Expose to the Right and High(er) ISO

John, thanks for that explanation. I may be incorrect, but you are basing your arguments on noise levels, are you not?

Noise levels are obviously not irrelevant, but to me at least, the purpose of ETTR is about tonal gradations more than about noise levels. These two are practically related of course.

If we use the LL article examples, a camera capable of 5 stops DR, with a 12-bit ADC provides 4096 distinct values. 2048 of those are used in the first (brightest) f-stop, 1024 for the next, etc.. Practically, this means I get better tonal gradations at the higher f-stops, which is what we want to exploit.

So, if I go to e.g. ISO 400, do I no longer have those 2048 levels available in the first f-stop any more? This is what I don't understand. Of course, I'm simplifying things, and it may be that although I do still have those 2048 levels, other parameters have worsened and whatever gains I have achieved by ETTR at ISO 400 have been made negligible by the other factors.

> Noise levels are obviously not irrelevant, but to me at least, the purpose of ETTR is about tonal gradations more than about noise levels.

Kris, while this is true if you're going to do some rather severe tonal adjustments like pulling hugh amounts of shadow detail or doing double conversions for HDR layering, imo for normal shots where you might just pull a bit of contrast it's largely irrelevant as the existing (normally exposed) data is more than adequate.

IMO even for noise issues, unless you're going to pull shadows ETTR gains you little if anything other than pixel-peeping quality.

- DL

I agree Don. Whether or not I employ the ETTR technique is another matter. I just want to understand it, and until I understand why its not applicable for higher ISOs, I don't feel I understand it.

Having said that, I almost always have to bump the exposure in LR by 0.5-1.5. So the ETTR technique caught my attention.

Hi, Kris,

KrisCarnmarker said:

So, if I go to e.g. ISO 400, do I no longer have those 2048 levels available in the first f-stop any more?

Assuming you use a corresponding exposure (one stop less exposure) when you go to from ISO 200 to ISO 400, then that matter is unchanged.

There will always be about 2048 discernible steps in the top half of the range of photometric exposure (H) to which the sensor can respond at a particular "ISO speed" setting. (I say "about" because I don't really know exactly how to come to grips with the black point being at about 128 units).

That top half will also be "the top one stop of the range of scene luminances" if we have fully exposed to the right (the highest scene luminance just barely receiving the saturation value of H and thus a digital output level of 4095).

If we expose one stop less than that, then our highest scene luminance will be at (about) 2047 units, and the top half of the range we are actually using contains (about) 1024 steps.

Note that if we are speaking of ISO speed settings that don't have bona fide gains of the analog amplifiers, but are constructed by digital scaling of the digitized outputs, then all bets are (somewhat) off.

Note also that these "step count" analyses really only works for a subject whose chromaticity is such that the r, g, and b sensor groups give the same output (not true, for example, for anything we could reasonably call "white" or "gray).

Excuse me for saying all this in such a tedious way, but if we don't, then there can be misunderstandings about what is meant.

Best regards,

Doug

Click to expand...

Click to expand...

Hi, John,

John Sheehy said:

What is the contrast setting in the 20D?

What is the average green RAW value for the grey card exposed with the metering at 0 EC, and what histogram peak does that result in on the camera?

Click to expand...

To give a more complete response:

1. The contrast setting was "minimum" ("-2").

2. For a metered exposure of a uniformly-illuminated neutral test card, using partial metering, with EC=0:

a. The average RGB values in the sRGB JPEG image were 118/118/118.

b. The camera histogram peak was at about 2.3 scale units (considering the scale to run from 0 at the far left to 5 at the far right, with each line representing an integer value).

c. In the raw data, the peak of the g1 or g2 histogram was at about 440 units (4095 scale).

Note with regard to the matter of the metered exposure, although ideally we would expect the metering equations to produce the same value of H (and thus of RGB) for a given part of a test scene between tests, in fact this is not quite so.

For example, the RGB value will vary somewhat at different ISO speed settings, or with different balances of shutter speed and aperture that make up the Ev the meter adopts. For example, if I use the Tv mode for metering, and set the shutter speed to 1/100 and take a metered shot of the test card, and then set the shutter speed to 1/200 sec and take another metered shot, the RGB values will differ.

The range of the discrepancy is not great - perhaps 1/2 stop at worst - and might in fact result from rounding phenomena in the whole exposure control chain, or perhaps from subtleties in the metering algorithm not embraced by our classical assumptions.

Note also that the Canon "handy check" of AE behavior (applicable to earlier EOS dSLR's, maybe not to the latest ones) would lead us to expect about RGB 116/116/116 for metered exposure of a uniform neutral test target.

Thanks for your input and insight.

Best regards,

Doug

Doug_Kerr said:

For example, the RGB value will vary somewhat at different ISO speed settings, or with different balances of shutter speed and aperture that make up the Ev the meter adopts. For example, if I use the Tv mode for metering, and set the shutter speed to 1/100 and take a metered shot of the test card, and then set the shutter speed to 1/200 sec and take another metered shot, the RGB values will differ.

Click to expand...

A likely candidate is light entering the viewfinder from behind the camera, and throwing off the brightness of the focus screen. Otherwise I've not seen large fluctuations such as you describe.

Bart

Hi, Bart,

Bart_van_der_Wolf said:

A likely candidate is light entering the viewfinder from behind the camera, and throwing off the brightness of the focus screen. Otherwise I've not seen large fluctuations such as you describe.

Click to expand...

Well, Duh, I knew that, but I have not at all been diligent about preventing that. I will certainly take the necessary precautions in the future (in fact, I have planned to redo some of my tests in just a few minutes!).

Thanks so much.

Best regards,

Doug

KrisCarnmarker said:

Having said that, I almost always have to bump the exposure in LR by 0.5-1.5. So the ETTR technique caught my attention.

Click to expand...

Then perhaps you're just under-exposing and should adjust your technique. I have a mental feedback loop between C1 and my exposure technique and try to keep my RC conversion to within +/- 0.5 stops. IMO regularly adjusting +1 or more is not a good idea.

- DL

Don Lashier said:

Then perhaps you're just under-exposing and should adjust your technique. I have a mental feedback loop between C1 and my exposure technique and try to keep my RC conversion to within +/- 0.5 stops. IMO regularly adjusting +1 or more is not a good idea.
- DL

Click to expand...

I agree with Don: >adjusting +1 or more is not a good idea.< as a general rule. The converters have different headroom, though.... some are better in the shadows and weaker at the highlights, and vice versa.

KrisCarnmarker said:

John, thanks for that explanation. I may be incorrect, but you are basing your arguments on noise levels, are you not?

Click to expand...

Yes, I am.

Noise levels are obviously not irrelevant, but to me at least, the purpose of ETTR is about tonal gradations more than about noise levels. These two are practically related of course.

Click to expand...

Yes, they are, but they are related in such a way that the noise renders the "number of levels" to a level of irrelevancy in actual practice. Noise and the spectral responses of the three color channels, along with clipping, are the only things that really affect image quality at the RAW level, in practical reality. The noise levels would have to be significantly lower than they currently are for the number of tonal levels to make a significant difference in image quality. In a DSLR at ISO 100, even an out-of-focus highlight area of flat illumination will have neighboring pixels in the same color channel differing by scores of RAW levels, due mainly to photon noise statistics. In the deep shadows, they are varying by several levels from both photon noise and read noise at ISO 100, and by scores for some cameras, at ISO 1600. Mild quantization is a relatively minor issue in this context.

If we use the LL article examples, a camera capable of 5 stops DR, with a 12-bit ADC provides 4096 distinct values. 2048 of those are used in the first (brightest) f-stop, 1024 for the next, etc.. Practically, this means I get better tonal gradations at the higher f-stops, which is what we want to exploit.

Click to expand...

I would call those "stops", not "f-stops", as they have nothing to do with ratios of focal length to aperture.

There is nothing there to exploit with most cameras. Exposing to the right works, but not for the main reason Michael Reichmann gives; it's just a coincidence that using more levels in a given ISO results in a higher signal-to-noise ratio, which *is* significant.

So, if I go to e.g. ISO 400, do I no longer have those 2048 levels available in the first f-stop any more?

Click to expand...

Most cameras use 12 bits of RAW data up to at least ISO 800. Not all cameras use all 4096 levels though. My Canon 30D only uses 3040 levels in some of its ISOs (160/320/640/1250), and those give better IQ than their neighboring lower ISOs (125/250/500/1000), which use 3967 levels of RAW data.

I'd include a demonstration comparing RAW data that uses various amounts of levels, and various S/N ratios, to show how levels mean almost nothing in practice, but that will take some work, so I will do that if I get time. Nikon realizes they mean almost nothing, and that is why they use lossy compression in their RAWs. You would literally have to start adding many images together, to reduce the noise to very low levels, before the number of tonal levels starts to come into play in the RAW data. Now conversions may be another story. Converters may posterize images more than necessary.

This is what I don't understand. Of course, I'm simplifying things, and it may be that although I do still have those 2048 levels, other parameters have worsened and whatever gains I have achieved by ETTR at ISO 400 have been made negligible by the other factors.

Click to expand...

For most cameras, ETTR at a higher ISO will give equal or better RAW data, so long as you don't reduce the absolute exposure to do so. IOW, if you you put a camera in manual mode, meter for ISO 100 and set the Av and Tv values accordingly, and then shot the same scene at every ISO, the one that had the highest ISO which didn't cause any unwanted clipping of highlights will give anywhere from slightly to significantly better results than the lower ISOs.

John Sheehy said:

IOW, if you you put a camera in manual mode, meter for ISO 100 and set the Av and Tv values accordingly, and then shot the same scene at every ISO, the one that had the highest ISO which didn't cause any unwanted clipping of highlights will give anywhere from slightly to significantly better results than the lower ISOs.

Click to expand...

That makes a lot of sense for me. With the fixed Ev (I mean real Ev) (I always mean real Ev) (and of course a certain scene, with a certain distribution of luminance), the higher ISO settings move the saturation photometric exposure (H) down, until it approaches the highest H in the image. (This actually works separately for each of the three CFA groups, of course, but that doesn't spoil the concept - I suppose that in most cases, the green collision happens first.) At this point, the best use is made of the range of code values.

So, not surprisingly, more generally, to attain this situation ("exposed fully to the right"), for a given scene, we can play with shutter speed, aperture, and/or sensitivity (ISO speed) setting, as best suits all our objectives, and it doesn't really matter which (until we're in a situation in which noise looms into importance in the shadow areas).

As one of our friends, a character actor who's busy now as a corny cowboy in an auto dealership commercial, says, "We don't care how yew git here - just git here!"

Thanks for helping me get this picture.

Regarding my own test data, now that Bart has reminded me not to forget to cork my eyepiece for metered exposures, I have started over. Results presently!

Best regards,

Doug

John Sheehy said:

YIOW, if you you put a camera in manual mode, meter for ISO 100 and set the Av and Tv values accordingly, and then shot the same scene at every ISO, the one that had the highest ISO which didn't cause any unwanted clipping of highlights will give anywhere from slightly to significantly better results than the lower ISOs.

Click to expand...

Interesting. I think most people would be surprised to find this out. Most people, me included, will avoid higher ISOs if at all possible.

Doug_Kerr said:

So, not surprisingly, more generally, to attain this situation ("exposed fully to the right"), for a given scene, we can play with shutter speed, aperture, and/or sensitivity (ISO speed) setting, as best suits all our objectives, and it doesn't really matter which (until we're in a situation in which noise looms into importance in the shadow areas).

Click to expand...

Yes, this is what I would have thought "intuitively". That the sensitivity was excluded from my choices was what was bothering me. Now that it is settled (in my mind) that this is not so, I can breathe again

Thank you all for taking the time to explain all this to me!

/Kris

KrisCarnmarker said:

Interesting. I think most people would be surprised to find this out. Most people, me included, will avoid higher ISOs if at all possible.

Click to expand...

Note the qualification: "so long as you don't reduce the absolute exposure to do so.".

It's not too surprising but also not immediately obvious.

- DL

Don Lashier said:

Note the qualification: "so long as you don't reduce the absolute exposure to do so.".

Click to expand...

Yes, that is a given.

Don Lashier said:

It's not too surprising but also not immediately obvious.

Click to expand...

Well, tell that to MR, the author of the article I was referring to It contradicts his statement.

KrisCarnmarker said:

Interesting. I think most people would be surprised to find this out. Most people, me included, will avoid higher ISOs if at all possible.

Click to expand...

Note that I am talking about a fixed absolute exposure (dictated by subject luminance and Av and Tv values); not about dialing the ISO and letting the metering do its ISO-related absolute exposure adjustments.