Ad
Ad
Ad
Pages: [1] 2 3 ... 12 »   Bottom of Page
Print
Author Topic: Expose to right, it is as simple as  (Read 35159 times)
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« on: July 31, 2011, 03:16:53 PM »
ReplyReply

Hi,

Agree with the conclusion, we should to expose to the right and the cameras should support this.

On the other hand Id say that Michaels explanation of why we should expose to the right is lacking. The reason to expose to the right is mainly that we want maximize signal (which is the number of photons detected). There is no simpler explanation, and it also happens to be the correct one.

Best regards
Erik
Logged

michael
Administrator
Sr. Member
*****
Offline Offline

Posts: 4895



« Reply #1 on: July 31, 2011, 03:19:40 PM »
ReplyReply

Eric,

Saying "maximize signal" isn't terribly useful unless you also explain why. Not everyone (in fact not that many) has the technical understanding needed for that simple phrase to make sense.

Which is why I wrote the article.

Michael
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #2 on: July 31, 2011, 03:33:36 PM »
ReplyReply

ETTR is right and good. The gotcha is the differential clipping of highlights as mentioned, which does make it a touch tricky, but certainly computable in a live-view situation. Although lack of code values in deepest shadows is an issue, the major problem is sensor read noise, and the brighter the signal, the less contribution read noise makes to the image. That exactly fits in with Erik's explanation.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« Reply #3 on: July 31, 2011, 04:14:55 PM »
ReplyReply

Michael,

The real issue I have with your article that you imply that noise would depend on the number of tonal steps in the dark areas, this is however not generally the case. Noise is coming from different factors, the obvious ones being shot noise (the natural variation of photons) and readout noise. Shot noise is as pointed out in your article proportional to the square root of the captured photons, while readout noise is fixed.

In general we talk about signal, which is essentially the light we detect, and noise which is the unwanted variation in sensor signal. We normally want to maximize signal/noise. If we look at shot noise we know that noise is the square root of the signal, so SNR (Signal Noise Ratio) is also the square root of the signal.

Let's assume that a sensor cell detects 10000 photons. The square root of 10000 is 100 so our SNR would be 100. Would we expose two stops less, the number of photons would be 2500 and SNR would be 50 (which is still very good).

If we assume that darks are three stops below midtones, and that midtones are at 10000 photons when correctly exposed to the right, SNR would be like sqrt(10000/8). With two stops less exposure we would have  sqrt(2500/8) = 17.6.

When we reduce exposure further we need also to take readout noise into account which can be something like 10 electrons. This would add up with the shot noise. This addition would be in quadrature so for two step underexposed (relative to ETTR) and three stops under midtones we would have:

Shot noise = 17.6
Readout noise = 10

Noise = sqrt(17.6^2 + 10^2) -> 20.3

Now, a 12 bit converter would see 4096 different values, if we assume that saturation is about  50000 photons the lowest bit would correspond to about 12 photons, while the variation on the number of photons would be around 20. This pretty much also illustrates that there is little practical value in 16 bit converters. Let's assume that we have a new device from Phase Zero, totally devoid of readout noise and having pixels holding maximum 50000 electrons. If we assume 14 bit conversion we would have 16384 steps. Each step would correspond to 3 electrons. With no readout noise SNR would be 1.7, except that Poisson statistics would not be valid for three electrons. There would be little difference in doing 16 bit conversion or 14 bit conversion, and multiplying the signal by four and adding 2 random bits!

Best regards
Erik


Eric,

Saying "maximize signal" isn't terribly useful unless you also explain why. Not everyone (in fact not that many) has the technical understanding needed for that simple phrase to make sense.

Which is why I wrote the article.

Michael

Logged

Schewe
Sr. Member
****
Offline Offline

Posts: 5500


WWW
« Reply #4 on: July 31, 2011, 04:21:01 PM »
ReplyReply

The reason to expose to the right is mainly that we want maximize signal (which is the number of photons detected).

More photons = better signal...underexposing to "preserve" highlight detail is a bad practice as there is a ton of info in the brightest stop if you know how to tease the data out. Clearly, when shooting a high dynamic range scene, you must take care to avoid clipping important textural detail. On the other hand the tools we currently have (flashy highlight warnings suck) really don address the issue.

The bottom line I would say is to take a neg shooter's point of view; shoot for the shadows, develop for the highlights. With so much data in those brightest areas, it really ain't that hard. See ETTR and look at the Niagara Falls image to see just how much detail there is in the uper portion of the capture/
Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5500


WWW
« Reply #5 on: July 31, 2011, 04:26:37 PM »
ReplyReply

In general we talk about signal, which is essentially the light we detect, and noise which is the unwanted variation in sensor signal. We normally want to maximize signal/noise. If we look at shot noise we know that noise is the square root of the signal, so SNR (Signal Noise Ratio) is also the square root of the signal.

Erik,

You might wanna dial down the math just a bit...(pretty sure Mike doesn't really care about that stuff). The main take away from Mike's article is that if you are shooting a low contrast scene, using your exposure meter and centering the image data in the center of the histo is a suboptimal idea...move the exposure to the right (to capture more photons and hence get a better capture). It's better to tone your curve down (darken) than it is to adjust digital captures to be lighter. Agreed? More photons=less noise?
Logged
cybis
Full Member
***
Offline Offline

Posts: 164



WWW
« Reply #6 on: July 31, 2011, 04:38:32 PM »
ReplyReply

What if you are shooting handheld in low light, does ETTR still make sense? Does ETTR only make sense at the lowest possible ISO setting? Should one trade ISO for speed in order to ETTR when conditions require it?

I regularly shoot handheld very low dynamic range scene in low light where all the data falls within one stop.  There are now essentially 4 variables in play to achieve an acceptable exposure: timing, aperture, ISO, and 'amount of ETTR'.

For instance, here are few possible exposure choices at a given aperture:

ISO 400 1/200 no-ETTR
ISO 1600 1/200 max-ETTR
ISO 400 1/50 max-ETTR
ISO 100 1/50 no-ETTR

Which is best?

Logged

ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« Reply #7 on: July 31, 2011, 04:41:24 PM »
ReplyReply

Jeff,

Sorry for the math...

Absolutely agreed, more photons = less visible noise! Incidentally, this is also one of the reasons that larger sensors yield better image quality. A larger sensor collect more photons!

Best regards
Erik



Erik,

You might wanna dial down the math just a bit...(pretty sure Mike doesn't really care about that stuff). The main take away from Mike's article is that if you are shooting a low contrast scene, using your exposure meter and centering the image data in the center of the histo is a suboptimal idea...move the exposure to the right (to capture more photons and hence get a better capture). It's better to tone your curve down (darken) than it is to adjust digital captures to be lighter. Agreed? More photons=less noise?
Logged

ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« Reply #8 on: July 31, 2011, 04:48:53 PM »
ReplyReply

Hi,

This may depend a bit on what you are shooting! Canon's and some Nikons have different characteristics from newer Nikons using Sony based sensors. With Canon and Nikon D3S and D700 you would probably increase ISO to reduce noise (perhaps up to 1000-1600). On recent Sony sensors it may have less significance, Nikon D3X, D7000. Pentax K5 and Sony Alpha 580 seem to be "ISO-less" cameras. You always want to collect as many photons as possible.

So advice is: On Canon increase ISO and expose to the right. On cameras with Sony sensors it may matter little.

Best regards
Erik


What if you are shooting handheld in low light, does ETTR still make sense? Does ETTR only make sense at the lowest possible ISO setting? Should one trade ISO for speed in order to ETTR when conditions require it?

I regularly shoot handheld very low dynamic range scene in low light where all the data falls within one stop.  There are now essentially 4 variables in play to achieve an acceptable exposure: timing, aperture, ISO, and 'amount of ETTR'.

For instance, here are few possible exposure choices at a given aperture:

ISO 400 1/200 no-ETTR
ISO 1600 1/200 max-ETTR
ISO 400 1/50 max-ETTR
ISO 100 1/50 no-ETTR

Which is best?


Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #9 on: July 31, 2011, 05:09:26 PM »
ReplyReply

Hi,

Agree with the conclusion, we should to expose to the right and the cameras should support this.

On the other hand Id say that Michaels explanation of why we should expose to the right is lacking. The reason to expose to the right is mainly that we want maximize signal (which is the number of photons detected). There is no simpler explanation, and it also happens to be the correct one.

Best regards
Erik

Eric,

You are exactly correct! The brightest f/stop of a digital exposure on current 12 bit sensors has nowhere near 2048 levels because of the presence of noise. Most of those 2048 levels are wasted in encoding noise. Furthermore, because of the Weber-Fechner law (see Norman Koren's Human Vision and Tonal Levels) the eye can appreciate only about 70 levels in the brighter f/stops and even fewer in the darker ones. However, extra levels are useful if you are doing an extreme amount of editing of the tone curve.

As you correctly state, we expose right to get a higher signal to noise value as explained in depth by Emil Martinec, who expressly discusses the fallacy of Micheal's original ETTR article. These concepts have been discussed at length on LULA and the Adobe forums and even Jeff Schewe has abandoned the levels argument and I was very discouraged to see Michael sticking to his original reasoning. The main improvement with ETTR is in the shadows, not the highlights.

As Emil discusses in his article, the Nikon compressed NEF format is visually lossless because it throws away redundant information in the brighter f/stops.

Regards,

Bill
Logged
kwalsh
Jr. Member
**
Offline Offline

Posts: 89


« Reply #10 on: July 31, 2011, 05:15:46 PM »
ReplyReply

There is someone besides the camera makers that need to get on board with this, the RAW converter makers.

ACR/LR causes ETTR problems with some camera profiles.  Unfortunately ETTR and then moving the exposure control doesn't actually work with all camera profiles because some place a color "twist" prior to the exposure compensation.  As a result you *will* get color errors when doing ETTR with such a profile.  If, on the other hand, the twist is applied post EC then everything is OK.

There doesn't seem to be a lot of rhyme or reason as to which camera profiles do it which way and from the discussions I've seen on the Adobe forums so far the response from Adobe has a been a bit obtuse on the issue.

At least with ACR/LR the problem is resolvable if you create your own profile with no pre-EC twists.  I don't know about other converters...

Ken

P.S. Nice article.  I completely agree we need some more intelligent digital exposure tools in our cameras.  True RAW histograms desired as well!  Thanks for revisiting an interesting and important topic.

P.P.S. As a signal processing engineer I find the "levels" rationalization at the start of the article a red herring.  Maybe it is an analogy that is helpful for some people to be motivated to ETTR, but it is in fact not at all relevant to ETTR.  The noise discussions later in the article are what are relevant to ETTR.  There are no cameras on the market in which quantization levels play any role in image noise, read noise and shot noise dominate in all of them.
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #11 on: July 31, 2011, 05:27:55 PM »
ReplyReply

There is someone besides the camera makers that need to get on board with this, the RAW converter makers.

ACR/LR causes ETTR problems with some camera profiles.  Unfortunately ETTR and then moving the exposure control doesn't actually work with all camera profiles because some place a color "twist" prior to the exposure compensation.  As a result you *will* get color errors when doing ETTR with such a profile.  If, on the other hand, the twist is applied post EC then everything is OK.

There doesn't seem to be a lot of rhyme or reason as to which camera profiles do it which way and from the discussions I've seen on the Adobe forums so far the response from Adobe has a been a bit obtuse on the issue.

At least with ACR/LR the problem is resolvable if you create your own profile with no pre-EC twists.  I don't know about other converters...

Ken

P.S. Nice article.  I completely agree we need some more intelligent digital exposure tools in our cameras.  True RAW histograms desired as well!  Thanks for revisiting an interesting and important topic.

P.P.S. As a signal processing engineer I find the "levels" rationalization at the start of the article a red herring.  Maybe it is an analogy that is helpful for some people to be motivated to ETTR, but it is in fact not at all relevant to ETTR.  The noise discussions later in the article are what are relevant to ETTR.  There are no cameras on the market in which quantization levels play any role in image noise, read noise and shot noise dominate in all of them.


Ken,

A great post! You may be a newbe to this forum, but obviously not to the realm of digital processing. Your point concerning hue twists is an important one and it is my understanding that with ACR the twists are often applied after exposure, or else no one would be reporting serious problems with them.

Regards,

Bill
Logged
dreed
Sr. Member
****
Offline Offline

Posts: 1260


« Reply #12 on: July 31, 2011, 05:33:03 PM »
ReplyReply

If ETTR were to be the way in which your camera took a picture, what do you want to see when reviewing on the LCD? The picture as taken or the picture as taken with exposure correction applied?

Next, which one of those do you put in the thumbnail of the raw file?

And what if you want different output for each?

For example, when I'm shooting, I want to see if I've blown highlights (and where they are) when I review on an LCD but if I'm browsing through pictures with Windows Explorer, having thumbnail pictures that are all ETTR'd is not terribly useful. It also doesn't play well with software such as DLNA servers that will "serve" a raw picture file by extracting the JPEG.

I don't believe that ETTR will ever default to "ON" in any consumer grade camera because either the shutter speed is going to be slower (greater chance of blurry pic) or the ISO increased to keep the shutter speed down and both will negatively impact the picture quality of the average photographer that doesn't whip out their tripod for each and every shot.
« Last Edit: July 31, 2011, 05:35:53 PM by dreed » Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« Reply #13 on: July 31, 2011, 06:06:50 PM »
ReplyReply

Hi,

I have the impression that Michael suggested a setting for ETTR. Would we have a setting for ETTR we would not really need to care about blown out highlights, because the camera would never overexpose non specular highlights.

There are some practical issues with automating ETTR exposure. To start with I would like to have histogram calculated from RAW pre color balance.

Best regards
Erik

If ETTR were to be the way in which your camera took a picture, what do you want to see when reviewing on the LCD? The picture as taken or the picture as taken with exposure correction applied?

Next, which one of those do you put in the thumbnail of the raw file?

And what if you want different output for each?

For example, when I'm shooting, I want to see if I've blown highlights (and where they are) when I review on an LCD but if I'm browsing through pictures with Windows Explorer, having thumbnail pictures that are all ETTR'd is not terribly useful. It also doesn't play well with software such as DLNA servers that will "serve" a raw picture file by extracting the JPEG.

I don't believe that ETTR will ever default to "ON" in any consumer grade camera because either the shutter speed is going to be slower (greater chance of blurry pic) or the ISO increased to keep the shutter speed down and both will negatively impact the picture quality of the average photographer that doesn't whip out their tripod for each and every shot.
Logged

dreed
Sr. Member
****
Offline Offline

Posts: 1260


« Reply #14 on: July 31, 2011, 06:13:05 PM »
ReplyReply

There are some practical issues with automating ETTR exposure. To start with I would like to have histogram calculated from RAW pre color balance.

Is there a reason why we can't have that now?

What I mean to say is that it is not very easy to map the histogram to what you see very easily by looking at the histogram for each of the RGB colour channels so why pretend that we can look at it and map it (in our heads) to what we see in the picture?
Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7655


WWW
« Reply #15 on: July 31, 2011, 06:16:53 PM »
ReplyReply

Hi!

I don't see why we won't have raw based histograms. As far as I can recall the Leica S2 has raw based histograms but most other cameras have histograms calculated from JPEG. >Don't know why!

Best regards
Erik


Is there a reason why we can't have that now?

What I mean to say is that it is not very easy to map the histogram to what you see very easily by looking at the histogram for each of the RGB colour channels so why pretend that we can look at it and map it (in our heads) to what we see in the picture?
Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #16 on: July 31, 2011, 06:45:38 PM »
ReplyReply

See ETTR and look at the Niagara Falls image to see just how much detail there is in the uper portion of the capture/

Jeff,

The amount of recovery in your Niagra Falls shot is astounding, but the raw file is not nearly so clipped as the ACR histogram would suggest, likely because of the BaselineOffset that ACR uses for your camera. Since I have your latest ACR book, I was able to download the raw file. The ACR histogram with camera default settings is shown below along with the raw histogram as shown by Rawnalyze. Before white balance, the red channel is entirely intact and the green and blue channels, while clipped, contain considerable data. I'm not so sure that the number of levels in the brightest stop has much to do with the recovery.

Regards,

Bill

Logged
michael
Administrator
Sr. Member
*****
Offline Offline

Posts: 4895



« Reply #17 on: July 31, 2011, 06:53:58 PM »
ReplyReply

It always amazes / annoys / amuses me when I publish an article that has been painstaking peer-reviewed by some of the brightest minds in the industry people who design sensors and write raw software (in this instances) and then "experts" whose credentials are unknown tell me (us) why the information in the article is wrong.

This was the case with my original ETTR article. I think I trust Thomas Knoll's knowledge of digital imaging (the original author of Photoshop and Camera Raw) over some online commentator.

Similarly in this instances. I'm sure that there are areas to quibble over (both scientists and artists love to quibble). But when it comes to arcania and subtleties in complex topics I'll continue to trust known experts whose bona fides are well established.

Michael
 
« Last Edit: July 31, 2011, 06:55:53 PM by michael » Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #18 on: July 31, 2011, 06:56:29 PM »
ReplyReply

I don't believe that ETTR will ever default to "ON" in any consumer grade camera because either the shutter speed is going to be slower (greater chance of blurry pic) or the ISO increased to keep the shutter speed down and both will negatively impact the picture quality of the average photographer that doesn't whip out their tripod for each and every shot.

With the latest cameras such as the Nikon D7000, it is the exposure on the image plane (lux seconds) that determines the image quality (SNR) and not the appearance of the histogram on the camera preview. Increasing the ISO will make the histogram appear more to the right and give a brighter preview, but increasing exposure in the raw converter will give the same results as a low ISO with a high shutter speed or large aperture, as long as the number of photons collected (the exposure) remains the same.

Regards,

Bill
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #19 on: July 31, 2011, 07:15:31 PM »
ReplyReply

It always amazes / annoys / amuses me when I publish an article that has been painstaking peer-reviewed by some of the brightest minds in the industry people who design sensors and write raw software (in this instances) and then "experts" whose credentials are unknown tell me (us) why the information in the article is wrong.

This was the case with my original ETTR article. I think I trust Thomas Knoll's knowledge of digital imaging (the original author of Photoshop and Camera Raw) over some online commentator.

I think that it was one of those rare occasions where Mr. Knoll misspoke when the made the comment that the brightest f/stop of a 12 bit capture contains 2048 discrete levels--he did not consider noise. Emil Martinec has a PhD in physics and is a full professor at the University of Chicago. His analysis is beyond refutation. Did you take the trouble to read it?

DXO gives the screen tonal range of the P65+ at 8.64 bits. That is the number of bits necessary to encode the number of discrete levels that the camera can detect. 2^8.64 = 399, which is far short of the theoretical number of levels in a 14 or 16 bit linearly encoded raw file.

In science, reason and data trump expert opinion. We no longer bleed patients for croup. That is how the "expert" physicians killed George Washington.

Regards,

Bill

Logged
Pages: [1] 2 3 ... 12 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad