Luminous Landscape Forum

Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: BernardLanguillier on December 08, 2005, 09:26:50 pm

Title: Revisiting the "expose to the right" dogma
Post by: BernardLanguillier on December 08, 2005, 09:26:50 pm
Most of us have by now accepted the idea that exposing to the right, just short of clipping, is the way to go with digital.

The basic reason being that, sensors being nearly linear devices, the brightest stop of a digital capture contains half of the bits available for sampling the tonal range that the sensor is able to capture. Keeping a gap opened between the right of out histogram and our brightest points results in keeping some of these precious bits un-used.

However, this dogma is based on an assumption that I feel is potentially unproven, and might also be largly un-tested.

This assumption is that a digital sensor works perfectly well up to the brightness levels resulting in 254, 254, 254 RGB values. In other words, we assume that we are not reducing the ability of the sensor to distinguish close tones by over-exposing the image as long as we stay short of pure white (255, 255, 255).

On the other hand, we also see in our digital images that the transition to fully blown highlights is typically not as smooth as it used to be with film, which hints at the fact that sensors are in fact not very good at dealing with these near white areas when the exposure reaches a high enough level.

Building on this, I am starting to wonder if, all things considered, there is actual value in over-exposing too much an image, even if it stays short of full blown highlights.

-> Aren't we better off being a bit more conservative?
-> Isn't this very much dependant on the actual behaviour of each sensor?

My view is that this is independant of the curve applied during RAW conversion, since this curve will never be able to get rid of posterizations that might have been introduced at capture by artifical over-exposure.

Am I missing something here?

Regards,
Bernard
Title: Revisiting the "expose to the right" dogma
Post by: Anon E. Mouse on December 08, 2005, 09:55:57 pm
You are applying a "rule of thumb" as some kind of "truth." Basically, you need to expose so that the tones you wish to capture in the scene are recorded. Since scenes and conditions vary so much, it is really hard to apply a "rule of thumb." They can be useful for beginners to get some ideas to improve their photography, but it will only go so far.

It seems the rule your are refering to is applying some fuzzy understanding to tone reproduction and implying data is like tones in negative film. You may have better luck with thinking digital is more like slide film where you need to place tones within the sensor response to give the correct relative tone. Post-processing should not be thought of as making the "correct exposure."
Title: Revisiting the "expose to the right" dogma
Post by: BernardLanguillier on December 08, 2005, 10:22:17 pm
Hi Anon,

If I read you correctly, you have not really followed the discussions on "exposure to the right", have you? If anything it does promote a slide like approach (avoiding blown highlights) but erring on the over-exposure side in order to maximize bit usage.

The very goal of my post is to discuss whether this "rule of thumb", considered by many here as the undisputable "truth", shouldn't indeed be challenged.

Regards,
Bernard
Title: Revisiting the "expose to the right" dogma
Post by: Goldilocks on December 08, 2005, 11:54:54 pm
Does the exposure to the right discussion concern only RAW shots or JPEGs as well. Could you please direct me to a place to understand this concept? I couldn't find it in a  search other than this "Replying to Revisiting the "expose to the right" dogma.
Thanks,
Linda (Goldilocks)
Title: Revisiting the "expose to the right" dogma
Post by: BernardLanguillier on December 09, 2005, 12:05:17 am
Linda,

This is the original article:

https://www.luminous-landscape.com/tutorial...ose-right.shtml (https://www.luminous-landscape.com/tutorials/expose-right.shtml)

Regards,
Bernard
Title: Revisiting the "expose to the right" dogma
Post by: kenstrain on December 09, 2005, 04:02:04 am
Quote
This assumption is that a digital sensor works perfectly well up to the brightness levels resulting in 254, 254, 254 RGB values. In other words, we assume that we are not reducing the ability of the sensor to distinguish close tones by over-exposing the image as long as we stay short of pure white (255, 255, 255).
[a href=\"index.php?act=findpost&pid=53090\"][{POST_SNAPBACK}][/a]

In terms of generic information the only published measurements that instantly spring to mind are those on clarkvision.com (in the SNR comparisons with film, curves are shown right up to white, and they do show some compression).

Possibly the answer is to do some tests with ones's own equipment, perhaps a progressively more overexposed colorchecker card?  

Ken
Title: Revisiting the "expose to the right" dogma
Post by: bobrobert on December 09, 2005, 05:47:11 am
Photographers are often advised to shoot so that details are retained in the highlights This means that some sort of underexposure is required In landscapes the sky is usually lighter than the foreground Surely if this happens then the histrogam will be to the left of the mid point meaning that there will be a gap on the right This to me appears contradictory Could someone enlighten me?
Title: Revisiting the "expose to the right" dogma
Post by: boku on December 09, 2005, 07:07:39 am
My assorted comments:

1) Expose to the right is much better applied to RAW files than JPEGs. In camera JPEG conversion algorithms intend to render a scene's tones accurately, "expose to the right" intends to optimize use of data.

2) I also take ETTR less as a "rule" as time goes on. It is a prudent concept to take advantage of a camera's capture capability, but you better know exactly how your camera is sensitized versus what it shows on the histogram. My Canon 5D is far different from my Canon 20D. I need to shy away from the right a bit on the 5D until I figure out what is happening.

3) When you expose to the right, you better remember what you pre-visuallized in the picture when you are sitting at the computer in front of the RAW converter. You need to know how to work with the conversion to bring back to your visualized rendition. Sometimes the default processing is significantly different and you need to remember what you were trying to achieve in the first place. This takes skill and insight.

4) I still think ETTR is a sound bit of advice, you just need to be careful as described. I would never use it for JPEGS, but then I don't shoot JPEGS.
Title: Revisiting the "expose to the right" dogma
Post by: DiaAzul on December 09, 2005, 09:07:36 am
There is an assumption that Sensor characteristics are linear with exposure which is not the case, particularly as the sensor approaches saturation. I think that the clarkvision website mentioned above does a good job of showing some of this non-linearity. Whilst the linearity of the sensor is pretty good for the first 80-90% of the exposure it does tend to exhibit a 'hockey stick' reponse as blow out is approached. This can result in unwanted colour shifts in highlights where the RAW convertor does not utilise a good quality profile - probably most noticable with a Canon 1-series camera in Photoshop CS (not CS2) where white highlights would skew to magenta just prior to blowing out.

Fortunately, most RAW convertors do seem to take this characteristic curve response into account during the conversion; however, it is probably still worth backing up the exposure a little if colour accuracy is required. Most importantly is to determine whether your chosen RAW convertor has a sufficiently accurate profile that maches your particular camera - though there is no evidence to suggest wide variations between cameras (of the same model) in terms of response curve.

Other than the above, in camera histograms are calculated on a subset of pixel data and as such may be inaccurate for gauging whether there are any blown out specular highlights - particular if the preview image is determined based upon an averaging type of interpolation algorithm.

I would tend to agree with Boku that the Histogram is more guide than fixed rule.
Title: Revisiting the "expose to the right" dogma
Post by: dazzajl on December 09, 2005, 09:48:02 am
Quote
This can result in unwanted colour shifts in highlights where the RAW convertor does not utilise a good quality profile - probably most noticable with a Canon 1-series camera in Photoshop CS (not CS2) where white highlights would skew to magenta just prior to blowing out.

Cheers for mentioning that bit. It explains exactly what I've got on a recent shoot when I was playing with how far to the right I could/should go to get a pure white background on some studio work.

This is with a 1D mkII and the Raw proccessor on PS CS verison. Just to clarify, are you saying this has been sorted with the CS2 version or just improved?

D  
Title: Revisiting the "expose to the right" dogma
Post by: DiaAzul on December 09, 2005, 09:51:14 am
Quote
This is with a 1D mkII and the Raw proccessor on PS CS verison. Just to clarify, are you saying this has been sorted with the CS2 version or just improved?


[a href=\"index.php?act=findpost&pid=53136\"][{POST_SNAPBACK}][/a]

It seems to have been sorted in CS2.
Title: Revisiting the "expose to the right" dogma
Post by: bob mccarthy on December 09, 2005, 09:53:43 am
One of the nice things about the pro cameras (D2x in my case) is separate histograms for RGB. I find luminance is not the best guide to setting the data to the right. Some histograms are green channel only (half of sensors).

I set the exposure so all channels are short of clipping.

At the same time, I look to see how much energy is at the clipping point. Some minor clipping is often warranted (specular highlights).

Bob
Title: Revisiting the "expose to the right" dogma
Post by: BJL on December 09, 2005, 01:27:27 pm
Quote
There is an assumption that Sensor characteristics are linear with exposure which is not the case, particularly as the sensor approaches saturation. ...

I would tend to agree with Boku that the Histogram is more guide than fixed rule.
[a href=\"index.php?act=findpost&pid=53135\"][{POST_SNAPBACK}][/a]
Thanks; indeed, electronic sensor also have a bit of a "shoulder", as discussed in the detailed spec. documents provided at the websites of sensor sellers like Kodak, Dalsa, Sony and FillFactory. Some camera makers might use the sensor only over its linear range, or put the right limit of the histogram at the end of that linear range, but that is not guaranteed.

In the spirit of AA, we probably still need to study the characteristics of our materials (now including sensors and related DSP) and learn how to expose them for best results, including highlight placement.
Title: Revisiting the "expose to the right" dogma
Post by: BernardLanguillier on December 09, 2005, 10:12:24 pm
Thank you for your feedback gentlemen, I gather that most of you have adopted an approach similar to mine.

It is conforting not to be alone, but I find it interesting that this key topic is not discussed more often.

Regards,
Bernard
Title: Revisiting the "expose to the right" dogma
Post by: Evan on December 10, 2005, 04:02:45 pm
The only thing I would add is that the histogram you see on the camera is based on the jpeg the camera would output with the current jpeg settings in the camera.  So if you're shooting raw and there is a spike at the highlights, there may still be room once you get it into your raw converter.  This takes practice to get a feel for what the histogram really means to the raw file.  I shoot a 10D and I don't know if this has changed with any of the newest high-end SLRs out there.  There was also a discussion about this on this site a while ago.
Title: Revisiting the "expose to the right" dogma
Post by: djgarcia on December 10, 2005, 07:59:22 pm
I pretty much follow ETTR as discussed. I would like to emphasize as has been mentioned, you need to understand the response characteristics of your equipment. This old axiom doesn't change for digital. Once you atune to your system's behavior repeatability becomes second-hand.

I shoot raw only, and I now use the camera's spot metering pretty much exclusively. Given my predominant subject of landscapes it works out. I find matrix / evaluative metering harder to second-guess in unusual lighting conditions, whereas the spot meter doesn't try to be smart so it's easy to predict when metering to some pre-visualization.

But in all I have found that when done properly ETTR'ing tends to keep the shadow detail away from posterization during post-processing.
Title: Revisiting the "expose to the right" dogma
Post by: devoman on December 11, 2005, 11:48:52 am
I just re-read the original article on ETTR:

https://www.luminous-landscape.com/tutorial...ose-right.shtml (https://www.luminous-landscape.com/tutorial...ose-right.shtml)

However, one paragraph strikes me as incorrect:

"This realization carries with it a number of important lessons, the most important of them being that if you do not use the right-hand fifth of the histogram for recording some of your image you are in fact wasting fully half of the available encoding levels of your camera."

It seems to me that the x-axis of the histogram is linear.  Therefore, a more accurate way to look at this is that the entire right half of the histogram represents the brightest F-stop (rather than the rightmost fifth).

Am I missing something?
Title: Revisiting the "expose to the right" dogma
Post by: Phuong on December 11, 2005, 12:39:56 pm
Quote
I just re-read the original article on ETTR:

https://www.luminous-landscape.com/tutorial...ose-right.shtml (https://www.luminous-landscape.com/tutorial...ose-right.shtml)

However, one paragraph strikes me as incorrect:

"This realization carries with it a number of important lessons, the most important of them being that if you do not use the right-hand fifth of the histogram for recording some of your image you are in fact wasting fully half of the available encoding levels of your camera."

It seems to me that the x-axis of the histogram is linear.  Therefore, a more accurate way to look at this is that the entire right half of the histogram represents the brightest F-stop (rather than the rightmost fifth).

Am I missing something?
[a href=\"index.php?act=findpost&pid=53232\"][{POST_SNAPBACK}][/a]

yes. you are thinking in decimal, while here you have to think in bits: the right most forth one has 2^10=1024 levels of brightness, while its next higher stop has 2^11 = (2^10)x2 = 1024x2 = 2048 levels of brightness which is twice as much as the forth one. therefore, the total levels of brightness of the whole 4 lower stops doesn't even equal to that of the fifth one. this is also the reason why you should be very careful when touching the shadow slider in raw converters since it will affect the dark tone very significantly. i'd rather do that later in PS (levels/curves) since in PS it's not linear anymore
Title: Revisiting the "expose to the right" dogma
Post by: devoman on December 11, 2005, 03:19:09 pm
With all due respect, I don't think you addressed my question.  Yes, I know that the brightest F-stop uses 2048 values (assuming 12 bits).

My point is that the scale (x-axis) on the histogram is linear.  Or at least I assume it is.  This means that if the rightmost endpoint represents the value 4095, then the midpoint represents the value 2048.   Thus the rightmost fifth of the histogram represents 819 values, not 2048.  Or in other words, the brightest F-stop uses 2048 values which represents the rightmost one-half of the histogram (not one-fifth).
Title: Revisiting the "expose to the right" dogma
Post by: Tim Gray on December 11, 2005, 04:55:47 pm
Quote
With all due respect, I don't think you addressed my question.  Yes, I know that the brightest F-stop uses 2048 values (assuming 12 bits).

My point is that the scale (x-axis) on the histogram is linear.  Or at least I assume it is.  This means that if the rightmost endpoint represents the value 4095, then the midpoint represents the value 2048.   Thus the rightmost fifth of the histogram represents 819 values, not 2048.  Or in other words, the brightest F-stop uses 2048 values which represents the rightmost one-half of the histogram (not one-fifth).
[a href=\"index.php?act=findpost&pid=53246\"][{POST_SNAPBACK}][/a]

Phuong has it right, you drop half in the first stop, not the right half of the histogram.  

"Or in other words, the birghtest F-stop used 2048 values (YES) which represents the righmost one-half of the histogram (NO)."  The right half of the histogram is more than one stop - it's one half of all the stops represented.
Title: Revisiting the "expose to the right" dogma
Post by: devoman on December 11, 2005, 05:01:29 pm
Quote
"Or in other words, the birghtest F-stop used 2048 values (YES) which represents the righmost one-half of the histogram (NO)."  The right half of the histogram is more than one stop - it's one half of all the stops represented.
[a href=\"index.php?act=findpost&pid=53251\"][{POST_SNAPBACK}][/a]

Thanks, but Just to be sure I understand this, you're saying that the X-axis of the histogram is NOT linear; it is logrithmic.  Is that correct?
Title: Revisiting the "expose to the right" dogma
Post by: boku on December 11, 2005, 05:03:14 pm
FACT: the X-axis of the histogram is NOT linear; it is logrithmic.
Title: Revisiting the "expose to the right" dogma
Post by: Gary Brown on December 11, 2005, 05:03:14 pm
Quote
My point is that the scale (x-axis) on the histogram is linear.  Or at least I assume it is.  [{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=53246\")

The values you see in the histogram (if my understanding is correct, anyway) have been converted from the linear raw values (where doubling or halving the value would be a one-stop difference) to the logarithmic RGB "pixel" values, for lack of a better term.

So, although the histogram x-axis is linear, in terms of RGB values, it's logarithmic, in terms of f-stops.

Somewhat related: Take a look at [a href=\"http://www.normankoren.com/digital_tonality.html#Human_vision]http://www.normankoren.com/digital_tonalit...ml#Human_vision[/url] and scroll down to the table showing "RAW levels in each zone" on the left, and "Gamma = 2.2 levels in each zone" on the right.
Title: Revisiting the "expose to the right" dogma
Post by: Jonathan Wienke on December 11, 2005, 09:42:54 pm
Quote
With all due respect, I don't think you addressed my question.  Yes, I know that the brightest F-stop uses 2048 values (assuming 12 bits).

My point is that the scale (x-axis) on the histogram is linear.  Or at least I assume it is.[a href=\"index.php?act=findpost&pid=53246\"][{POST_SNAPBACK}][/a]

You know what they say about assumptions...

The histogram is based on an in-camera JPEG conversion of the RAW data, which means that a tone response curve has been applied to the histogram data, and it no longer has a linear gamma. Each of the 5 segments of the displayed histogram represents 1-1.5 stops of exposure depending on the camera and camera settings.
Title: Revisiting the "expose to the right" dogma
Post by: BernardLanguillier on December 11, 2005, 11:45:29 pm
The ability to display a linear RGB histogram based on demosacied RAW data would be great to have, since the brightest stop would represent half of the histogram width for more accurate evaluation of the highlights situation.

Unfortunately, and for reasons that I have a hard time understanding, no camera manufacturer that I know of is currently proposing this.

Regards,
Bernard
Title: Revisiting the "expose to the right" dogma
Post by: devoman on December 12, 2005, 09:21:37 am
Thanks, all, for the clarifications and elaborations!
Title: Revisiting the "expose to the right" dogma
Post by: Jonathan Wienke on December 12, 2005, 01:37:50 pm
Quote
The ability to display a linear RGB histogram based on demosacied RAW data would be great to have, since the brightest stop would represent half of the histogram width for more accurate evaluation of the highlights situation.

Unfortunately, and for reasons that I have a hard time understanding, no camera manufacturer that I know of is currently proposing this.

There really isn't even any need to demosaic/interpolate a RAW histogram. I'd be ecstatic if a camera offered a 16-channel bar graph of the uninterpolated RAW data, where the rightmost bar represented clipped pixels and the remaining bars went down the tonal scale in 1/2 or 1/3-stop increments (set to match the camera's exposure adjustment increment) as they progressed to the left. Make the rightmost bar red, the next one to the left yellow, and the rest green to make it even easier to evaluate. It would be trivial to program and far less CPU intensive than a JPEG-based histogram, and far more useful for evaluating exposure.

Simply adjust exposure until you get a bit of yellow bar but no red bar unless there's some specular highlights you don't mind clipping. And knowing that each exposure adjustment increment would move things one bar to the left or right would make dialing in the perfect exposure trivially easy in most circumstances, especially if the test shot was somewhat underexposed.
Title: Revisiting the "expose to the right" dogma
Post by: BJL on December 12, 2005, 03:03:27 pm
Jonathan's post suggests a related wish: a histogram which covers the full RAW highlight gamut, with a marking (flashing bars?) for the highlight clipping limit of the current JPEG conversion settings. And a clear description of where the f-stop levels lie, at least in the manual.
Title: Revisiting the "expose to the right" dogma
Post by: BernardLanguillier on December 12, 2005, 09:42:14 pm
Agreed, that would already be great.

Cheers,
Bernard
Title: Revisiting the "expose to the right" dogma
Post by: bob mccarthy on December 13, 2005, 11:35:30 am
Quote
Simply adjust exposure until you get a bit of yellow bar but no red bar unless there's some specular highlights you don't mind clipping. And knowing that each exposure adjustment increment would move things one bar to the left or right would make dialing in the perfect exposure trivially easy in most circumstances, especially if the test shot was somewhat underexposed.
[a href=\"index.php?act=findpost&pid=53350\"][{POST_SNAPBACK}][/a]

Isn't that basically what the auto exposure system within the camera does . Or perhaps better, what the exposure system of the camera "should" do!

Bob
Title: Revisiting the "expose to the right" dogma
Post by: Tim Gray on December 13, 2005, 11:57:22 am
Quote
Isn't that basically what the auto exposure system within the camera does . Or perhaps better, what the exposure system of the camera "should" do!

Bob
[a href=\"index.php?act=findpost&pid=53452\"][{POST_SNAPBACK}][/a]

I think there's a difference between what can be accomplished by today's autoexposure systems, and what can be accomplished via a live histogram, which is what I think you're describing.  Maybe given what Sony has been able to do with the R1, a live histogram type AE isn't that far off...

What I'd like to be able to to is simply dial in my tolerance of what is clipped and the camera sets the exposure based on that.  So nomally I want 0 clipping - but if there are specular highlights, maybe 2% is ok - so that's what I'd dial in and that's what would  drive the exposure to be captured in the RAW file.
Title: Revisiting the "expose to the right" dogma
Post by: bob mccarthy on December 13, 2005, 02:00:19 pm
Quote
I think there's a difference between what can be accomplished by today's autoexposure systems, and what can be accomplished via a live histogram, which is what I think you're describing.  Maybe given what Sony has been able to do with the R1, a live histogram type AE isn't that far off...

What I'd like to be able to to is simply dial in my tolerance of what is clipped and the camera sets the exposure based on that.  So nomally I want 0 clipping - but if there are specular highlights, maybe 2% is ok - so that's what I'd dial in and that's what would  drive the exposure to be captured in the RAW file.
[a href=\"index.php?act=findpost&pid=53457\"][{POST_SNAPBACK}][/a]

If we ever get wide DR sensors, then we could dial in contrast range (expansion-contraction) and all be digital Zone system shooters. The only difference would be setting the exposure off the highlight rather than the shadows. With tone mapping, maybe even a shoulder/toe effect could be accomplished

Whew, that would be cool

bob
Title: Revisiting the "expose to the right" dogma
Post by: Jonathan Wienke on December 13, 2005, 02:22:25 pm
Quote
If we ever get wide DR sensors, then we could dial in contrast range (expansion-contraction) and all be digital Zone system shooters. The only difference would be setting the exposure off the highlight rather than the shadows. With tone mapping, maybe even a shoulder/toe effect could be accomplished
If you shoot RAW, you can already do this with the RAW conversion settings and level/curve adjustments. The future is here.
Title: Revisiting the "expose to the right" dogma
Post by: bob mccarthy on December 13, 2005, 02:35:01 pm
Quote
If you shoot RAW, you can already do this with the RAW conversion settings and level/curve adjustments. The future is here.
[a href=\"index.php?act=findpost&pid=53475\"][{POST_SNAPBACK}][/a]

It's just with film and development (expansion) I can get 2 to 3 more stops of DR without going through the HDR or equivalent process.

It was just a thought.

Bob
Title: Revisiting the "expose to the right" dogma
Post by: Andrew Larkin on December 13, 2005, 05:14:04 pm
The histogram on the 5D is marked into five equal sections with vertical lines.  Each line is one stop of exposure apart.

This makes the histogram linear only in terms of exposure - not in terms of raw sensor values.

Andrew
Title: Revisiting the "expose to the right" dogma
Post by: Tim Gray on December 13, 2005, 06:51:26 pm
Quote
The histogram on the 5D is marked into five equal sections with vertical lines.  Each line is one stop of exposure apart.

This makes the histogram linear only in terms of exposure - not in terms of raw sensor values.

Andrew
[a href=\"index.php?act=findpost&pid=53485\"][{POST_SNAPBACK}][/a]

I thought the 5d would have somewhere in the neighbourhood of 8 - 9 stops of dynamic range, 5 doesn't seem right.
Title: Revisiting the "expose to the right" dogma
Post by: Jonathan Wienke on December 14, 2005, 10:27:22 am
The 1-stop thing is an approximation at best, and varies depending on camera settings, especially contrast. Remember that histograms are derived from camera JPEGs even when shooting RAW.