Expose to right, it is as simple as

[hjulenissen]

For anything gamma-related, I think the works of Poynton is of great value:
http://www.poynton.com/PDFs/GammaFAQ.pdf

12 What is contrast ratio? Contrast ratio is the ratio of intensity between the brightest white and the
darkest black of a particular device or a particular environment. Projected
cinema film, or a photographic reflection print, has a contrast ratio of
about 80:1. Television assumes a contrast ratio, in your living room, of
about 30:1. Typical office viewing conditions restrict the contrast ratio of a
CRT display to about 5:1.

13 How many bits do I need
to smoothly shade from
black to white?
At a particular level of adaptation, human vision responds to about a
hundred-to-one contrast ratio of intensity from white to black. Call these
intensities 100 and 1. Within this range, vision can detect that two intensities are different if the ratio between them exceeds about 1.01, corresponding to a contrast sensitivity of one percent.
To shade smoothly over this range, so as to produce no perceptible steps,
at the black end of the scale it is necessary to have coding that represents
different intensity levels 1.00, 1.01, 1.02, and so on. If linear light coding is
used, the “delta” of 0.01 must be maintained all the way up the scale to
white. This requires about 9,900 codes, or about fourteen bits per component.
If you use nonlinear coding, then the 1.01 “delta” required at the black
end of the scale applies as a ratio, not an absolute increment, and
progresses like compound interest up to white. This results in about 460
codes, or about nine bits per component. Eight bits, nonlinearly coded
according to Rec. 709, is sufficient for broadcast-quality digital television
at a contrast ratio of about 50:1.
If poor viewing conditions or poor display quality restrict the contrast
ratio of the display, then fewer bits can be employed.
If a linear light system is quantized to a small number of bits, with black
at code zero, then the ability of human vision to discern a 1.01 ratio
between adjacent intensity levels takes effect below code 100. If a linear
light system has only eight bits, then the top end of the scale is only 255,
and contouring in dark areas will be perceptible even in very poor
viewing conditions.

However, high DR monitors are becoming available and more than 8 bits are needed for them.

Regards,

Bill

Even for new monitors that could gain from >8 bits, it is difficult to actually do. Photoshop does not support >8bits for rendering afaik. Most graphics APIs, drivers etc is hardwired for 8bpc. DVI/HDMI should be able to support >8bits, but for some reason it is not used. Rather, my shiny new Dell monitor supports 10 bits over DisplayPort, but then goes on to warn me against using it because of issues with wake from resume...

-h

[Christopher Sanderson]

In terms of arrival date for this tutorial, what sort of expectations should we have? CY2012? CY2011?

The first files of Camera to Print & Screen  will be released on or about August 8th.

Forum readers may wish to check the Store sometime on Sunday August 7th as that is when I will be making the product 'live' prior to its announcement on What's New on August 8th

[bjanes]

Yes but not based on ETTR (not without having a specific idea how much more exposure can be used prior to blowing out highlights you don’t want blown out). Incident meter, reflective meter, matrix, all are only as good as the person setting them for ideal exposure (in this context raw).

The recent Nikon Matrix metering is fairly sophisticated. It uses 1005 sensors and can detect the contrast of a scene as an aid in determining exposure according to its internally stored database. However, with my D3, the matrix often underexposes with high contrast scenes and you have to watch the histogram and correlate how the histogram displayed by the camera correlates with the raw data. It is still 12-18% overall scene based and a shot in dense fog will give a low contrast image with the histogram peaking in the center and a large empty space in the right portion of the histogram.

It would seem that these 1005 sensors could be used for automated ETTR exposure. Not as good as a live view method, but perhaps usable. It could ignore the brightest 5% (I am only guessing here) or so of the brightest areas as specular highlights and then set exposure such that the next brightest pixels are near saturation.

Regards,

Bill

[digitaldog]

It would seem that these 1005 sensors could be used for automated ETTR exposure.

That’s the key. 1005 or 105 sensors, its being programed to deal with raw, ETTR that’s the issue. But you can do that with a single spot meter read too. Its about ‘calibrating’ the exposure system and targeting the exposure that’s key and that’s mainly and educational process. We have to stop thinking as if we are exposing film (and would that film be neg or chrome?).

[hjulenissen]

It would seem that these 1005 sensors could be used for automated ETTR exposure. Not as good as a live view method, but perhaps usable. It could ignore the brightest 5% (I am only guessing here) or so of the brightest areas as specular highlights and then set exposure such that the next brightest pixels are near saturation.

But if the science of exposure is being able to have the brightest pixel just at the clipping point, then the art of exposure would be figuring out the scene-dependent proper threshold for clipping some percentage of the pixels, would it not? If there is a light-bulb in your picture, you might want to clip those pixels and instead get good exposure of the remaining 95%. But of you are shooting fireworks at night-time, the situation may be opposite. I dont see how any automated algorithm could guess my intentions from those two?

-h

[michael]

That's why, as proposed in the article, there needs to be a user configurable threshold adjustment.

Michael

[bjanes]

That’s the key. 1005 or 105 sensors, its being programed to deal with raw, ETTR that’s the issue. But you can do that with a single spot meter read too. Its about ‘calibrating’ the exposure system and targeting the exposure that’s key and that’s mainly and educational process. We have to stop thinking as if we are exposing film (and would that film be neg or chrome?).

True, but with a 1 degree spot meter you have to find the brightest area of the subject where you want highlight detail and then increase exposure by about 2.5 stops as Bruce Fraser suggested years ago. An automated system can't make that judgement for you but with 1005 readings and some threshold for neglecting specular highlights that can be blown, it should be able to come close.

I don't see the relationship to film, except that most spot meters are calibrated for mid-gray but merely read luminance and one can convert to lux (lumens per square meter) if desired and do whatever one wants with the reading. However the ISO saturation standard for digital sensors (ISO 12232) also involves mid gray (18%) with a half stop for headroom. A digital camera calibrated to the ISO standard for light meters should give 12% saturation when one exposes according to the meter reading, the same as with film.

Regards,

Bill

[fdisilvestro]

I don't know where you get your 8 bit gamma 2.2 DR of 17.6 f/stops. In his post on encoding, Greg Ward gives the DR of 8 bit sRGB at 1.6 orders of magnitude. 10^1.6 = 39.81 or 5.32 f/stops.

I had an error in my previous post writing the equation, but the result is correct.
With gamma encoding, you are able to represent a higher dynamic range from the original scene with less bits, otherwise it would be a waste of time to do it so:

Minimum number of bits = DR / Gamma

If you look in the table in the Norman Koren page you referenced, you´ll see that for exposure zone 11, you don´t have any levels if you use 10 bit linear RAW but you still have 3 levels with 8bits at 2.2 gamma encoding

I haven´t studied yet the article by Greg Ward, but maybe that´s something related to sRGB and not to gamma encoding

[Anders_HK]

True, but with a 1 degree spot meter you have to find the brightest area of the subject where you want highlight detail and then increase exposure by about 2.5 stops as Bruce Fraser suggested years ago. An automated system can't make that judgement for you but with 1005 readings and some threshold for neglecting specular highlights that can be blown, it should be able to come close.

Actually the beauty is that with a 1 degree spot meter we are back to the same simplicity as Ansel Adams!  It means that we as a photographer are in control for judging exact and precise where we wish to place the threshold for bright point (whether we prefer zone 0, I or II) and that we do not leave such decision to the automation meter in our camera. This is the ultimate full control.

Same as Ansel we also need to know what is critical for our exposure for the specific media we use, thus in our case how much above neutral is the bright point, and which is not dependent on Bruce Fraser's recommendation of 2.5 stops but on the performance of the specific sensor in our camera. Thus we need test sensor for this in order to be assertive. Thus indeed we SHOULD think of it like FILM, albeit the film in our camera being a sensor of an emulsion with DIFFERENT PERFORMANCE and CHARACTER than B&W film, slides or polaroids, all of which had different performances and characters. With digital we look primarily at bright point and a specific and large DR for our specific sensor.

Why neutral? This is what our aperture, shutter and the whole of stops and metering in photography is based upon.  

Contrary to Ansel though, we do not necessarily need to scan a scene with the spot meter to find more points if we are lazy, which we could... if we wanted to determine also the other end of DR and want to evaluate the scene for intermediate, e.g. mid tone. Using a spot meter though, it happens we could essentially also put tape over our histogram..., low and behold  

[ErikKaffehr]

Hi,

My suggestion would be that camera exposes ETTR when shooting RAW and in camera JPEG is compensated for the overexposure, histogram and blinking highlights should be based on the raw image.

The camera should make some educated guess on what highlights are spectacular. Most cameras already have exposure compensation, so that option could also be used to bias/shift automatic ETTR exposures.

On the other hand, I don't feel we need to overdo this. There are two reason to ETTR, one is to keep down shoot noise, and on many cameras the best option is to use base ISO. For low contrast subjects I'd suggest that exposing for mid gray would give good enough image quality, even if we could improve the image quality somewhat by exposing fully to the right.

When we have a high contrast scene we need to fully optimize dynamic range and want to be able to produce clean shadows which can be show read noise (which is more disturbing than "shot noise"), the best way to keep shadows clean is to collect as many photons as possible so "signal" in the shadows area will be as good as possible.

In low contrast situations I'd say that increasing exposure by one stop will have very small effect on image quality. Increasing exposure two stops would double signal to noise ratio and may be noticable.

In high contrast situations doubling exposure doubles dynamic range, so even half a stop may matter.

Best regards
Erik

That's why, as proposed in the article, there needs to be a user configurable threshold adjustment.

Michael

[Josh-H]

The camera should make some educated guess on what highlights are spectacular

Canon already does this (to some degree) with the mode 'highlight tone priority'. This is a genuinely under used mode that seeks to place emphasis on preservation of highlight detail. Only available on a 1DSMKIII at ISO200. Not sure if its available on other models (I think it is). http://www.luminous-landscape.com/forum/index.php?topic=20984

[Anders_HK]

For low contrast subjects I'd suggest that exposing for mid gray would give good enough image quality, even if we could improve the image quality somewhat by exposing fully to the right.

Exactly, or rather... depending on the capability of the specific sensor, if the scene/subject is suffice within what we judge as the "quality zone" within the limits of DR, then there seem little point in ETTR.

In such cases it seems we are back to exposing for averaged mid tone...

[NikoJorj]

True, but with a 1 degree spot meter you have to find the brightest area of the subject where you want highlight detail and then increase exposure by about 2.5 stops as Bruce Fraser suggested years ago.

But wouldn't that be completely blown when aiming at something colored?

Be it based on a matrix exposure sensor or a liveview reading, I'd think ETTR automation is way easier that smile recognition implemented in many P&S.

Canon already does this (to some degree) with the mode 'highlight tone priority'.

HTP is only an underexposure (and then, a curve applied in post by DPP or LR as well)... Nor much to do with specular highlight detection as far as I can understand?

[bjanes]

Same as Ansel we also need to know what is critical for our exposure for the specific media we use, thus in our case how much above neutral is the bright point, and which is not dependent on Bruce Fraser's recommendation of 2.5 stops but on the performance of the specific sensor in our camera. Thus we need test sensor for this in order to be assertive.

That is correct. The 2.5 stops comes from the fact that white is about 2.5 stops from 18% (log2 [1/0.18] = 2.47). Many cameras allow 0.5 EV headroom for the highlights, so you would have to add 3 EV for a totally ETTR exposure. All one has to do is expose a gray card (or other uniform surface--white is OK) and determine the resulting raw value in the green channel of the raw file (Iris, DCRaw, or Rawnayze can do the job) and calculate the percent saturation of the sensor. Green is used, since the white balance multiplier is 1.0, but less for the red and blue channels.

Regards,

Bill

[digitaldog]

That is correct. The 2.5 stops comes from the fact that white is about 2.5 stops from 18% (log2 [1/0.18] = 2.47).

And yet, at least when I shoot such a subject, I can open up beyond that another stop, stop and a half without blowing the highlights and normalizing using Exposure in ACR/LR.

If I take an incident reading, same deal. I can open up 1.5 stops over its recommendation and ETTR. So where’s the disconnect here between using a meter as we’ve done for years (and getting proper exposure, for lets say chrome) and finding that these recommendations do not provide optimal exposure for the raw?

[bjanes]

And yet, at least when I shoot such a subject, I can open up beyond that another stop, stop and a half without blowing the highlights and normalizing using Exposure in ACR/LR.

If I take an incident reading, same deal. I can open up 1.5 stops over its recommendation and ETTR. So where’s the disconnect here between using a meter as we’ve done for years (and getting proper exposure, for lets say chrome) and finding that these recommendations do not provide optimal exposure for the raw?

That's a good question. The Kodak Q14 reflection gray scale is useful here. This image is a photograph exposed so that Patch M has a sensor saturation of 18%, which corresponds to a pixel value of 116 in sRGB. Patch A has a density of 0.05 or about 90% reflectance and gives an sRGB value of about 242. This is about the maximum that can be achieved on a paper target (the Babel reflector has a reflectance of 99% and is an ideal white target, but is rather expensive and is also small).



The chart shown below was inspired by a post by Julia Borg which does not seem to be currently online. It shows the density of the patches along with the reflectance and exposure correction needed to place an exposure meter reading at a given luminance. Values for both 18% and 12% meter calibrations are shown, and corresponding pixel values for gamma 1.0, 1.8, and 2.2 are also shown.



Objects with a reflectance greater than 90% are unlikely to be observed in non-specular images in a natural scene, so one could place the luminance for a reflectance of 98% (100% can't be shown as density, which is a log scale). The chart shows the exposure compensation needed to place the reading at a given pixel value. For an 12% meter (used by the Nikon D3 and many other cameras), you would need +3 1/8 EV to place the reading at a pixel value in sRGB of 252 in a gamma 2.2 encoded space. What you need for your camera depends on its calibration.

Regards,

Bill

[OldRoy]

The two threads generated by the original exceptionally lucid article are more interesting than anything I've previously read on ETTR. I understand the issues in a general way however I wouldn't pretend to be able to contribute other than anecdotally. Many thoughts arise, however the two of mine that immediately surface are:

What are the histograms in View NX and Capture NX2 derived from (when viewing in RAW mode)?

In the other thread it's asserted that for a D3 (I use a D700 predominantly) there's little point in using ISO values >800 - it's better to push in post. I'm aware that there are also variables related to the contrast range in the scene, however that was the generalisation that I came away with. This seems to fly in the face of the near-universal assumption (even if not thereby validated!) that this range of sensors is notable for high-ISO performance. I'm here relating this to RAW shooting, rather than JPEGS, which may even be what most people usually shoot.

I was prompted to take a look at some exterior shots that were accidentally over exposed to a degree that I'd considered them write-offs; I'd previously been shooting HDR bracketed interiors. Amazingly NX2, which is restricted to +/- 2 stops exposure compensation, pulled back the building at -2 to an acceptable exposure: the overcast sky was still blown however. I was very surprised indeed.

Now when shooting outside on even fairly bright days I've always tended to use a little -ve compensation, particularly with matrix metering, as I hate blown out looking clouds and I've always tended to use "highlight protection" in NX2 conversions ras opposed to -ve exposure compensation followed by "shadow protection" to raise the shadows. Now I'm wondering if my approach has been fundamentally wrong even though I've usually been broadly satisfied with the results.
Hmm...
Roy

[imagico]

Michael,
while i commend your reiterated plea for a new exposure paradigm for digital i still do not see this happening.  The first generations of digital cameras were designed to encourage film photographers to switch and therefore they pretended exposing digital was just like film.  And current designs continue this since the companies do not want to alienate their customers by telling them now things have changed.  What i do see happening is a quite, invisible change to ETTR in cameras without RAW output. Maybe this is already in progress - there is no way to tell without a direct comparison or raw output.  I would be very happy to be proven wrong of course.

I think you should be careful when dismissing technical arguments concerning your text on the basis of the perceived credentials of the one bringing then up though.  I don't want to participate in the technical argument here but ideas should be judged by their own merit, not by the merit of the one who states them.  This is especially important here i think since you also criticize an established opinion (which is also still taught in lots of books and schools by people with widely recognized credentials).

[bjanes]

In the other thread it's asserted that for a D3 (I use a D700 predominantly) there's little point in using ISO values >800 - it's better to push in post. I'm aware that there are also variables related to the contrast range in the scene, however that was the generalisation that I came away with. This seems to fly in the face of the near-universal assumption (even if not thereby validated!) that this range of sensors is notable for high-ISO performance. I'm here relating this to RAW shooting, rather than JPEGS, which may even be what most people usually shoot.


Roy

Roy,

For a detailed explanation of how to read DXO dynamic range graphs and how to use them for exposure decisions, see this post by Emil Martinec. You might also enjoy reading Emil's post on Signal to Noise and Exposure Decisions, which discusses ETTR and Michael's erroneous statements about the number of levels in the brightest f/stop of a 12 bit digital capture by current sensors.

Regards,

Bill

[Nigel Johnson]

As Nicolas (NikoJorj) implied with his comment 'But wouldn't that be completely blown when aiming at something colored?' the various posts suggesting using spot-meters completely miss the fact that this would not address the problem of blowing out an individual channel.

The automatic ETTR system proposed by Michael has a huge advantage in that it would separately address the exposures for each of the RGB channels of the sensor and select an ETTR exposure that would maximise the exposure in one or more of the channels. If the area giving this exposure is near neutral in colour, this is likely to be similar to the exposure that could be calculated using a spot-meter (having previously determined the compensation needed to produce near saturation). However, if the area giving the individual channel ETTR exposure is a saturated colour, the exposure is likely to be very different from that obtained using a spot-meter.

The suggestion of using a multi-segment meter such as the sensor used for matrix metering would probably also have the same problem as using a spot-meter (unless it was able to separately measure RGB channels, with similar spectral sensitivity to the imaging sensor).

It would obviously be possible, but usually impractical, to use coloured filters with a spot-meter (with appropriate spectral characteristics and individual correction factors) to determine individual channel ETTR exposures and then set the lowest of the three exposures.

An additional advantage of the system proposed by Michael is that it could also be applied to those sensors that use more than three colours (I can't remember which manufacturer it is who has green and blue-green filters for the two green pixels in the Bayer pattern) since it would be applied to the individual channels prior to the demosaicing and conversion to a conventional RGB image.

Regards
Nigel