Jump to content

Expose to the right (ETTR) or underexpose?


Muizen

Recommended Posts

 

The traditional way would be to expose for a middle gray like a gray card, assuming that as long as you get the mid tones correct, highlights and shadows will care for themselves. With digital photography that isn’t true anymore; for high contrast situations this is a recipe for blowing out highlights. Thus the advice to expose for the highlights and leave it to the raw conversion process to care for the shadows and mid tones.

 

Please, let's stay away from semantics and buzz words.

Does anybody have an issue with Michael's description above within the context set by Jaap in post #41?

 

K-H.

Link to post
Share on other sites

There is some truth in Michael's words. For landscapes and seascapes in the late afternoon, I will maximise the exposure for the sky ensuring there are no blown highlights at ISO 160. There is often a lot of latitude in the shadows which can be drawn out. If there is any marginal highlight blown in the clouds, sky or reflections, I find it almost impossible to fix with RAW converters. (C1 Pro/CS5)

  • Like 2
Link to post
Share on other sites

The many reactions to my question to whether or not ETTR, has left me somewhat confused. Perhaps I am looking for a rule I could adopt, but such a rule is not possible because of the strongly varying light circumstances?

Some key opinions (although taken out of context)

- ETTR is recommended by Reichman

- ETTR should be avoided (Hans)

- Underexposing with half a stop yields perfect results (horoso)

- M9 does not tolerate ETTR better or worse than other cameras (denoir)

- ETTR mostly a waste of time (Sandy/Keithu)

- Meter normally for the M9 mostly all the time (= no ETTR) (dwbell)

- ETTR but constrained to not blown highlights ((k-hawinkler)

- For the M9 (ETTR) is not a good idea (denoir)

- Reichmans article makes perfect sense in situations where there is plenty of light..ETTR is mythical. M9 handles underexposure spectacularly well, despite the theory (Blakley)

 

As a result of this interesting discussion I tend to conclude that ETTR (and it goes without saying: without blowing highlights) is not better than exposing normally with perhaps a tendency to some underexposing. But that much depends on the available light.

Link to post
Share on other sites

Yes- but some kittens must be murdered:eek::rolleyes:

 

One of the problems in this discussion is that the word highlight covers two things.

1. it is meant as the lightest part of the image we would wish to show some detail.

2. it is meant as specular highlight which is supposed to show no detail.

 

If we interpret ETTR (please, no exact definitions and theoretical nit-picking ;), I know the whole story...)

as the exposure which just preserves highlights #1 and pushes highlights #2 off the histogram to the right we will have pulled as much of the shadows above the noise floor and have the fullest contrast range as possible in our file. The histogram will show this.

If the essence of the photograph is to get all detail in the shadows, highlights #1 will sadly have to go as well, only to be saved by bracketing and HDR techniques.:o

Edited by jaapv
  • Like 4
Link to post
Share on other sites

Advertisement (gone after registration)

I think the ETTR vs. underexposure issue is a red herring. ETTR can be equivalent to either overexposure or underexposure, depending on the circumstances.

 

If the scene contrast is low and smaller than the camera’s dynamic range, exposure is unproblematic. Still one can make a case for ETTR so that all the tonal values one cares for sit flush with the right hand side of the histogram which by traditional standards amounts to overexposure. This results in a kind of high-key image which may not be what one was after, but you can pull down the shadows and maybe the mid tones to get more contrast. This will do nothing for dynamic range – as there was never a need to –; the effect is reduced shadow noise.

 

If, on the other hand, the contrast is taxing the camera’s dynamic range, ETTR underexposes to ensure that highlights (not specular highlights) won’t be blown. Shadows will be squished and some shadow detail may be lost in the noise, but all in all one stands a much better chance of recovering shadow detail from noise than highlights gone for good because of clipping. Thanks to ETTR, useful dynamic range is optimized, at the expense of some added noise.

  • Like 5
Link to post
Share on other sites

We should, of course, not ignore the relative impact and importance of this technique.

 

To paraphrase: "An ETTR exposure of a fuzzy concept is still a sh1t image!"

 

Further, these small qualitative gains can probably not be seen on anything smaller than A2 (ish ?) sized prints.

 

I do it on the 5DII when required / possible because bringing up low tone detail really can get messy down there. The M9 is much cleaner in the murky areas, in my experience. With the 5D I was often luminance mask noise reducing and so on, with the M9 I just whap the fill slider up and I'm done! :D

 

I only say this to those who are wondering if it's worth it to bother (learning and understanding the principle and methods is always worth it) using these techniques in practice and my honest answer is "probably not".

Link to post
Share on other sites

Your graphs prove the benefit of ETTR - however your implementation of ETTR is at fault.

 

The ideal method is not to increase ISO and over expose, this just causes a 'no gain' effect. The best method of ETTR is to 'over-expose' at a base ISO.

 

For example using your chart, an image covering luminance values of 5 through 8 will have a lower S/N than an image exposed at levels 2 through 4.

 

Of course that is a 'low contrast' scene. An subject with 11 stops will not benefit from any shift to the left, in fact ETTR would under-expose this compared to straight metering - because ETTR should be stated as "shift it to the right without clipping important highlights."

 

Pumping up the ISO is not proper ETTR.

 

 

 

With a mathematical model of the M9, much can be taken out of the guesswork.

 

The accurate model that I have, contains all existing kinds of noise, Photon Noise, Sensor noise, A/D Noise, Photo response non uniform etc. giving a reliable picture of the Signal to Noise of the complete camera.

I have calculated the noise of the two combinations ISO 160/320 and ISO 800/1600, in both cases the lower ISO value with “normal” exposure and the higher ISO value with +1 stop ETTR.

First the graphical representation of both combinations, where yellow is the higher and magenta is the lower ISO value of the two.

 

[ATTACH]273092[/ATTACH]

 

 

According to Sandy, there is a negative effect of ETTR on color reproduction. ETTR should therefore be avoided.

 

Hans

 

The issue of color reproduction is worth looking at. Coming from the 5dII, moderate levels of ETTR produce better images - with the M9 I will need to try comparison shots and bracket landscapes.

 

From your charts - it shows that ETTR should be avoided if hand-held speeds require an increase of ISO.

Link to post
Share on other sites

One of the problems in this discussion is that the word highlight covers two things.

1. it is meant as the lightest part of the image we would wish to show some detail.

2. it is meant as specular highlight which is supposed to show no detail.

I'll go with 1, jaapv. Surely 2. has to be irrelevant?

Link to post
Share on other sites

I'll go with 1, jaapv. Surely 2. has to be irrelevant?

Yes it is - until you read the forums - and it was much worse before with the M8 and older PP sofware- with posters trying to save street lights, using "A" in high-contrast night shots, etc and then complaining about poor noise performance.:rolleyes:

Link to post
Share on other sites

The whole issue in my posting is, whenever you feel the need to do + 1 stop ETTR at a certain ISO, you can just as well go one stop down in ISO without ETTR.

By doing that, you loose nothing in terms of S/N and you prevent overexposure or color shifts.

 

Hans

Link to post
Share on other sites

First time posting on this forum, looks like there is quite a lot going on with ETTR and exposure.

 

Hans, your data is showing that the M9 is shot noise limited (i.e. the noise is pretty much caused by the arrival rate of the photons itself, and the sensor circuitry adds very little noise if any, at least for a good part of the ISO / gain range). All good cameras behave in a narrow band when it comes to this..

 

From a sensor output stanpoint, ETTR (if this not explained somewhere else already):

 

Take for instance, your favorite M9,

 

1. The ADC (analog-digital converter - the block which is responsible for digitizing pixel output) output is 14 bit for the M9, which means this block is capable of breaking down a pixel output into 16384 levels (or tones), and assume that the ADC assigns these codes in a monotonic fashion with 0 for the purest black and 16383 for the purest white.

 

So, purest black: 00 0000 0000 0000 ( 'd0: short hand for decimal 0),

Purest white: 11 1111 1111 1111 ('d16383).

In practice, the purest black and purest white are not exactly at the ends of this range, but somewhat lower than the ends of this range.

 

Now take a scene where all the tones in the histogram are uniformly distributed (i.e. all these 16384 levels occur at least once). If you underexpose this scene, what happens is that some of the levels between 8191 and 16383 (I used half of the full range as an example) get incorrectly shifted towards the lower half of the full range. With the severity of underexposure, more of the upper half of the tonal range gets shifted. However, the problem arises when this "RAW" data from the sensor needs to be gamma mapped (i.e the sensor response needs to be mapped to what a human eye sees). Now if one were to map an underexposed scene (generally involves stretching both the ends of the captured histogram), RAW converters / camera processors have limited information (especially at the high end), and this information needs to be mapped to the higher end of the histogram. ie we are asking to create information out of thin air. The histogram can be shifted, but it is like adding more light to a certain part of the histogram. If one is not careful, subtle shifts in tonality are lost. RAW converters / onboard processing has come a long way in this regards. And the reverse is not the case with an over exposed histogram, because it involves redistributing already existing data to lower values, and setting the corresponding black level. Also, one doesn't want to lift the shadows too much because of noise concerns (usually at low gain settings, sensor circuitry tends to be noisier, and hence lower S/N ratio).

 

Also, some sensors try to allocate extra information at the ends of the tonal range.

 

From a sensor perspective, does ETTR help? Yes and no.

 

It largely depends on the scene histogram distribution, photographic goals and noise / tonality concerns. If a scene has large shadow areas, one might be better off exploiting this technique (without blowing out the highlights). If the scene is uniformly distributed, then there is very little to achieve by ETTR.

 

Through this simple shifting of bits around, color tonality gets affected. Human eyes are exceptional at perceiving shifts in color tonality than what the current state of art cameras can achieve.

 

Also, RAW developers are starting to develop floating point arithmetic for RAW analysis. This should help with some of these issues.

 

There is no free lunch. :)

 

 

With a mathematical model of the M9, much can be taken out of the guesswork.

 

The accurate model that I have, contains all existing kinds of noise, Photon Noise, Sensor noise, A/D Noise, Photo response non uniform etc. giving a reliable picture of the Signal to Noise of the complete camera.

I have calculated the noise of the two combinations ISO 160/320 and ISO 800/1600, in both cases the lower ISO value with “normal” exposure and the higher ISO value with +1 stop ETTR.

First the graphical representation of both combinations, where yellow is the higher and magenta is the lower ISO value of the two.

 

On the horizontal axis, 13 is the highest or lightest level, just before overexposure.

Going to the left is decreasing the luminance.

On the vertical axis is the Signal to Noise in stops.

Zero means that the signal is just as strong as the noise, and 6 means that the signal is 2exp6 = 64 times stronger as the noise.

 

[ATTACH]273092[/ATTACH]

 

The graphs are roughly showing that the S/N is the same for all luminance values when the higher ISO value is exposed with +1 stop ETTR.

For instance the yellow dot at Luminance 7 has the same S/N as the Magenta dot at Luminance 6, and so on.

But the graphs are not accurate enough to show the exact difference.

So here are the S/N values at the Luminance levels 12 –7 for ISO 160, compared to S/N at the Luminance levels 13-8 ( +1 stop ETTR ) for ISO 320.

 

[ATTACH]273094[/ATTACH]

 

At lower Luminance levels (darker area’s) there is an absolute insignificant S/N advantage of 0,05 stops in favor of ISO 320 with +1 stop ETTR. At the higher Luminance levels (lighter area’s) there is no difference at all.

 

Now for ISO 800 and ISO 1600.

 

[ATTACH]273095[/ATTACH]

 

Although the S/N in absolute value is worse than at ISO 160/320, the relative value between S/N at ISO 800 is exactly the same as the S/N value at ISO 1600 with 1 stop ETTR. There is simply no difference.

 

When just looking at noise, the conclusion is:

Instead of shifting the exposure +1 stop ETTR, one can just as well go one Stop down in ISO without overexposure. This is true for low as well as high ISO values, and it is true over the whole range from light to dark.

 

According to Sandy, there is a negative effect of ETTR on color reproduction. ETTR should therefore be avoided.

 

Hans

  • Like 3
  • Thanks 1
Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...