Alley in Pisa, Italy

by Lars Kotthoff

submit your photo


Hall of Fame
View past winners from this year

Please participate in Meta
and help us grow.

Take the 2-minute tour ×
Photography Stack Exchange is a question and answer site for professional, enthusiast and amateur photographers. It's 100% free, no registration required.

Picking up from this answer and this question, What exactly is ETTR? How it may reduce the image noise? And how is it difference from film to digital sensors?

In the answer linked above, what are the 5 stops and is it related to ETTR?

In real life how can I apply this technique when I'm shooting?

share|improve this question
    
The question of the meaning of a stop in this context is answered under What is one “stop”? –  mattdm May 1 '12 at 22:14
    
@mattdm I understand what a stop means however the answer linked in the question mentioned "5 stop range", is this a standard range for tone brightness? –  akram May 2 '12 at 18:42
    
Oh, I see the confusion. That number comes from a quote from the Luminous Landscape ETTR article, and 5 stops was chosen as a reasonable number to represent the total dynamic range of a DSLR of the time the article was written. You can work the same calculation with any other arbitrary number for total stops. Five is just the example. –  mattdm May 2 '12 at 19:13
    
@mattdm oh okay that makes much more sense, thanks –  akram May 2 '12 at 19:29

4 Answers 4

up vote 25 down vote accepted

"Expose to the right" means record the brightest image you can and then reduce the brightness in post to achieve the desired level.

The word "right" comes from the histogram, where conventionally brightness increases left to right, thus increasing brightness shifts the whole histogram to the right.

ETTR helps reduce noise simply by capturing more light, which reduces photon noise, and gives a better signal to [electrical] noise ratio (by virtue of a bigger signal). The reason high ISO photos look noisy is due to low levels of light and amplifying a weak signal.

The technique works provided you don't increase the exposure to the point where it hits the maximum possible value and gets cut off, as this will result in a loss of information (known as clipping/blowing the highlights). Typically this is seen as an area of the image (usually sky) which has gone pure white.

In principle the technique works for film, certainly exposing the left and then having to push your image when printing will increase grain. However film has a different cutoff characteristic, as highlights gently roll off rather than hitting a hard limit.

Here's an experiment I did to demonstrate the effect (and rebuff a blog article which claimed ETTR didn't work):

Here's the camera metered exposure:

Here I've used ETTR and increased the camera meter's exposure by 1 stop using a longer exposure:

Finally, to show the difference here's the standard exposure with the ETTR image offset in the centre:

The reduction in noise is visible, particularly in the purple patch in the bottom left.

share|improve this answer
3  
+1, especially for providing a nice example and for stressing the issue with clipped highlights, an important practical consideration. –  mattdm May 1 '12 at 20:37

To be short ETTR is a smart usage of two fact:

  1. There is more information in the high light (the right of the level curve) than in the low light (the left of the level curve). This is due to the fact that capter has linear response to the light intensity while human perception is rather log (what you perceive as twice brighter is in fact not twice the amount of light but much more)

  2. The noise is present everywhere but what you perceive is the ratio noise over signal: if the signal is big you cannot see the noise, if the signal is of the same order or smaller than the noise you will see noise. So the more you collect light the bigger is your signal and the smaller is the noise perception

When overexposing your image (and in particular a globally dark image) you are using the right part of level curve for storing your image rather than the left one. Doing that you have two advantages (1) more information (more distinct tones) and (2) by collecting more light you increase the signal/noise ratio (so get less visible noise)

In post-treatment you can then correct your level and get the tone you want.

Back to film camera (I get the B&W picture which is equivalent to the color one but easier to figured out) each grain has a threshold (a number of photon) above which it will turn black and bellow which it will stay white (and be washed out in the film processing) the "noise" was the size of the grain which was related to the sensitivity.

share|improve this answer
    
+1 I liked "what you perceive as twice brighter is in fact not twice the amount of light but much more" –  akram May 2 '12 at 18:23
    
"more information" is slightly misleading. There are the same number of bits for the right half of the histogram as there are for the left half aren't there? –  Joe May 3 '12 at 6:59
    
@Joe you are right. However your perception act as "compressing" the right part and "inflating" the left part of the histogram, so there is more tones in the bright lights –  floqui May 3 '12 at 9:11

The answers you cite contain the information you want. It may not be "accessible" enough without reading and re and re-re reading. I'll try to summarise what was said in those references and in many other places, but do note that this is a summary and lots of detail are available elsewhere.

A digital camera sensor tends to produce an output that is linearly related to light level. this does not have to be the case, and here may be advantages in doing otherwise, but that's the norm so far.

With a linear sensor, if you halve the brightness you halve the numerical "reading" or light level. If the 'reading' is 4000 at 100% of sensor max light level capability, then it will be 2000 at 50% of sensor max level,
and it will be 1000 at 25% of max
500 at 12.5% of max
250 at 6.25% of max
125 at 3.125% OF MAX
62 AT ...

BUT each halving of light level is equivalent to one stop, or one EV level. It's far more intuitive to think in EV units but it can be equally expressed in stops.

So the first "stop" of sensor range has a certain EV of actual brightness at the top of this range and 1 EV less at the bottom, and the sensor has max reading of 4000 and minimum of 2000 and there are 2000 "counts" across this or EV level.
Areas in the image which are one EV level less bright than maximum brightness = the second stop / EV level in the image and have light levels from 1000 to 2000 and a 1000 range
The third stop has light levels from 500 to 1000 and a 500 range
The fourth stop has light levels from 250 to 500 and a 250 range

This means that the first stop of exposure has many numeric values between its top and bottom levels. Noise of a given magnitude that is a certain percentage of its range will be an increasing percentage of the range of a stop as light level falls. eg say noise was +/- 5 units relative to the sensors 4000:1 dynamic range.
In the top stop noise is 5/2000 = 1/400 = 0.25% of the range.
In the 2nd stop noise is 5/1000 = 0.5% .
By the time we are down to the 8th stop the dynamic range available
= 4000 /(2 x 2 x 2 x 2 x 2 x 2 x 2 x 2) ~+ 16 sensor steps, and the 5 units of noise are 5/16 or about 31% of the range. ie at the op end of brightness a given level of noise may have little effect but as brightness falls the noise double for every 1 stop decrease and the % that the noise is of signal variation doubles.

Translating this into practice - take a highish ISO photo where the image is starting to get noisy. Now look in the shadow areas - you will find that they are far more affected - in about inverse proportion to their brightness.

So - EV levels that are close to the top of the sensors maximum light handling level are less noise affected. It does not matter about what the light level is as long as it can be corrected in due course. Rather, we push all brightness levels up until the brightest level is almost clipping. This allows the lower levels to have as much sensor variation as possible.

Note that 5 stops was just a convenient range to consider - this effect of right shifting matters right across the range.

Film tends to have a logarithmic response to light so comoresses a wider variation of levels into a lower effective range.

share|improve this answer
    
I would compute sensor DR stops a bit differently. A/D converters are binary devices, and can only encode, at most, as much information as their bit depth. Since, in binary, every additional digit is a doubling of the numeric space of all previous digits, modern cameras are effectively limited to 14 stops, or 2^14 levels. Its extremely difficult in reality to actually achieve 14 stops of dynamic range, however, given the necessary overhead of converting an electronic charge into ADU (analog-to-digital units). Maximum saturation is usually less than 2^14, so real-world performance... –  jrista May 1 '12 at 20:58
    
...is usually limited to around 13 stops of dynamic range or less (assuming a very forgiving method of computing dynamic range...many would dispute even that much is actually possible, and offer that 10-11 stops is all we can really get in reality with more conservative methods.) The binary nature of an ADC also leads to every additional bit adding almost twice as many possible luminance levels as the previous, so a 15-bit sensor would offer about 32000 levels vs. the approximage 16000 of a 14-bit sensor. –  jrista May 1 '12 at 20:59
    
The dynamic range of the best modern camera systems slightly exceeds the number of bits in the ADC. This apparent impossibility is well cobered in prior stack exchange answer and relates in par the ability to "dither" an ADC output to beyond he number of bits provided if the signal and measurement systems are able to support such accuracy. Rushing out, else more ... –  Russell McMahon May 1 '12 at 23:21

There are those who think ETTR is folklore, not fact. Ctein (who has multiple decades of experience and is a master printmaker) has written that ti's all bull. (link: http://theonlinephotographer.typepad.com/the_online_photographer/2011/10/expose-to-the-right-is-a-bunch-of-bull.html) I'd suggest at least looking at his commentary.

Me? I respect Ctein a lot, but I tend to expose towards the right a bit (typically about 3/4 of a stop of compensation), depending on the subject. At its worst, ETTR seems to be placebo, not harmful. Whether it's really helpful? Not everyone agrees about it..

share|improve this answer
2  
Before getting too riled up by the inflammatory title of the linked article, note that this paragraph summarizes the key point: These days, noise is really not a big source of image quality loss[....] Cameras and sensors are so much better. Clipped highlights, as Mike and I discussed last week, haven't gone away. It's still a big issue when trying to get real quality in a digital photograph. The argument is that blown pixels are a bigger real-world problem than noise in most situations. –  mattdm May 1 '12 at 22:50

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.