is around 7. So the dark side is getting noisy.

What this shows, clearly I hope, that the image gets more noisy when exposure is reduced. For photon arrival statistics SNR is proportional to the square root of the exposure.

What this means is that we want to keep exposure as high as feasible to reduce noise in the darker areas. That is all ETTR is about.

But, increasing exposure means that highlights can be clipped and that can cause ugly artifacts.

So, ETTR is about using maximum feasible exposure without clipping highlights. Significant highlights that is.

So, how to achieve that? Modern digital cameras have some tools to handle that. These tools are:

  • Histograms
  • Blinkies
  • Zebras

Just to say, on most cameras these tools are not really showing raw data.

  • Histograms show the distribution of luminance in the image.
  • Blinkies show parts of the image that is clipped.
  • Zebras show parts of the images that would be overexposed as black and white stripes.

This is a real world image. Here I knew, from experience, that moon is pretty bright. It is just a sunlit block of pretty dark stone 380 000 km away. So, I checked the zebras on the and back down exposure until they went away and that increased exposure until they were just visible.

So, my take is:

  • Download and buy RawDigger, it is a great way to understand how light metering on your camera works.
  • Use zebras, if available. But, keep in mind that they may be conservative.
  • If you have a histogram with little signal on the right side, you can probably improve your IQ exposing more.
  • If you see a spike on the uttermost right side of the histogram, it probably say that there are a lot of pixels clipped.

There is no big reason to overdo ETTR. Most cameras have pretty good SNR at base ISO.
On a camera with a Sony sensor, it may not be a great idea to increase ISO get more ETTR exposure. ETTR is about exposure, not ISO.

Best regards

Erik

-- hide signature --

Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

SrMi
SrMi Senior Member • Posts: 1,658
Re: Why expose ETTR and how to do it?

Erik Kaffehr wrote:

<snip>

Nice summary, thanks.

There is no big reason to overdo ETTR. Most cameras have pretty good SNR at base ISO.

What does "good SNR at base ISO" mean?

I assume that you meant that most cameras have at base ISO large enough DR  (max. signal to noise "distance") so that losing a bit of DR due to underexposure is not a biggie.

On a camera with a Sony sensor, it may not be a great idea to increase ISO get more ETTR exposure. ETTR is about exposure, not ISO.

Dual gain sensors have kind of two base ISOs. Jim Kasson wrote how to optimize for that by increasing ISO in discrete steps.

Cheers,

- Srdjan

Best regards

Erik

Erik Kaffehr
OP Erik Kaffehr Veteran Member • Posts: 5,180
Re: Why expose ETTR and how to do it?
1

SrMi wrote:

Erik Kaffehr wrote:

<snip>

Nice summary, thanks.

There is no big reason to overdo ETTR. Most cameras have pretty good SNR at base ISO.

What does "good SNR at base ISO" mean?

I assume that you meant that most cameras have at base ISO large enough DR (max. signal to noise "distance") so that losing a bit of DR due to underexposure is not a biggie.

No, I mean that SNR is good enough not cause significant noise in mid tones, like blue sky or facial tones. SNR is proportional to exposure, so if exposure is reduced SNR will also reduce.

This figure from DxO-mark illustrates this, they peg 'good SNR' at 38 dB. All three cameras are above 38 dB at base ISO. Panasonic crosses the line at 117 ISO, Sony A7rIV at around 500 ISO and Hasselblad X1D at 800 ISO.  Note that dual gain does not affect SNR. That is because it does not change the photon count, just the voltage swing that the photon count is resulting in allowing better readout.

On a camera with a Sony sensor, it may not be a great idea to increase ISO get more ETTR exposure. ETTR is about exposure, not ISO.

Dual gain sensors have kind of two base ISOs. Jim Kasson wrote how to optimize for that by increasing ISO in discrete steps.

Dual gain improves DR, mostly. DR is maximum signal, divided by readout noise. What dual gain does is to reduce the capacitance of the photodiode, so it reduces maximum signal, but when reducing exposure, the full capacitance would not be utilized anyway. But reducing the capacitance increases the voltage swing and that allows a cleaner readout.

Here the effect of dual gain is clearly visible.

Best regards

Erik

-- hide signature --

Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

Dirk Dittert Regular Member • Posts: 408
Re: Why expose ETTR and how to do it?

Erik,

thanks for sharing!

I wonder if it would be possible to design an image sensor that can stop the exposure when a certain number of photosite reaches 100% capture capacity. Basically that would be a sensor with protection against clipping.

 Dirk Dittert's gear list:Dirk Dittert's gear list
Canon EOS 5D Mark IV Fujifilm X-T3
Erik Kaffehr
OP Erik Kaffehr Veteran Member • Posts: 5,180
Re: Why expose ETTR and how to do it?

Dirk Dittert wrote:

Erik,

thanks for sharing!

I wonder if it would be possible to design an image sensor that can stop the exposure when a certain number of photosite reaches 100% capture capacity. Basically that would be a sensor with protection against clipping.

I don't really think so. Stopping exposure is almost the same as clipping. What really happens is that the photodiode does not really saturate, but the voltage exceeds the maximum of the ADC.

The reason that we don't have global shutters in digital cameras is due to not being able to stop photon detection.
With interline CCDs, global shutter was possible as half the sensor area was used to save electron charges. With CMOS, it would also be possible to add a buffer area for charges, but that means reducing full well capacity.

The task sensor designers have is managing sensor estate. Adding more features reduces the area of the sensor collecting electron charges.

Keep also in mind that camera sensors are basically analogue devices and photosites are probably the most complex analogue devices made.
Modern CMOS technology allows the sensor to output digital signals, but most of the sensor is still an analogue device.

Best regards

Erik

-- hide signature --

Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

JimKasson
JimKasson Forum Pro • Posts: 31,384
Re: Why expose ETTR and how to do it?

Dirk Dittert wrote:

Erik,

thanks for sharing!

I wonder if it would be possible to design an image sensor that can stop the exposure when a certain number of photosite reaches 100% capture capacity. Basically that would be a sensor with protection against clipping.

Yes. I call those photon-counting sensors. Eric Fossum and others are working on them, but they’re a long way from commercial consumer camera use.

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Fujifilm GFX 100 Sony a7R IV Sony a9 II +1 more
SrMi
SrMi Senior Member • Posts: 1,658
Re: Why expose ETTR and how to do it?
2

Dirk Dittert wrote:

Erik,

thanks for sharing!

I wonder if it would be possible to design an image sensor that can stop the exposure when a certain number of photosite reaches 100% capture capacity. Basically that would be a sensor with protection against clipping.

Note that it is OK to clip highlights that are not relevant.

I would be happy if we had live histogram and highlights warnings that are based on raw data.

strawbale Senior Member • Posts: 1,496
Re: Why expose ETTR and how to do it?

Erik Kaffehr wrote:

SrMi wrote:

Erik Kaffehr wrote:

<snip>

Nice summary, thanks.

There is no big reason to overdo ETTR. Most cameras have pretty good SNR at base ISO.

What does "good SNR at base ISO" mean?

I assume that you meant that most cameras have at base ISO large enough DR (max. signal to noise "distance") so that losing a bit of DR due to underexposure is not a biggie.

No, I mean that SNR is good enough not cause significant noise in mid tones, like blue sky or facial tones. SNR is proportional to exposure, so if exposure is reduced SNR will also reduce.

This figure from DxO-mark illustrates this, they peg 'good SNR' at 38 dB. All three cameras are above 38 dB at base ISO. Panasonic crosses the line at 117 ISO, Sony A7rIV at around 500 ISO and Hasselblad X1D at 800 ISO. Note that dual gain does not affect SNR. That is because it does not change the photon count, just the voltage swing that the photon count is resulting in allowing better readout.

On a camera with a Sony sensor, it may not be a great idea to increase ISO get more ETTR exposure. ETTR is about exposure, not ISO.

Dual gain sensors have kind of two base ISOs. Jim Kasson wrote how to optimize for that by increasing ISO in discrete steps.

Dual gain improves DR, mostly. DR is maximum signal, divided by readout noise. What dual gain does is to reduce the capacitance of the photodiode, so it reduces maximum signal, but when reducing exposure, the full capacitance would not be utilized anyway. But reducing the capacitance increases the voltage swing and that allows a cleaner readout.

Here the effect of dual gain is clearly visible.

Best regards

Erik

Beginner's question: "how much better" is a SNR (18%) of, say, 42 compared to 36 (at same ISO)? Is 3dB difference halving the noise?

With 'good' (arbritrarily?) set at 38, even a A7Riv is only 'good' up to 500 ISO?

 strawbale's gear list:strawbale's gear list
Nikon Coolpix P7800
Erik Kaffehr
OP Erik Kaffehr Veteran Member • Posts: 5,180
Re: Why expose ETTR and how to do it?
1

strawbale wrote:

Erik Kaffehr wrote:

SrMi wrote:

Erik Kaffehr wrote:

<snip>

Nice summary, thanks.

There is no big reason to overdo ETTR. Most cameras have pretty good SNR at base ISO.

What does "good SNR at base ISO" mean?

I assume that you meant that most cameras have at base ISO large enough DR (max. signal to noise "distance") so that losing a bit of DR due to underexposure is not a biggie.

No, I mean that SNR is good enough not cause significant noise in mid tones, like blue sky or facial tones. SNR is proportional to exposure, so if exposure is reduced SNR will also reduce.

This figure from DxO-mark illustrates this, they peg 'good SNR' at 38 dB. All three cameras are above 38 dB at base ISO. Panasonic crosses the line at 117 ISO, Sony A7rIV at around 500 ISO and Hasselblad X1D at 800 ISO. Note that dual gain does not affect SNR. That is because it does not change the photon count, just the voltage swing that the photon count is resulting in allowing better readout.

On a camera with a Sony sensor, it may not be a great idea to increase ISO get more ETTR exposure. ETTR is about exposure, not ISO.

Dual gain sensors have kind of two base ISOs. Jim Kasson wrote how to optimize for that by increasing ISO in discrete steps.

Dual gain improves DR, mostly. DR is maximum signal, divided by readout noise. What dual gain does is to reduce the capacitance of the photodiode, so it reduces maximum signal, but when reducing exposure, the full capacitance would not be utilized anyway. But reducing the capacitance increases the voltage swing and that allows a cleaner readout.

Here the effect of dual gain is clearly visible.

Best regards

Erik

Beginner's question: "how much better" is a SNR (18%) of, say, 42 compared to 36 (at same ISO)? Is 3dB difference halving the noise?

I would think that 6dB would be halving the noise.

Just to say, this is a very good paper related to moise:

https://www.photonstophotos.net/Emil%20Martinec/noise.html

Best regards

Erik

-- hide signature --

Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

Dirk Dittert Regular Member • Posts: 408
Re: Why expose ETTR and how to do it?

How about doing that in software? You could take a pre-exposure starting at ISO 100.000 and decrease until you can calculate the where at which exposure you will run into clipping. That doesn‘t give you unlimited dynamic range but would avoid bad exposures.

 Dirk Dittert's gear list:Dirk Dittert's gear list
Canon EOS 5D Mark IV Fujifilm X-T3
strawbale Senior Member • Posts: 1,496
Re: Why expose ETTR and how to do it?

Erik Kaffehr wrote:

strawbale wrote:

Erik Kaffehr wrote:

SrMi wrote:

Erik Kaffehr wrote:

<snip>

Nice summary, thanks.

There is no big reason to overdo ETTR. Most cameras have pretty good SNR at base ISO.

What does "good SNR at base ISO" mean?

I assume that you meant that most cameras have at base ISO large enough DR (max. signal to noise "distance") so that losing a bit of DR due to underexposure is not a biggie.

No, I mean that SNR is good enough not cause significant noise in mid tones, like blue sky or facial tones. SNR is proportional to exposure, so if exposure is reduced SNR will also reduce.

This figure from DxO-mark illustrates this, they peg 'good SNR' at 38 dB. All three cameras are above 38 dB at base ISO. Panasonic crosses the line at 117 ISO, Sony A7rIV at around 500 ISO and Hasselblad X1D at 800 ISO. Note that dual gain does not affect SNR. That is because it does not change the photon count, just the voltage swing that the photon count is resulting in allowing better readout.

On a camera with a Sony sensor, it may not be a great idea to increase ISO get more ETTR exposure. ETTR is about exposure, not ISO.

Dual gain sensors have kind of two base ISOs. Jim Kasson wrote how to optimize for that by increasing ISO in discrete steps.

Dual gain improves DR, mostly. DR is maximum signal, divided by readout noise. What dual gain does is to reduce the capacitance of the photodiode, so it reduces maximum signal, but when reducing exposure, the full capacitance would not be utilized anyway. But reducing the capacitance increases the voltage swing and that allows a cleaner readout.

Here the effect of dual gain is clearly visible.

Best regards

Erik

Beginner's question: "how much better" is a SNR (18%) of, say, 42 compared to 36 (at same ISO)? Is 3dB difference halving the noise?

I would think that 6dB would be halving the noise.

Just to say, this is a very good paper related to moise:

https://www.photonstophotos.net/Emil%20Martinec/noise.html

Best regards

Erik

Thank you!

 strawbale's gear list:strawbale's gear list
Nikon Coolpix P7800
JimKasson
JimKasson Forum Pro • Posts: 31,384
Re: Why expose ETTR and how to do it?

Dirk Dittert wrote:

How about doing that in software? You could take a pre-exposure starting at ISO 100.000 and decrease until you can calculate the where at which exposure you will run into clipping. That doesn‘t give you unlimited dynamic range but would avoid bad exposures.

What you're asking for is something that many of us have asked for for more than a decade, it's basically automatic ETTR. There's no reason to do any more pre-exposure than is currently done to generate the EFV image; the data just needs to be processed differently by the camera.

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Fujifilm GFX 100 Sony a7R IV Sony a9 II +1 more
Paul JN Regular Member • Posts: 489
Re: Why expose ETTR and how to do it?

It’s over 30 years since I failed my physics A level so i cant remember what it was I failed to understand Nevertheless I find this read fascinating. Thanks for taking the time to write it.

 Paul JN's gear list:Paul JN's gear list
Ricoh GR III Leica Q Fujifilm X100V Sony a9 Fujifilm GFX 50R +10 more
SrMi
SrMi Senior Member • Posts: 1,658
Re: Why expose ETTR and how to do it?

JimKasson wrote:

Dirk Dittert wrote:

How about doing that in software? You could take a pre-exposure starting at ISO 100.000 and decrease until you can calculate the where at which exposure you will run into clipping. That doesn‘t give you unlimited dynamic range but would avoid bad exposures.

What you're asking for is something that many of us have asked for for more than a decade, it's basically automatic ETTR. There's no reason to do any more pre-exposure than is currently done to generate the EFV image; the data just needs to be processed differently by the camera.

Magic Lantern (custom firmware for some Canon cameras) has implemented automatic ETTR (link).

incoherent1 Contributing Member • Posts: 578
Re: Why expose ETTR and how to do it?

Erik Kaffehr wrote:

Dirk Dittert wrote:

Erik,

thanks for sharing!

I wonder if it would be possible to design an image sensor that can stop the exposure when a certain number of photosite reaches 100% capture capacity. Basically that would be a sensor with protection against clipping.

I don't really think so. Stopping exposure is almost the same as clipping. What really happens is that the photodiode does not really saturate, but the voltage exceeds the maximum of the ADC.

Yabut. Before actual exposure, during metering, the camera could determine which of the multiple metering spots was brightest ("has voltage exceeding maximum of ADC"), and adjust exposure (by whatever exposure mode the photograher has chosen) to point of full clipping (maximum voltage) (or some level below, adjustable by exposure compensation we are used to). What am I missing?

 incoherent1's gear list:incoherent1's gear list
Nikon D810 Nikon AF-S Micro-Nikkor 105mm F2.8G IF-ED VR Nikon AF-S Nikkor 24-120mm F4G ED VR Sigma 50mm F1.4 DG HSM | A Nikon 200-500mm F5.6E ED VR +1 more
JimKasson
JimKasson Forum Pro • Posts: 31,384
Re: Why expose ETTR and how to do it?

incoherent1 wrote:

Erik Kaffehr wrote:

Dirk Dittert wrote:

Erik,

thanks for sharing!

I wonder if it would be possible to design an image sensor that can stop the exposure when a certain number of photosite reaches 100% capture capacity. Basically that would be a sensor with protection against clipping.

I don't really think so. Stopping exposure is almost the same as clipping. What really happens is that the photodiode does not really saturate, but the voltage exceeds the maximum of the ADC.

Yabut. Before actual exposure, during metering, the camera could determine which of the multiple metering spots was brightest ("has voltage exceeding maximum of ADC"), and adjust exposure (by whatever exposure mode the photograher has chosen) to point of full clipping (maximum voltage) (or some level below, adjustable by exposure compensation we are used to). What am I missing?

Not much, but no manufacturer as far as I know, has bothered to implement this. There is Magic Lantern, if you have the right Canon.

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Fujifilm GFX 100 Sony a7R IV Sony a9 II +1 more
Keyboard shortcuts:
FForum MMy threads