Exposing to the right! An example

[billhercus]

It is a cold, windy spring day here so looking out of my attic window it came to me that I should experiment with 'exposing to the right.' After all, what else is a chap to do?

Two pictures below are from a 100% crop of two images taken within a few seconds from my attic window to the small ex fishing village which is about 4 miles away. The camera is on a tripod and the lens is a Sigma 100-300 f4 EX with a Sigma 1.4 TC at full stretch.
The top image was exposed at 1/160th and the botttom at 1/40th - both at f13.
Images were processed in RawShooter identically, except for exposure correction. There is no clipping in the longer exposure's histogram. Limited corrections were made in Photoshop to levels + some sharpening and sizing. Identical corrections were made although both images would require different processing to get the best results, I suspect.

Others will know far more than I about that!

From this simple experiment, the differences in these pictures are quite marked with the bottom (exposed to the right) image looking as if it has been through some noise reduction software (it hasn't).

I'd be interested in any guidance/comments here about how the bottom image should be processed for best effect or any general views on this technique.

http://www.mayfieldghouse.freeserve.co.uk/Webpics/Exposetest.jpg

[scottbergerphoto]

Shooting to the right is done to make sure that you capture the most data you can for an image. 1/2 the available data is in the far right stop of the histogram. It may be indistinguishable from a shot not done that way on a computer monitor. Your second shot is blurry, I suspect from the longer exposure on a shaky tripod? I don't think any of what you are seeing is do to shooting to the right.

[PacAce]

It is a cold, windy spring day here so looking out of my attic window it came to me that I should experiment with 'exposing to the right.' After all, what else is a chap to do?

Two pictures below are from a 100% crop of two images taken within a few seconds from my attic window to the small ex fishing village which is about 4 miles away. The camera is on a tripod and the lens is a Sigma 100-300 f4 EX with a Sigma 1.4 TC at full stretch.
The top image was exposed at 1/160th and the botttom at 1/40th - both at f13.
Images were processed in RawShooter identically, except for exposure correction. There is no clipping in the longer exposure's histogram. Limited corrections were made in Photoshop to levels + some sharpening and sizing. Identical corrections were made although both images would require different processing to get the best results, I suspect.

Others will know far more than I about that!

From this simple experiment, the differences in these pictures are quite marked with the bottom (exposed to the right) image looking as if it has been through some noise reduction software (it hasn't).

I'd be interested in any guidance/comments here about how the bottom image should be processed for best effect or any general views on this technique.

http://www.mayfieldghouse.freeserve.co.uk/Webpics/Exposetest.jpg
I'm curious to know how you determined the exposure to use for your first sample image. Was the shot properly exposed? The reason I'm asking is because the shot seems noisy and it looks like it might have been underexposed and corrected to get the right exposure.

The idea with "shooting to the right" is to use as much of the sensor's dynamic range as possible and that means making sure that the brightest part of the image is always at the right edge of the histogram without blowing out the highlight details that you would care about. Shooting to the right is most effective for low key shots where the bulk of the data is in the left half of the histogram. For normal scenes where there is data already near the right side of the histogram, shooting toward the right may not produce a noticeable change in the final image and, in fact, may risk blowing highlight details.

After you've taken a picture shot to the right, you now have to correct the exposure to get it where it's supposed to be, especially for low-key shots where it'll be obvious that the shot was over exposed. Doing this results in a cleaner image, expecially in the dark shadow areas.

[slin100]

Leo is right. The purpose of exposing to the right is to maximize dynamic range. RAW data is linear, so the data is even distributed throughout all stops. 1/2 of the data in the brightest stop is only true for JPGs due to gamma-encoding, which effectively redistributes bits into the brighter regions.

[scottbergerphoto]

The idea with "shooting to the right" is to use as much of the sensor's dynamic range as possible and that means making sure that the brightest part of the image is always at the right edge of the histogram without blowing out the highlight details that you would care about. Shooting to the right is most effective for low key shots where the bulk of the data is in the left half of the histogram. For normal scenes where there is data already near the right side of the histogram, shooting toward the right may not produce a noticeable change in the final image and, in fact, may risk blowing highlight details.

After you've taken a picture shot to the right, you now have to correct the exposure to get it where it's supposed to be, especially for low-key shots where it'll be obvious that the shot was over exposed. Doing this results in a cleaner image, expecially in the dark shadow areas.
Well said.

[scottbergerphoto]

1/2 of the data in the brightest stop is only true for JPGs due to gamma-encoding, which effectively redistributes bits into the brighter regions.
According to this article http://www.luminous-landscape.com/tutorials/expose-right.shtml , which has been cited numerous times as the reason for exposing to the right, that is not correct.

Home on The Dynamic Range
Let's assume for the purposes of illustration that a digital SLR has a dynamic range of 5 stops (it's usually closer to 6 stops, but let's not quibble). When working in RAW mode, which you should be, most cameras record a 12 bit image. (Yes, we say it's in 16 bit mode, but the reality is that it's only recording 12 bits in a 16 bit space. Better than 8, but not as good as a real 16 bits would be).

A 12 bit image is capable of recording 4,096 (2^12) discrete tonal values. One would think that therefore each F/Stop of the 5 stop range would be able to record some 850 (4096 / 5) of these steps. But, alas, this is not the case. The way that it really works is that the first (brightest) stop's worth of data contains 2048 of these steps — fully half of those available.
Why? Because CCD and CMOS chips are linear devices. And, of course, each F/Stop records half of the light of the previous one, and therefore half the remaining data space available. This little table tells the tale.

Within the first F/Stop, which contains the Brightest Tones 2048 levels available
Within the second F/Stop, which contains Bright Tones 1024 levels available
Within the third F/Stop, which contains the Mid-Tones 512 levels available
Within the fourth F/Stop, which contains Dark Tones 256 levels available
Within the fifth F/Stop, which contains the Darkest Tones 128 levels available

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.
But, we all know (or at least should by now) that the worst sin in digital imaging is to blow out the highlights — just as it was when shooting slide film. Once they're blown (past the right-hand edge of the histogram) it's bye-bye data.
Quoted from the above linked article.

[PacAce]

To further illustrate what Scott posted above, I took a picture which has data that covers both ends of the range recorderable by a typical digital camera This is shown in first attachment below. I shot it in RAW and converted it as a standard TIFF image. What the converter did under the covers was apply a tone curve that more evenly distributed the image tones to what we're used to seeing. The data was also normalized from 12-bit to 16-bit so that the white point of the original 12-bit data from the camera lines up with the white point of the 16-bit format.

The 2nd attachment show you what the image from the camera looks like without any tone curve applied or normalized to 16-bit data. The white part of the image really isn't white because the maximum value you can specify with 12 bits falls short of the maximum value in 16 bit. You'll also note that you really can't make out the two dark tones on the left side of the image. That's because the data is all bunched up together.

Continued on next post....
.

[PacAce]

To get the linear image to a more useful image, we would have to apply a tone curve similar to what the RAW processor applies when converting RAW to JPEG or TIFF. The first attachment show a sample tone curve that I appled to the linear image. It got the image close to what the RAW converter put out. And the 2nd attachment is the linear image after the tone curve was applied to it in PSCS.

BTW, if you go back to my Part 1 above and compare the histograms of image 1 and image 2 you can see the the two spikes on the right representing the bright tones are very close together whereas in the second image, they are pretty spead apart. This further illustrates what Scott posted above.
.

[slin100]

Thanks for the excellent examples. I stand corrected regarding bit allocations.

[billhercus]

To be honest, I had assumed that a simple correction to exposure on the 'overexposed' image would be all that this technique requires but it seems that there is more to it.

To answer some of the questions which may explain the irregular quality of the pictures used, I submit the following:-

Exif for first shot:
(Second only difference is shutter speed)
100-300mm @ 420mm
1/160 sec, f/13
Mode: Av
Metering: Evaluative
ISO: 200
AF mode: AI Focus AF
Drive: Single frame shooting
White balance: Auto
Flash: Off
File size: 7,088KB
Image size: 3504 x 2336
Saturation: Normal
Sharpness: Normal
Contrast: Normal
Tone: Normal

The tripod is a Kenlock 3000 (about 20 years old) and I used a remote cord (but no mirror lockup) - the image is some 4 miles away.
I had left the focus on AI Focus AF which was not intended but I don't know if that would have an effect on this - can't think why.

For the first shot I used +0.35 EC in RSE which means exposure was not correct I know but would that not be within an acceptable tolerance?

For the exposed to the right image, I corrected -0.80 in RSE

The 'normal' image RSE histogram showed no detail in the right hand eigth of the display.
Capture sharpening by Photokit used High-Res and superfine edge
I used USM amount 22, radius 50 and threshold zero.
Output sharpening on Web and 800 pixel Superfine.
Any guidance on post processing errors will be gratefully received. :oops:

Scott's explanation from The Luminous Landscape was my original motivation for playing with this technique. There is obviously more to it than checking histogram, overexposing then correcting exposure in your RAW converter which is what I'd thought.
I shall certainly be taking more images 'to the right' bearing in mind comments here, as I feel sure this will be worth the effort, principally because of the pure logic of the case.

PacAce, I will study your reply at length ......

Once again, thanks for the distilled wisdom here!

[Harry Settle]

Excuse me, but I'm a little slow at some of this. When I read the "Shoot to the right", article at Luminous Landscape I assume that they are saying to shoot the picture exposed to the right of the histogram in order to ensure you have captured more detail in the low end. What you all seem to be doing here is adjusting a shot in post.

[PacAce]

Excuse me, but I'm a little slow at some of this. When I read the "Shoot to the right", article at Luminous Landscape I assume that they are saying to shoot the picture exposed to the right of the histogram in order to ensure you have captured more detail in the low end. What you all seem to be doing here is adjusting a shot in post.
Yes, you are correct. Shooting to the right does mean to set the exposure so the the data shows up as far right on the histogram as possible without blowing the highlights. And, as I mentioned in a previous post, this is most effective for dark or low-key scenes. However, after you have shot to the right, you just can't leave it that way or else the image will look over exposed. So, you'll need to post process the image to bring the exposre back in line with what it's supposed to be. In effect, in post processing, you will have to move the exposure to the left again.

[Harry Settle]

Thanks

[PacAce]

Here are a couple of examples of what shooting to the right does for you for low-key or dark subjects and scenes. All shots were taken at ISO 1600.

In the 1st attached image, A was shot using the correct exposure. B was shot 1 Stop under exposed. And C was shot to the right so that it was really over exposed as you can see from the bottom left image. The exposure for C was then corrected during RAW conversion processing. The image on the bottom right is the corrected image of C.

The 2nd attached image shows a 100% crop of A (the correctly exposed image), B (the under exposed image) after it was corrected in PSCS for the correct exposure, and C (shot to the right and corrected). As you can see high ISO noise is marginal in A, really bad in B and practically non-existent in C. Check out how silky smooth the background of C is.
.

[PhotosGuy]

Great job, Leo!

[scottbergerphoto]

Leo,
What a great demonstration and explanation of this technique. I have made it a sticky for the next week.

[PacAce]

Thanks, guys. Just doing my part to help clarify any obscurities there may be regarding "shooting to the right". :D

[billhercus]

That is a superb explanation, brilliantly illustrated Leo!!

Many thanks :D

[billhercus]

That is a superb explanation, brilliantly illustrated Leo!!

My second picture shows a similar noise reduction to your 'C', I think?

Many thanks :D

[lawmans3]

Leo, Great job on this! Makes everything crystal clear now! Thank you....

[PacAce]

Leo, Great job on this! Makes everything crystal clear now! Thank you....
;) Glad you found it helpful. :)

[Tom W]

Yes, Leo, it is a great illustration of why "exposing to the right" works.

It deserves another week of stickiness IMHO.

[EddieA]

As my first contribution to this excellent forum, and regarding the requirement of preserving the detail in the tones represented in the left of the histogram, I would like to remind you of that great pioneer of photography who was among the first to be confused by this.
In the words of 'Painless' Peter Potter or Peter Potter Junior (aka Bob Hope) c. 1948/52 (can't remember whether it was Paleface' or 'Son of')..
"He leans to the left so shoot to the right.." :)
I'm sure many of you will have seen this.
Forgot to say this : Great explanation!

[Sean-Mcr]

Thank's for a great demo Leo

Really helpful;)

[marathon]

wow... now this is something useful. thank you for clarifying this so well. you've made my day :)

[rdenney]

...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.


While the conclusions of this thread are correct and will modify my own practice, I would like to state for the record that this statement on Reichmann's site is wrong.

He makes the point that the top stop of dynamic range contains half the tonal values. That is correct. What he forgets is that the histogram is not scaled in stops, but in those linear tonal values. So that top stop actually includes the right half of the histogram. Zones (in the Zone System sense) and stops follow a geometric rather than linear progression.

So, while it's important to make full use of that top stop of dynamic range, it's also important not to consign that stop to the right fifth of the histogram. Doing so will surely make blown-out highlights. Remember that we only know if highlights are blown out if the histogram is continuous right up to Zone X. If there is a gap, we might have a big hump of histogram off the right end of the scale and not even know it. In addition to looking at the histogram, we also need to turn on that highlight warning display so that we can see how much of the image is in the top bit.

Rick "who thinks the LCD images often lead us astray by blowing out more easily than the underlying data" Denney

[Poe]

While the conclusions of this thread are correct and will modify my own practice, I would like to state for the record that this statement on Reichmann's site is wrong.

He makes the point that the top stop of dynamic range contains half the tonal values. That is correct. What he forgets is that the histogram is not scaled in stops, but in those linear tonal values. So that top stop actually includes the right half of the histogram. Zones (in the Zone System sense) and stops follow a geometric rather than linear progression.

So, while it's important to make full use of that top stop of dynamic range, it's also important not to consign that stop to the right fifth of the histogram. Doing so will surely make blown-out highlights. Remember that we only know if highlights are blown out if the histogram is continuous right up to Zone X. If there is a gap, we might have a big hump of histogram off the right end of the scale and not even know it. In addition to looking at the histogram, we also need to turn on that highlight warning display so that we can see how much of the image is in the top bit.

Rick "who thinks the LCD images often lead us astray by blowing out more easily than the underlying data" Denney

This is very good to know, as I do see some dividing lines on my histrogram of my Rebel XT. Now I know that its an actual linear distribution of the data and not geometrical as I had originally thought.

I've been reading about the zone system on this website http://www.cicada.com/pub/photo/zs/quicktour/1.html and it really has pieced together everything I've picked up from reading Peterson's "How to see creatively" and "Understanding Exposure." Although geared toward one who photographs on film since it describes development techniques, I'm sure it can be applicable when adjusting raw files in your editor of choice.

[PacAce]

While the conclusions of this thread are correct and will modify my own practice, I would like to state for the record that this statement on Reichmann's site is wrong.

He makes the point that the top stop of dynamic range contains half the tonal values. That is correct. What he forgets is that the histogram is not scaled in stops, but in those linear tonal values. So that top stop actually includes the right half of the histogram. Zones (in the Zone System sense) and stops follow a geometric rather than linear progression.

So, while it's important to make full use of that top stop of dynamic range, it's also important not to consign that stop to the right fifth of the histogram. Doing so will surely make blown-out highlights. Remember that we only know if highlights are blown out if the histogram is continuous right up to Zone X. If there is a gap, we might have a big hump of histogram off the right end of the scale and not even know it. In addition to looking at the histogram, we also need to turn on that highlight warning display so that we can see how much of the image is in the top bit.

Rick "who thinks the LCD images often lead us astray by blowing out more easily than the underlying data" Denney
I must have missed this post by Rick when he first posted it a while ago but I have to correct an error on his part. Altough he is correct in stating that the histogram is not scaled in full stops, it is very close except at the left and right ends of the histogram.

However, it is simply not correct to say that the right half of the histogram represents a top stop of the image. This is because the histogram is a representation of the image that has been converted to JPEG which has already been normalized, not the RAW image itself. If anybody would like to test this out, simply take a picture of a white or gray card and add plus EC in one stop increments to the exposure and watch the histogram as the graph moves towards the right edge of the scale.

[PacAce]

... Now I know that its an actual linear distribution of the data and not geometrical as I had originally thought.

Not so. See my previous post above.

[Poe]

Are RAW images 12 bit or 16 bit?

[rdenney]

However, it is simply not correct to say that the right half of the histogram represents a top stop of the image. This is because the histogram is a representation of the image that has been converted to JPEG which has already been normalized, not the RAW image itself. If anybody would like to test this out, simply take a picture of a white or gray card and add plus EC in one stop increments to the exposure and watch the histogram as the graph moves towards the right edge of the scale.

So, why don't you or someone who knows tell us the values of each of the zones of a 10-zone gray scale, in terms of luminance on the 0-255 scale that I see on my histogram. That would be good to know. Don't have my camera here or I'd try it myself.

I do know that if I try to follow all the advice on exposing to the right too aggressively, I end up with blown highlights or posterized skies when I try to burn them in the way I like them. A blown highlight is gone forever, just like on Velvia, heh, heh.

Rick "pondering a way to calibrate the histogram in zones" Denney

[PacAce]

Are RAW images 12 bit or 16 bit?
12 bit if we're talking EOS digital cameras.

[Robert_Lay]

So, why don't you or someone who knows tell us the values of each of the zones of a 10-zone gray scale, in terms of luminance on the 0-255 scale that I see on my histogram. That would be good to know. Don't have my camera here or I'd try it myself.

I do know that if I try to follow all the advice on exposing to the right too aggressively, I end up with blown highlights or posterized skies when I try to burn them in the way I like them. A blown highlight is gone forever, just like on Velvia, heh, heh.

Rick "pondering a way to calibrate the histogram in zones" Denney
Dear Rick,

I'm with you - it would be nice to have that calibration. However, I'm not at all sure that it's even possible. Considering that the industry tries very hard to keep their demosaicing algorithms top secret, it is unlikely that such a calibration can be obtained unless one takes it upon themselves to do it for their own camera. And then there is no assurance that it would be transferable to another camera of the same make and mode.

I made such a calibration for my G5 about a year ago. I also applied that calibration to a program that I wrote in VB6 which is an exposure learning tool. The program accepts a calibration curve for use instead of the linear curve that is the default. Nice as it is, it is not useful enough to really do anything with it. In other words, even though I have such a curve for my camera, I only use that curve from my memory - not as a printed card. The reason is that it is basically the same shape you would expect from film except that the toe and the shoulder are a bit more abrupt than most film.

The curve that I'm talking about is only of value when shooting JPG. For RAW mode, I have never tried to do such a calibration, because I'm not at all sure how I would do it. It would have to be based upon some arbitrary setting of all of the sliders in ACR and would probably come out different if you used RSE.

[DocFrankenstein]

I think the luminance histogram is can be improved on many levels.

It is not the average in terms of gray values. It's biased toward the green - the noisiest channel.

Second - the histogram my rebel gives is the JPEG histogram. And it changes with contrast/saturation settings I set in jpeg.

If you move the liminance to the right, you're guaranteed to blow the red channel in portraits. The resulting image is flat and looks not adequate.

What a photographer needs is the 3 channel histogram. You should view it full screen to make sure no highlight detail is lost... and it would be nice for the histogram to be linear. The current manufacturers shift everything to the left, giving me viewing space of 3 pixels... and in these 3 pixels lies the difference between lost highlights and good exposure.

So I just rely on a handheld meter pointed towards the light. The color histograms come out perfect.

[PacAce]

So, why don't you or someone who knows tell us the values of each of the zones of a 10-zone gray scale, in terms of luminance on the 0-255 scale that I see on my histogram. That would be good to know. Don't have my camera here or I'd try it myself.

I do know that if I try to follow all the advice on exposing to the right too aggressively, I end up with blown highlights or posterized skies when I try to burn them in the way I like them. A blown highlight is gone forever, just like on Velvia, heh, heh.

Rick "pondering a way to calibrate the histogram in zones" Denney
Rick, I'm not sure I understand what you are asking for here. Why does there need to be an absolute value for each of the 10-zone gray scale as represented in the histogram. The idea behind exposing to the right is to take what ever zone happens to be the brightest in the scene you are shooting and expsose it such that that zone is at the right edge of the histogram. So, if you are shooting a scene where the brightest element is in Zone III, then your exposure should be set so that the data from Zone III lies in right edge of the histogram. During post processing, you would adjust the levels of the image so that Zone III lies close to where it should be (or where you want it to be) in the histogram of the processed image.

[PacAce]

I think the luminance histogram is can be improved on many levels.

It is not the average in terms of gray values. It's biased toward the green - the noisiest channel.

Second - the histogram my rebel gives is the JPEG histogram. And it changes with contrast/saturation settings I set in jpeg.

If you move the liminance to the right, you're guaranteed to blow the red channel in portraits. The resulting image is flat and looks not adequate.

What a photographer needs is the 3 channel histogram. You should view it full screen to make sure no highlight detail is lost... and it would be nice for the histogram to be linear. The current manufacturers shift everything to the left, giving me viewing space of 3 pixels... and in these 3 pixels lies the difference between lost highlights and good exposure.

So I just rely on a handheld meter pointed towards the light. The color histograms come out perfect.
How will a linear representation of the image on the histogram be any different from a normalized representation? In either case, you still can't tell if the image barely gets blown, is blown by a little or is blown by a lot.

Also, what do you mean by "current manufacturers shift everything to the left, giving me viewing space of 3 pixels... and in these 3 pixels lies the difference between lost highlights and good exposure"?

[Poe]

To further illustrate what Scott posted above, I took a picture which has data that covers both ends of the range recorderable by a typical digital camera This is shown in first attachment below. I shot it in RAW and converted it as a standard TIFF image. What the converter did under the covers was apply a tone curve that more evenly distributed the image tones to what we're used to seeing. The data was also normalized from 12-bit to 16-bit so that the white point of the original 12-bit data from the camera lines up with the white point of the 16-bit format.

The 2nd attachment show you what the image from the camera looks like without any tone curve applied or normalized to 16-bit data. The white part of the image really isn't white because the maximum value you can specify with 12 bits falls short of the maximum value in 16 bit. You'll also note that you really can't make out the two dark tones on the left side of the image. That's because the data is all bunched up together.

Continued on next post....
.

Leo, how did you get the data to not normalize from 12 bit to 16 bit in the second photo? I'm working in ACR right now and the tone curve and histrogram seem to scale to 8 bit. All I did was open my RAW file from the bridge, into ACR, then have it open the RAW file into CS2, so I have a .CR2 file open. Do I need to change my process in some way?

I've been toying around in ACR for a while now. I set my exposure, shadows, contrast and saturation to zero in ACR for the RAW file in question. This gives me a rather dull photograph compared to the way the settings were determined automatically. I assume that the dull photograph is how the photograph was actually rendered on the sensor at the time of exposure. By keeping those settings zero and then going and playing with the tone curve, I'm able to actually take the photograph back to how the computer would automatically set exposure, shadow, contrast and saturation (which is very cool indeed).

I'd still like to know why the tone curve in ACR as well as that in CS2 is still scaled from 0 to 255 for the RGB rather than 0 to 65535 as I'm expecting it to be in 16 bit space.

[PacAce]

Leo, how did you get the data to not normalize from 12 bit to 16 bit in the second photo? I'm working in ACR right now and the tone curve and histrogram seem to scale to 8 bit. All I did was open my RAW file from the bridge, into ACR, then have it open the RAW file into CS2, so I have a .CR2 file open. Do I need to change my process in some way?

Poe, the 2nd image on the post you referenced is a straight linear conversion from raw to tiff (using EVU, I think it was). That image has no tone curve applied to it nor has it been normalized (white point adjusted). Since the image was converted to 16-bit but the raw file itself only had 12 bits of data, there was a 4-bit area "nothingness" in the rightmost side of the histogram.


I've been toying around in ACR for a while now. I set my exposure, shadows, contrast and saturation to zero in ACR for the RAW file in question. This gives me a rather dull photograph compared to the way the settings were determined automatically. I assume that the dull photograph is how the photograph was actually rendered on the sensor at the time of exposure. By keeping those settings zero and then going and playing with the tone curve, I'm able to actually take the photograph back to how the computer would automatically set exposure, shadow, contrast and saturation (which is very cool indeed).

I'd still like to know why the tone curve in ACR as well as that in CS2 is still scaled from 0 to 255 for the RGB rather than 0 to 65535 as I'm expecting it to be in 16 bit space.
That's a good question to which I do not have an answer. If you look at the other tools, like the color picker or the info palette, you can see that they are all scaled from 0 to 255, too. To tell you the truth, I never really gave it that much thought because I always thought that the 0 to 255 scale was used primarily for display purposes and that internally, PS is working with the correct scale for the bit mode in use.

[DocFrankenstein]

How will a linear representation of the image on the histogram be any different from a normalized representation. In either case, you still can't tell if the image barely gets blown, is blown by a little or is blown by a lot.
You can tell better because lineraly it's not shifted to the right. So the highlight information is crucial. Let the shadows fall where they may.

Also, what do you mean by "current manufacturers shift everything to the left, giving me viewing space of 3 pixels... and in these 3 pixels lies the difference between lost highlights and good exposure"?
I meant to say to the right. Current manufacturers shift the histogram to the right.

So in the itsy bitsy fifth column of your histogram you have 2048 levels. (luminous/reichmann) Yet, it's the part you care about the most in digital medium.

Question is - why shove these 2048 levels into smaller the rightmost column of the histogram.

With linear histogram you'd have the fifth column as wide as 50% of the histogram. The second column 25% of the histogram... etc...

Makes sense?

[Poe]

o tell you the truth, I never really gave it that much thought because I always thought that the 0 to 255 scale was used primarily for display purposes and that internally, PS is working with the correct scale for the bit mode in use.

If PS is working behind the scenes then it seems like one wouldn't really gain much from working in 16bit space compared to 8bit if you can't specificy precise color luminosity. It seems to me that a chosen paramter could vary from 1x to 256x the amount you specify using the 8 bit display scale.

BTW, what is EVU?

[PacAce]

If PS is working behind the scenes then it seems like one wouldn't really gain much from working in 16bit space compared to 8bit if you can't specificy precise color luminosity. It seems to me that a chosen paramter could vary from 1x to 256x the amount you specify using the 8 bit display scale.

BTW, what is EVU?
The advantage of processing in 16-bit mode is that it provides a finer granularity for editing than 8-bit does. That means less susceptibility to banding and posterizaton, especially when the data is already true 16-bit (or at least 12-bit if it's from raw data files) to begin with.

The fact that you can't precisely specify the color via the tools is of little consequence, IMHO. To demonstate this, go to color palette in PS and set the color to R=255, B=0, G=0 and bucket fill a rectangular selection with this color. Now set the color to R=254, B=0, G=0. Make another selection adjacent to the previous selection and bucket fill that selection with the new color. Can you tell the two colors apart? No, right? So, if you subdivided the color difference in this example into 255 parts instead of one, is it really going to make a difference? No. :)

EVU = Canon EOS Viewer Utility

[rdenney]

Rick, I'm not sure I understand what you are asking for here. Why does there need to be an absolute value for each of the 10-zone gray scale as represented in the histogram. The idea behind exposing to the right is to take what ever zone happens to be the brightest in the scene you are shooting and expsose it such that that zone is at the right edge of the histogram. So, if you are shooting a scene where the brightest element is in Zone III, then your exposure should be set so that the data from Zone III lies in right edge of the histogram. During post processing, you would adjust the levels of the image so that Zone III lies close to where it should be (or where you want it to be) in the histogram of the processed image.

It's just a way of understanding the histogram in a way that can be adjusted. If I knew where the zonal boundaries were on the histogram, then I would know how many stops I need to move my exposure to get a certain movement on the histogram. The alternative is trial and error. Using the histogram already means that we are burning a shot just as an exposure check. I might do the same using Polaroid with a view camera. But Polaroid has a different response than the film, so I have to understand the differences. Here, I want to understand the differences between the LCD and the histogram and what ends up on the sensor.

I have marked my Pentax spotmeter so that the exposure scale is calibrated in zones (Zone VI offered this modification for sale, but I did my own). The sensitivity of film has a definite toe, but tests I've seen indicate that the sensor has a toe as well on the highlight end, and one that is quite similar to color negative film. Knowing where I am in relation to that toe would be useful. If the LCD would really show me where I was blown out, and if the camera let me set the blown highlight threshold, then I would feel more comfortable pushing my histograms all the way to the right edge. I see that I grossly underexposing many of my images from early in my 10D ownership, but that was a reaction to blown highlights that I could not detect until it was too late.

If I knew where those boundaries were, I could compare my important highlights with the known peaks on the histogram and be able to predict accurately whether they are blown.

This really should be a lot easier than the testing we Zone System nuts used to do with film!

Rick "who thinks a small printout of a sample histogram with zones marked on it would make a nice laminated cheat sheet for the camera bag" Denney

[rdenney]

I'd still like to know why the tone curve in ACR as well as that in CS2 is still scaled from 0 to 255 for the RGB rather than 0 to 65535 as I'm expecting it to be in 16 bit space.

All the PS histograms and tone curves are scaled to 8 bits. In 16-bit space, each bit on the display integrates 8 bits in the file. The file is still in 16-bit space, but we can only see a summary on the histogram. Providing a 16-bit scale wouldn't do much that is useful, because there aren't 64K pixels on the screen display to show the histogram at that level of detail, and I'd hat to have to scroll a 16-bit histogram.

As I understand it, ACR scales the 12 bits from the camera to 16-bit space by a straight re-sampling--else we've have an unused portion of the histogram at one end or the other.

Rick "content with the 8-bit summary in the displays" Denney

[rdenney]

If PS is working behind the scenes then it seems like one wouldn't really gain much from working in 16bit space compared to 8bit if you can't specificy precise color luminosity. It seems to me that a chosen paramter could vary from 1x to 256x the amount you specify using the 8 bit display scale.

BTW, what is EVU?

I don't think there is much to be gained from being able to specify luminosity more precisely than using an 8-bit scale. The point of working in 16-bit space is that you can stretch the tone curve radically and still have good tonal separation in 8-bit space. Remember that monitors can't display more than 8 bits (even if they say they can) and printers convert everything to 8-bits before they print. 255 levels really is quite a fine gradation. But the trick is to convert it to 8 bits as the final step after making all the tonal manipulations (I just let the printer driver or the "save for web" process do that for me).

Rick "who doesn't like posterization but who does sometimes want extreme tonal movements" Denney

[Robert_Lay]

You can tell better because lineraly it's not shifted to the right. So the highlight information is crucial. Let the shadows fall where they may.


I meant to say to the right. Current manufacturers shift the histogram to the right.

So in the itsy bitsy fifth column of your histogram you have 2048 levels. (luminous/reichmann) Yet, it's the part you care about the most in digital medium.

Question is - why shove these 2048 levels into smaller the rightmost column of the histogram.

With linear histogram you'd have the fifth column as wide as 50% of the histogram. The second column 25% of the histogram... etc...

Makes sense?

Unfortunately, the fact that the histogram is divided into 5 equal portions is still making people believe that each of those columns is supposed to be one full stop, or whatever.

It isn't!

The five separate columns have no meaning whatsoever, other than to divide the 256 levels into 5 groups, more or less.

[Robert_Lay]

Poe, the 2nd image on the post you referenced is a straight linear conversion from raw to tiff (using EVU, I think it was). That image has no tone curve applied to it nor has it been normalized (white point adjusted). Since the image was converted to 16-bit but the raw file itself only had 12 bits of data, there was a 4-bit area "nothingness" in the rightmost side of the histogram.


That's a good question to which I do not have an answer. If you look at the other tools, like the color picker or the info palette, you can see that they are all scaled from 0 to 255, too. To tell you the truth, I never really gave it that much thought because I always thought that the 0 to 255 scale was used primarily for display purposes and that internally, PS is working with the correct scale for the bit mode in use.
Dear Leo,

There is a way to configure that Info Palette so that it shows values in 16 bit mode. Actually, it will show a range of 0 through 32768 rather than 0 through 65535 as you would expect. I'm trying now to figure out how to put it in that mode, but haven't succeded yet. Obviously, it only works with images that are in 16 bit mode.
Edit - Wrong! It works with images in either mode, but you have to take the mapped values with a grain of salt.

[Robert_Lay]

Dear Leo,
Now I've figured out how to do it.

First, select the Info Palette.
Then, look for the little eyedropper tool in the left half of the palette.
Left click on it and you get a menu that includes the means to select between 8, 16 and 32 bits.

BTW, there is an interesting story behind the range of values 0 - 32768.

Ordinarily you would not think that you can fit all of those values into 15 bits, and you can't. In 15 bits you have a range of only 0 - 32767.

The extra value is provided for a strange reason - they wanted an unambiguous mid-point between zero and 32768, and that was the way they did it. It's all mentioned in Bruce Fraser's book.

[PacAce]

Dear Leo,
Now I've figured out how to do it.

First, select the Info Palette.
Then, look for the little eyedropper tool in the left half of the palette.
Left click on it and you get a menu that includes the means to select between 8, 16 and 32 bits.

BTW, there is an interesting story behind the range of values 0 - 32768.

Ordinarily you would not think that you can fit all of those values into 15 bits, and you can't. In 15 bits you have a range of only 0 - 32767.

The extra value is provided for a strange reason - they wanted an unambiguous mid-point between zero and 32768, and that was the way they did it. It's all mentioned in Bruce Fraser's book.
Thanks for the info, Bob. I'll try it when I get home. I'm assuming this will work for PSCS since that's what I have. At work I have PS 7 and I can't get it to switch to 16-bit display.

[Poe]

I was thinking more about the exposing to the right concept and the zone system.

Reichman ( http://www.luminous-landscape.com/tutorials/expose-right.shtml ) lists a chart giving the number of tonal values per fstop. Based upon the zone system, the middle number of the chart, 512, would be used as Zone 5, if I were to meter on a subject. and the other tonal numbers would fall into their respective zones, either above or below. This means that if I want the subject I metered on to have the most detail I'd have to overexpose the given subject by 2 stops to put the subject in zone 7 which would be the 2048 tonal value range, note my 2 stop overexposure, and then in ACR or an equivolent program, adjust the exposure down two stops.

I disagree with the idea of metering on a highlight as he states at the end of the tutorial since the meter in all cameras work on the principle that the light that is analyzed is considered middle grey. Unless the meters in digital cameras are designed differently than the TTL meters of film bodies and the separate light meters one can buy separately (excluding incident light meters since those work differently than reflectance light meters which are the more common and built in meter). If they are different I'd like to know. By metering on a highlight with a meter that puts that highlight in zone 5, all the shadow detail would be thrown to zone 3 and probably lower depending on how great the difference is in luminosity of the shadows and highlights in the composition. You'd get underexposed photogrpahs and if you were to bump the exposure up in ACR, more noise would be generated in the final photo.

[Robert_Lay]

Thanks for the info, Bob. I'll try it when I get home. I'm assuming this will work for PSCS since that's what I have. At work I have PS 7 and I can't get it to switch to 16-bit display.
I am currently using PSCS2, but I am 99.9% sure that it was implemented in the same fashion on version 8.0

[Robert_Lay]

I was thinking more about the exposing to the right concept and the zone system.

Reichman ( http://www.luminous-landscape.com/tutorials/expose-right.shtml ) lists a chart giving the number of tonal values per fstop. Based upon the zone system, the middle number of the chart, 512, would be used as Zone 5, if I were to meter on a subject. and the other tonal numbers would fall into their respective zones, either above or below. This means that if I want the subject I metered on to have the most detail I'd have to overexpose the given subject by 2 stops to put the subject in zone 7 which would be the 2048 tonal value range, note my 2 stop overexposure, and then in ACR or an equivolent program, adjust the exposure down two stops.

The first error in that thinking comes from taking Reichman's 5 stop range too literally. He has not shown all of the stops of dynamic range - namely the sixth stop, the seventh stop, etc. It would be a misinterpretation of the Zone system to characterize it as having only 5 f-stops of range. Even if you throw out zones 0 and IX as being nothing more than bookends, you still have 8 f-stops or Zones - so, please don't think in terms of 5 f-stops of range - it just is not the way the zone system was designed.

Looking at it from the digital perspective and just for the sake of establishing that different authors have different treatments of this same subject, you will be surprised to learn that in Bruce Fraser's book on page 6 he shows a very similar illustration in which he clearly shows a sixth stop labeled "64" which is not shown at all in Reichman. If you were going to have to pick between Reichaman and Fraser, who would you choose?

Edit: The Canon EOS 10D has been characterized by an Imatest analysis which shows that it has a dynamic range of 8.97 stops. See the Norman Koren Web page on the Zone system at http://www.normankoren.com/zonesystem.html#Intro

So, in summary, your next step, which was to assume that metering system, which is supposed to be centered on Zone V, would be placing the metered subject at a level of 512 is in error. the Truth of the matter is that after the RAW range of values has been processed through the demosaicing algorithms and converted to JPG, the metering system has placed the metered subject on Zone V, but it's at 128 out of 256, or very close to that - not at 512 out of 4096. Remember, once the image has been converted from RAW to JPG, it has a range of 0 - 255 - not 0 - 4095! Therefore, all of your procedures and conclusions that bring you to a 2 stop overexposure are in error.



I disagree with the idea of metering on a highlight as he states at the end of the tutorial since the meter in all cameras work on the principle that the light that is analyzed is considered middle grey. Unless the meters in digital cameras are designed differently than the TTL meters of film bodies and the separate light meters one can buy separately (excluding incident light meters since those work differently than reflectance light meters which are the more common and built in meter). If they are different I'd like to know. By metering on a highlight with a meter that puts that highlight in zone 5, all the shadow detail would be thrown to zone 3 and probably lower depending on how great the difference is in luminosity of the shadows and highlights in the composition. You'd get underexposed photogrpahs and if you were to bump the exposure up in ACR, more noise would be generated in the final photo.


I believe you misunderstood what Reichman says about metering the highlight. He does not say to meter on the brightest white and use that exposure. What he actually says is "set your exposure such that the whitest white in the scene comes close to, but not quite reaching, the full digital scale (255 for 8-bit capture".

That is NOT the same thing as what you are saying, at all. What Reichman is saying is that you have to "place" the whites where they belong - at a digital value of 255, or as close to that as you can. If that takes an exposure correction of one or two f-stops to get it placed correctly, so be it. It's up to you and the scene and how you personally use the Zone System principles.

You are exactly correct about the meters in digital cameras - they do essentially what all other reflective light meters have been doing since the original Weston meter. Please re-read what he actually said at the end of the tutorial as I have quoted it above. It does not say what you are saying!

[PacAce]

I am currently using PSCS2, but I am 99.9% sure that it was implemented in the same fashion on version 8.0
I just checked and although there is a selection for 16-bit, it is grayed out so I can't pick it. Not sure why not, though. Will investigate further when I get a chance but I don't really need to have it since I'm content with the 8-bti displays. :)

[Robert_Lay]

I just checked and although there is a selection for 16-bit, it is grayed out so I can't pick it. Not sure why not, though. Will investigate further when I get a chance but I don't really need to have it since I'm content with the 8-bti displays. :)
Strange!
Are you using a Mac? Maybe it is a little different on the Mac. When I get that menu in Windows XP it offers 8, 16 and 32. I will have to see if I still have a version 8.0 installed somewhere and see if there is a difference.

What surprises me is that it still functions on my system even if I am looking at an 8 bit JPG image.

[PacAce]

Strange!
Are you using a Mac? Maybe it is a little different on the Mac. When I get that menu in Windows XP it offers 8, 16 and 32. I will have to see if I still have a version 8.0 installed somewhere and see if there is a difference.

What surprises me is that it still functions on my system even if I am looking at an 8 bit JPG image.
Bob, PSCS on both my Mac and my PC behave the same way. I only have 16-bit in the popup menu and it is grayed out. There are no 8-bit or 32-bit options.

However, I did a little more digging around and it turns out that in order to activate the 16-bit option in the popmenu display when the eyedroppers in the info palette are clicked, you have to first enable it by clicking on the triangle on the upper right hand corner of the info palette window, selecting "Palette Options..." and then checking the "Show 16 bit values" box at the bottom. Once that's enabled, you can toggle between 16-bit and 8-bit color displays via the popup menu.

Thanks again for bringing this feature to my attention. :)

[Robert_Lay]

Bob, PSCS on both my Mac and my PC behave the same way. I only have 16-bit in the popup menu and it is grayed out. There are no 8-bit or 32-bit options.

However, I did a little more digging around and it turns out that in order to activate the 16-bit option in the popmenu display when the eyedroppers in the info palette are clicked, you have to first enable it by clicking on the triangle on the upper right hand corner of the info palette window, selecting "Palette Options..." and then checking the "Show 16 bit values" box at the bottom. Once that's enabled, you can toggle between 16-bit and 8-bit color displays via the popup menu.

Thanks again for bringing this feature to my attention. :)
That sounds familiar - I think I hit that button without realizing it at some point. For anyone else following this, I did also find that there are significant differences between PSCS (ver8.0) and PSCS2(ver9.0), both in the functionality and in the user interface.
Version 8.0 uses a checkmark by "16 bit" in the menu to enable or disable it, while version 9.0 makes you pick from 8, 16 or 32 bits. In version 8.0 the 16 bit display of the numeric value only works when the image is in 16 bit mode, whereas in Version 9.0 it works with an image that is either 8 bit or 16 bit mode. I did not even try to find or make any 32 bit mode images.

[Poe]

The first error in that thinking comes from taking Reichman's 5 stop range too literally. He has not shown all of the stops of dynamic range - namely the sixth stop, the seventh stop, etc. It would be a misinterpretation of the Zone system to characterize it as having only 5 f-stops of range. Even if you throw out zones 0 and IX as being nothing more than bookends, you still have 8 f-stops or Zones - so, please don't think in terms of 5 f-stops of range - it just is not the way the zone system was designed.

Looking at it from the digital perspective and just for the sake of establishing that different authors have different treatments of this same subject, you will be surprised to learn that in Bruce Fraser's book on page 6 he shows a very similar illustration in which he clearly shows a sixth stop labeled "64" which is not shown at all in Reichman. If you were going to have to pick between Reichaman and Fraser, who would you choose?

So, in summary, your next step, which was to assume that metering system, which is supposed to be centered on Zone V, would be placing the metered subject at a level of 512 is in error. the Truth of the matter is that after the RAW range of values has been processed through the demosaicing algorithms and converted to JPG, the metering system has placed the metered subject on Zone V, but it's at 128 out of 256, or very close to that - not at 512 out of 4096. Remember, once the image has been converted from RAW to JPG, it has a range of 0 - 255 - not 0 - 4095! Therefore, all of your procedures and conclusions that bring you to a 2 stop overexposure are in error.


I believe you misunderstood what Reichman says about metering the highlight. He does not say to meter on the brightest white and use that exposure. What he actually says is "set your exposure such that the whitest white in the scene comes close to, but not quite reaching, the full digital scale (255 for 8-bit capture".

That is NOT the same thing as what you are saying, at all. What Reichman is saying is that you have to "place" the whites where they belong - at a digital value of 255, or as close to that as you can. If that takes an exposure correction of one or two f-stops to get it placed correctly, so be it. It's up to you and the scene and how you personally use the Zone System principles.

You are exactly correct about the meters in digital cameras - they do essentially what all other reflective light meters have been doing since the original Weston meter. Please re-read what he actually said at the end of the tutorial as I have quoted it above. It does not say what you are saying!

I used a 5 f-stop range due to the nature of photographic paper. This was from reading this website http://www.cicada.com/pub/photo/zs/quicktour/1.html. The site considers photographic paper to have a 5 fstop luminance range. Whether or not there have been advances to increase the range are unknown to me. Please clarify. Part of the brilliance that I found of the zone system is that you can decided the placement of highlights, shadows and midtones on your final product.

I did some more pondering after reading your reply, and I realize the error of my thinking. Middle grey would be tonal value 2048 (not sure of exact placement, but it falls just at the end of the 1024 region and the beginning of the 2048 section). I went ahead and wrote out the full geometrical progression of tonal values. I then realized that these are tonal values of luminosity and middle grey should be just that, in the middle. Thus it would be higher up and not at the 512th tone.

But I would still use a grey card and not a white card to get the correct exposure for middle grey, as metering on a white (tonal value greater than 2048) would cause a shift of the tones to the left and an underexposure of the image. If one had more room on the histogram to overexpose after a grey card exposure, then one should do that then adjust down appropriately in ACR to the grey card expsoure.

[Robert_Lay]

I used a 5 f-stop range due to the nature of photographic paper. This was from reading this website http://www.cicada.com/pub/photo/zs/quicktour/1.html. The site considers photographic paper to have a 5 fstop luminance range. Whether or not there have been advances to increase the range are unknown to me. Please clarify. Part of the brilliance that I found of the zone system is that you can decided the placement of highlights, shadows and midtones on your final product.

It is quite true that a photographic print has a more limited range of tones than can be supported by the photographic negative or diapositive materials. It is generally the case that transparency materials are better than reflective materials in this respect. Of course, the shortcomings of prints should not influence our efforts in capturing as great a range of tonality as possible in the original capture (more to say on that topic). There is a world of difference in photo printing papers. Some of the better papers use special coatings to enhance the reflectivity in the whites.

As you say, the "placement" of tones is a cornerstone of the Zone System, and that is exactly what Reichman was advocating when he says to expose to the right. He is saying that the brightest highlights of the scene should be rendered at that level which is the brightest white in the digital range - namely, the value 255. The value 255 is the highest numeric value possible in an 8 bit per channel range of digital brightness.


I did some more pondering after reading your reply, and I realize the error of my thinking. Middle grey would be tonal value 2048 (not sure of exact placement, but it falls just at the end of the 1024 region and the beginning of the 2048 section). I went ahead and wrote out the full geometrical progression of tonal values. I then realized that these are tonal values of luminosity and middle grey should be just that, in the middle. Thus it would be higher up and not at the 512th tone.

I'm afraid that the 12 bit range of 0 to 4095 is still confusing you. For that range of values, the middle gray is not at upper end of 1024 region and lower end of 2048 region. For example, if you choose to call the values from 0 through 7 as Zone 0 (not unrealistic), then

Zone I will be from 8 through 15;
Zone II will be from 16 through 31;
Zone III will be from 32 through 63;
Zone IV will be from 64 through 127;
Zone V will be from 128 through 255;
Zone VI will be from 256 through 511;
Zone VII will be from 512 through 1023;
Zone VIII will be from 1024 through 2047;
Zone IX will be from 2048 through 4095

As you can see, Middle Gray, or Zone V will be the range from 128 to 255. But that is arbitrary. We could just as easily say that we will have 11 Zones instead of 10 zones and that the first zone is the range 0 through 3. The point I am making here is that the mapping of brightness values that we choose to be represented by the digital values is arbitrary, just as when developing a negative we can alter the development time and produce a negative with either a long range of densities or a short range of densities. The only way that you will arrive at the true calibration of the digital levels versus the brightness levels for a given camera is to calibrate it, as some people have done. I made such a calibration for my Powershot G5 and published it here on this forum in July of 2005.

One very important point which I mentioned in my previous message must be reiterated. When you are shooting in JPG mode, the camera delivers an 8 bit per channel image, and that is a range of 0 to 255, not 0 to 4095. So, you have to completely re-write your table of digital values for the 8 bit world. Unfortunately, the range of 0 to 255 does not support the level of detail possible in the 12 bit world, so there is a definite compression of tonal range in the JPG image as compared to the 12 bit RAW image. I should also point out that by so doing, we have not changed the dynamic range of the camera - we have only changed the number of individual steps that we use to represent an f-stop of change in brightness.


But I would still use a grey card and not a white card to get the correct exposure for middle grey, as metering on a white (tonal value greater than 2048) would cause a shift of the tones to the left and an underexposure of the image. If one had more room on the histogram to overexpose after a grey card exposure, then one should do that then adjust down appropriately in ACR to the grey card expsoure.
The issue of whether you use a Gray card or a White card for metering was not mentioned in your previous message nor in mine. Apparently, you have interpreted Reichman's instructions as meaning that you should meter from a White card and call it middle gray. If that were what Reichman said, then he would be wrong, but he didn't say that. What he said was to arrange the exposure such that the whites appear at 255. In order to do what he says, you have to keep increasing the exposure until the whites fall at 255, which means that they should be starting to clip. A good part of this misunderstanding arose from that very simple misinterpretation of Reichman's directions.

[PhotosGuy]

Apparently, you have interpreted Reichman's instructions as meaning that you should meter from a White card and call it middle gray. If that were what Reichman said, then he would be wrong, but he didn't say that. What he said was to arrange the exposure such that the whites appear at 255. In order to do what he says, you have to keep increasing the exposure until the whites fall at 255, which means that they should be starting to clip. A good part of this misunderstanding arose from that very simple misinterpretation of Reichman's directions. For those who might have missed it, as in... Gray Card…White Paper. What’s best? (http://photography-on-the.net/forum/showthread.php?t=58677)

[Poe]

Now I'm more confused then I was to begin with.

[Robert_Lay]

Now I'm more confused then I was to begin with.
Sorry, I have tried my best.

Is it something I did, or are you just waiting for the world to stop spinning so you can get off :lol:

[Poe]

Epiphany:

Middle gray is a tonal value of 2047/2048 (which ever you prefer). The difference is too small to be noticed to the human eye.

A CMOS sensor can then detect 2047 levels darker and 2047 levels brighter. Each "level" is a intensity tone. It does not have 2048 levels devoted strictly to "brightest" tones and 1024 levels devoted to "bright" tones. In order to normalize such bright tones between the 2047/2048 to 4095 digital luminosity scale, with middle gray being the middle, 1024 bright/brightest tones would have to be compressed into the others. Similarly 1024 dark + darkest tones would have to be expanded to 2048.

By metering and exposing for middle gray, one is determining the change in voltage that indicates a change of one unit of intensity so that tones can be calculated, the image displayed on the lcd, a displayed histrogram, and the data making up the image on the CF card.

The reason why exposing to the right then correcting in post-processing works is because the signal to noise ratio is much higher with the bright tones than the dark tones. If there was zero noise in the sensor there would be no need to expose to the right and the correct exposure would work just the same. Post processing algorithms are much better at the interpolation of colors than using the data of a noisy sensor.

[PacAce]

Epiphany:

Middle gray is a tonal value of 2047/2048 (which ever you prefer). The difference is too small to be noticed to the human eye.

So far, so good. :)


A CMOS sensor can then detect 2047 levels darker and 2047 levels brighter. Each "level" is a intensity tone. It does not have 2048 levels devoted strictly to "brightest" tones and 1024 levels devoted to "bright" tones. .

A CMOS can detect 4096 levels for brightness, starting at a light level that is too low to register (0) and going up to a level that just barely saturates the CMOS (4095). The sensitivity of CMOS to light is linear so that the voltage generated by the CMOS when excited by light photons is directly proportional to the intensity of the light striking the CMOS.


In order to normalize such bright tones between the 2047/2048 to 4095 digital luminosity scale, with middle gray being the middle, 1024 bright/brightest tones would have to be compressed into the others. Similarly 1024 dark + darkest tones would have to be expanded to 2048

Our eyes do not respond to light in a linear fashion like the CMOS sensor does. The sensitivity of our eyes to light is logarithmic. Therefore, what the CMOS would consider midpoint in light intensity (at 2048 ) is to us just one stop less bright than the max brightness registered by the CMOS. That is the reason the data generated directly by the CMOS needs to be normalized for our viewing. So what ends up happening is that the data representing the bright data gets compresses towards the bright end of the normalized histogram and the data from the dark areas get expanded out.

Below is an illustration that tries to depict how linear data is normalized. The graph is the histogram of the linear data from the CMOS sensor. The bottom graph is the same data that's been normalized. The numbered areas show how the data is compressed (bright end) and expanded (dark end).

(Notice that the top two stops are compressed into the rightmost fifth of the normalized histogram and the darkest two are compressed into the leftmost fifth of the histogram. I gather this is due to the S-curve tone curve that's applied to the JPEG file when converted and normalized from RAW.)

65805

[Poe]

I think it has just hit me.

I'd just like to add one thing. With your example there of the linear data - it could go both ways, you could have your data either to the left or the right. If the data lies to the left, the shadows are inflated and the highlights are compressed. If the data lies to the right, the shadows are compressed and the highlights inflated (this is describing the process of going from linear data to normalized histogram). It makes me think that I should actually take two photographs of a subject, one underexposed and one over exposed, then piece the two together to get an exposure that would have a much greater range of information on the tones of the subject and not clip a subject's shadows or highlights.

[PacAce]

I think it has just hit me.

I'd just like to add one thing. With your example there of the linear data - it could go both ways, you could have your data either to the left or the right. If the data lies to the left, the shadows are inflated and the highlights are compressed. If the data lies to the right, the shadows are compressed and the highlights inflated (this is describing the process of going from linear data to normalized histogram). It makes me think that I should actually take two photographs of a subject, one underexposed and one over exposed, then piece the two together to get an exposure that would have a much greater range of information on the tones of the subject and not clip a subject's shadows or highlights.
And that's what exposing to the right is all about except that you don't want to underexpose because if you do, you'll need to increase the exposure in post processing and that will amplify the noise. What you want to do is overexpose, if possible, and then underexpose during post processing to get the most in noise reduction (or better S/N ratio). :)

[benca1]

And that's what exposing to the right is all about except that you don't want to underexpose because if you do, you'll need to increase the exposure in post processing and that will amplify the noise. What you want to do is overexpose, if possible, and then underexpose during post processing to get the most in noise reduction (or better S/N ratio). :)

Funny. After reading the entire thread, on the last post, when nearly all hope was lost for my feeble brain - AH HA! got it. Thanks much to everyone for the thoughtful posts.

[I Simonius]

Funny. After reading the entire thread, on the last post, when nearly all hope was lost for my feeble brain - AH HA! got it. Thanks much to everyone for the thoughtful posts.

Well my feeble brain is still struggling, I feel like Im in a hall of mirrors!

And just to thin k I read all this because I noticed my 5D seems to produce shots thaht look (to my film eye) about one stop overexposed

can someone answer me ever so simply - is that good or bad???? i.e. should I compensate or does it have an advantage being where it is exposurewise?
Im guessing I hope the answer is yes it is good as this is the cameras tendency to 'expose to the right' but I would have thought it would be the other way round that adarker looking image indicates more info in the highlights?



(whimper)

[Robert_Lay]

Well my feeble brain is still struggling, I feel like Im in a hall of mirrors!

And just to thin k I read all this because I noticed my 5D seems to produce shots thaht look (to my film eye) about one stop overexposed

can someone answer me ever so simply - is that good or bad???? i.e. should I compensate or does it have an advantage being where it is exposurewise?
Im guessing I hope the answer is yes it is good as this is the cameras tendency to 'expose to the right' but I would have thought it would be the other way round that adarker looking image indicates more info in the highlights?



(whimper)
In my opinion, it's bad to be one stop underexposed. I would prefer to be one stop Overexposed.

First, as has been pointed out quite clearly, there are no advantages in being underexposed, but there are several disadvantages.

Second, the reason that I advocate up to one stop overexposure (when shooting RAW), is that RAW processing can recover up to one stop of overexposure. Please note that I am saying "up to" one stop overexposure. It is better, of course, to have your highlights just barely hitting 255 (in the scale of 0 - 255).

[I Simonius]

In my opinion, it's bad to be one stop underexposed. I would prefer to be one stop Overexposed.

First, as has been pointed out quite clearly, there are no advantages in being underexposed, but there are several disadvantages.

Second, the reason that I advocate up to one stop overexposure (when shooting RAW), is that RAW processing can recover up to one stop of overexposure. Please note that I am saying "up to" one stop overexposure. It is better, of course, to have your highlights just barely hitting 255 (in the scale of 0 - 255).

Thanks, yes I get that, and I didn't mean to suggest I thought under exposing was good what I meant to say was that I thought accurate exposure ( by that I mean one that gives me correct midtones I guess) was better than over exposure

I guess Im not explaining myself very well

The example I had in mind was when recently I was shooting ascene with a lot of grass in, no highlights, virtually no sky, i.e. a landscape pointing downwards
The 5D exposed it so that it looked decidedly over exposed, I would guess by one stop approx

I tookseveral more of the scene to double check, essentially I was metering off the grass, but what I got instead of the expected mid tone grass was an overexposed grass


What I wondered was whether the 5D is set to expose this way, if so is it set that way to deliberately 'expose to the right' and is overexposing slightly equal to 'exposing to the right'?

[In2Photos]

...What I wondered was whether the 5D is set to expose this way, if so is it set that way to deliberately 'expose to the right' and is overexposing slightly equal to 'exposing to the right'?

Not sure about the 5D but have read several reports of other Canon cameras needing some =/- EC to compensate for over/under exposure.

I am sure you know this, but either EC or, if you shoot RAW with PS CS2, you can save the defaults to adjust the exposure by one stop when opened.

[I Simonius]

Not sure about the 5D but have read several reports of other Canon cameras needing some =/- EC to compensate for over/under exposure.

I am sure you know this, but either EC or, if you shoot RAW with PS CS2, you can save the defaults to adjust the exposure by one stop when opened.

yes thanks, that's right..,but I just wondered about this 'expose to the right' thing ( that I can't get my single grey cell around:lol: ), because IF my 5D over-exposing slightly is the same as 'exposing to the right' then I might not bother with the + or - EC and leave it where it is...???

[Robert_Lay]

Thanks, yes I get that, and I didn't mean to suggest I thought under exposing was good what I meant to say was that I thought accurate exposure ( by that I mean one that gives me correct midtones I guess) was better than over exposure

I guess Im not explaining myself very well

The example I had in mind was when recently I was shooting ascene with a lot of grass in, no highlights, virtually no sky, i.e. a landscape pointing downwards
The 5D exposed it so that it looked decidedly over exposed, I would guess by one stop approx

I tookseveral more of the scene to double check, essentially I was metering off the grass, but what I got instead of the expected mid tone grass was an overexposed grass


What I wondered was whether the 5D is set to expose this way, if so is it set that way to deliberately 'expose to the right' and is overexposing slightly equal to 'exposing to the right'?

Regarding your grass scene. A properly calibrated exposure system will integrate the entire scene, with whatever weighting factors are built into the exposure measuring function, and arrive at a setting that gives Zone V, or 127 on the scale of 0 - 255.

Whether that appears over, under or normally exposed to you is quite another matter. I would suggest looking at the histogram of that image and see whether or not it averages out to 127 (which it should do).

[rdenney]

We should remember that traditional exposure metering is not metering the range of brightness of the scene, but rather looking at the zones of the frame and seeking an exposure that will accommodate them all. This metering technology was developed in the days of film, when we could not look at a histogram. We can now see the brightness range in an image with far more precision than was previously possible.

To get the very most out of film, we did not use even the most sophisticated metering methods in cameras. We used a 1-degree spot meter and something like the Zone System, to make sure we measured all the parts of the scene important to us and placed them on the desired tone in a print. We manipulated technique to perform that placement, and sometimes we missed. Sometimes, we had to depend on a little film forgiveness, or we had to work a little harder to make full-range prints. In-camera meters have improved a lot and can get us close. But the best results always came when the tones were properly placed on the zonal scale, and that has always required measuring those tones specifically to make sure they were within our ability to get them on the zone we wanted.

We still have to do that. But the technique is a bit different now. Instead of measuring specific parts of the scene optically using a 1-degree spot meter, we can measure them statistically by looking at the histogram. A test shot (if we have time) might be more effective than spending the same amount of time pointing a spot meter. Thus, if the 5D puts the middle tones on the middle zones, then that's about the best that can be expected unless we select exposure based on spot metering. Or, we can note that the histogram is a bit too bunched up at an extreme and adjust accordingly.

Rick "who has put his spot meter away" Denney

[I Simonius]

We should remember that traditional exposure metering is not metering the range of brightness of the scene, but rather looking at the zones of the frame and seeking an exposure that will accommodate them all. This metering technology was developed in the days of film, when we could not look at a histogram. We can now see the brightness range in an image with far more precision than was previously possible.

To get the very most out of film, we did not use even the most sophisticated metering methods in cameras. We used a 1-degree spot meter and something like the Zone System, to make sure we measured all the parts of the scene important to us and placed them on the desired tone in a print. We manipulated technique to perform that placement, and sometimes we missed. Sometimes, we had to depend on a little film forgiveness, or we had to work a little harder to make full-range prints. In-camera meters have improved a lot and can get us close. But the best results always came when the tones were properly placed on the zonal scale, and that has always required measuring those tones specifically to make sure they were within our ability to get them on the zone we wanted.

We still have to do that. But the technique is a bit different now. Instead of measuring specific parts of the scene optically using a 1-degree spot meter, we can measure them statistically by looking at the histogram. A test shot (if we have time) might be more effective than spending the same amount of time pointing a spot meter. Thus, if the 5D puts the middle tones on the middle zones, then that's about the best that can be expected unless we select exposure based on spot metering. Or, we can note that the histogram is a bit too bunched up at an extreme and adjust accordingly.

Rick "who has put his spot meter away" Denney

good points
I am aware thet you have to be very careful with a spot metter esp when metering grass and other non solid objects as there may bepatches of tiny shadow or specular highlights tat you didnt immediately notice and which the brain ignores, so you can in fact get an off reading from a spot metering of grass etc

I still have to find time to do this justice( checking it out) but I was using the 35zone metering (or whatever it's called) when I took the example I originally refferred to and it looked way overexposed ( one stop estimated) on the LCD screen. Now it might have been the screen even I guess.

I really do need to stop thinking film and make digital the habit i.e. I need to learn to rely on histograms not LCDs (my eye)

the problem is I don't really fully understnd histos, but I guess like the zone system I'l just have to get my reluctant brain around it. I mean I know that you want it all nicely spread out in the middle, that the RHGS is the brightness and the LHS is the darkness but this exposing to the right business is something I still have to get my head around, one minute it makes sense, next it doesn't

Im amazed I ever learned even to drive!

Im going to read tis whole thread again and see if I can get my head around it

I need some super simple instructions! Im afraind talk of light being linear and of having this many levels and of proportionally excited photons is all jolly interesting but it's more than I can take on board to actually make sense of what I do with my camera

[RgB]

I'm curious to know how you determined the exposure to use for your first sample image. Was the shot properly exposed? The reason I'm asking is because the shot seems noisy and it looks like it might have been underexposed and corrected to get the right exposure.

PacAce isn't it true that large level adjustments really damage your photo creating lots of noise, i would assume that also goes for Exposure adjustments in PP.

[Robert_Lay]

good points
I am aware thet you have to be very careful with a spot metter esp when metering grass and other non solid objects as there may bepatches of tiny shadow or specular highlights tat you didnt immediately notice and which the brain ignores, so you can in fact get an off reading from a spot metering of grass etc

I still have to find time to do this justice( checking it out) but I was using the 35zone metering (or whatever it's called) when I took the example I originally refferred to and it looked way overexposed ( one stop estimated) on the LCD screen. Now it might have been the screen even I guess.

I really do need to stop thinking film and make digital the habit i.e. I need to learn to rely on histograms not LCDs (my eye)

the problem is I don't really fully understnd histos, but I guess like the zone system I'l just have to get my reluctant brain around it. I mean I know that you want it all nicely spread out in the middle, that the RHGS is the brightness and the LHS is the darkness but this exposing to the right business is something I still have to get my head around, one minute it makes sense, next it doesn't

Im amazed I ever learned even to drive!

Im going to read tis whole thread again and see if I can get my head around it

I need some super simple instructions! Im afraind talk of light being linear and of having this many levels and of proportionally excited photons is all jolly interesting but it's more than I can take on board to actually make sense of what I do with my camera

Here is something else to mull over...
With film the rule was to expose for the shadows and develop for the highlights.

Exposing for the shadows could be considered to be a good way to end up over-exposing. So, that's why the second clause was added.

With digital, exposing to the right is essentially saying the same thing, except that now the right hand border of the histogram is being used as your limiting factor. The only thing wrong with that is that it does not take full advantage of the fact that RAW mode allows you to recover up to one stop of overexposed highlights - thus extending the region into which you can "expose to the right". However, it would unwise to always try to use up that safety factor provided by RAW mode.

[PacAce]

PacAce isn't it true that large level adjustments really damage your photo creating lots of noise, i would assume that also goes for Exposure adjustments in PP.

The only time noise becomes a problem is if you underexpose and then try to compensate by increasing the brightness level during PP or, if shot in raw, increasing the exposure level during raw conversion. And the noise really becomes noticeable if the change is by a stop or more.

However, if the original shot was overexposed and then corrected in PP by toning down the brightness or decreasing the exposure, that actually decreases the noise level so that the adjust image looks a lot better than if the shot had been correctly exposed in the first place. Of course, if there are highlights that were blown, then nothing can be done to fix that.

[I Simonius]

Here is something else to mull over...
With film the rule was to expose for the shadows and develop for the highlights.

Exposing for the shadows could be considered to be a good way to end up over-exposing. So, that's why the second clause was added.

With digital, exposing to the right is essentially saying the same thing, except that now the right hand border of the histogram is being used as your limiting factor. The only thing wrong with that is that it does not take full advantage of the fact that RAW mode allows you to recover up to one stop of overexposed highlights - thus extending the region into which you can "expose to the right". However, it would unwise to always try to use up that safety factor provided by RAW mode.

Thanks BOb , yes I used agfapan 25 exclusively, for my personal work, and had to do exactly that as it tended to be pretty contrasty anyway

However I really need to get out there and jst look and test with the histogram thingy, the problem is for some reason I am having trouble getting my head round it

[I Simonius]

I just want to check something - in case I am making a huge assumption, - before I try to really get my head around it all

When using levels in photoshop, is the right and left side of the histogram exactly the same as the histogram on te camera?

[I Simonius]

Yes, you are correct. Shooting to the right does mean to set the exposure so the the data shows up as far right on the histogram as possible without blowing the highlights. And, as I mentioned in a previous post, this is most effective for dark or low-key scenes. However, after you have shot to the right, you just can't leave it that way or else the image will look over exposed. So, you'll need to post process the image to bring the exposre back in line with what it's supposed to be. In effect, in post processing, you will have to move the exposure to the left again.

So that would make sense of my pics of the metered grass then that looked over exposed, if the camera (5D) has been created to generally expose to the right?

It would mean that actually it IS the best thing for it to look slightly overexposed on the LCD?

[I Simonius]

Great job, Leo!

yes thanks Leo

So does ACR do the exposure correction OK or is it best to doing manually?

I;m abit wary of trying to do it myself with curves as I have no idea whether Im doing it right of completely wrong - I would much rather know that I can rely on ACR to do it automatically for me, or at least that I can do it myself without getting confused as ACR seems easy to use

[I Simonius]

While the conclusions of this thread are correct and will modify my own practice, I would like to state for the record that this statement on Reichmann's site is wrong.

He makes the point that the top stop of dynamic range contains half the tonal values. That is correct. What he forgets is that the histogram is not scaled in stops, but in those linear tonal values. So that top stop actually includes the right half of the histogram. Zones (in the Zone System sense) and stops follow a geometric rather than linear progression.

So, while it's important to make full use of that top stop of dynamic range, it's also important not to consign that stop to the right fifth of the histogram. Doing so will surely make blown-out highlights. Remember that we only know if highlights are blown out if the histogram is continuous right up to Zone X. If there is a gap, we might have a big hump of histogram off the right end of the scale and not even know it. In addition to looking at the histogram, we also need to turn on that highlight warning display so that we can see how much of the image is in the top bit.

Rick "who thinks the LCD images often lead us astray by blowing out more easily than the underlying data" Denney

it is all SLOWLY all starting to make sense...:D

[PacAce]

So that would make sense of my pics of the metered grass then that looked over exposed, if the camera (5D) has been created to generally expose to the right?

It would mean that actually it IS the best thing for it to look slightly overexposed on the LCD?

As long as you are not blowing the hightlight that are important to you, then there's nothing wrong with overexposing your shots as long as you go in afterwards to correct the exposure if you feel it is necessary. Of course, some images are overexposed on purpose without it being corrected later to achieve a certain artistic look. :)

[I Simonius]

I must have missed this post by Rick when he first posted it a while ago but I have to correct an error on his part. Altough he is correct in stating that the histogram is not scaled in full stops, it is very close except at the left and right ends of the histogram.

However, it is simply not correct to say that the right half of the histogram represents a top stop of the image. This is because the histogram is a representation of the image that has been converted to JPEG which has already been normalized, not the RAW image itself. If anybody would like to test this out, simply take a picture of a white or gray card and add plus EC in one stop increments to the exposure and watch the histogram as the graph moves towards the right edge of the scale.

It would be really good were this all to be written up into one article with all the questions and points raised and addressed.

In fact that might be BRILLIANT project for quite a few of the tuts on this , the best of all camera forii (plural of forum?):D

[PacAce]

yes thanks Leo

So does ACR do the exposure correction OK or is it best to doing manually?

I;m abit wary of trying to do it myself with curves as I have no idea whether Im doing it right of completely wrong - I would much rather know that I can rely on ACR to do it automatically for me, or at least that I can do it myself without getting confused as ACR seems easy to use

Manually correcting your exposure is not that big a deal. Usually you would just drag the Exposure slider to the left until the image looks correct to you. Letting ACR do it automatically is not usually going to give you the result you'll be looking for because all it's going to do is set the brightest point to the extreme right and the darkest point to the extreme left. And that's not what you want to do when correcting an overexposed image.

[I Simonius]

As long as you are not blowing the hightlight that are important to you, then there's nothing wrong with overexposing your shots as long as you go in afterwards to correct the exposure if you feel it is necessary. Of course, some images are overexposed on purpose without it being corrected later to achieve a certain artistic look. :)

yeah Im not going for the high key job

Normally I don't chimp but have taken to doing so lately to better understand what Im doing exposure wise be4cause I have got overconfident (from film days) and it don't always work the same with digital!

Certainly the shots of the grass were the first where I had noticed the apparent overexposure because I don't normally take shots of the grass but on this occasion had fired off a shot by mistake at the ground and just happened to notice it on the screen, then I took some more because it semmed overexposed , just to test it , and found misself befuddled thereof

[I Simonius]

Manually correcting your exposure is not that big a deal. Usually you would just drag the Exposure slider to the left until the image looks correct to you. Letting ACR do it automatically is not usually going to give you the result you'll be looking for because all it's going to do is set the brightest point to the extreme right and the darkest point to the extreme left. And that's not what you want to do when correcting an overexposed image.

But it is something that is done in ACR? That's OK I can handle ACR

What worried me was that it was something one had to do in PHOTOSHOP itself with curves ( or levels) and Im really not that sure that I understand what the effects are on dynamic range when I use those tools

[PacAce]

But it is something that is done in ACR? That's OK I can handle ACR

What worried me was that it was something one had to do in PHOTOSHOP itself with curves ( or levels) and Im really not that sure that I understand what the effects are on dynamic range when I use those tools

Yes, it's something that can and should be done in ACR if you can help it. AAMOF, the example I gave in my previous post re the use of the Exposure slider to correct the exposure was for ACR. What I was saying not to do was not to use the AUTO feature of ACR if you have PSCS2.

[rdenney]

Here is something else to mull over...
With film the rule was to expose for the shadows and develop for the highlights.


This rule was for negatives only. For transparencies, the opposite rule was relevant. Digital has always seemed to me more like transparencies than negatives, because with both transparencies and digital, once the highlights are blown they are gone forever. So, exposing the right has a hard limit at the point where highlights are blown.

Based on my experience with transparencies, I underexposed quite a lot of images on my 10D. The images were acceptable, and some even good by photographer standards, but this thread has shown me the error of my ways (I might have even said so somewhere in one of those previous pages, heh, heh).

Rick "who, like Simon, has had to learn not to be bound by rules that apply only to film" Denney

[I Simonius]

Yes, it's something that can and should be done in ACR if you can help it. AAMOF, the example I gave in my previous post re the use of the Exposure slider to correct the exposure was for ACR. What I was saying not to do was not to use the AUTO feature of ACR if you have PSCS2.

Thanks - but sometimes the auto feature gets it right???
Do you mean that the auto feature doesn't egt it right or that it somehow messes things up in another way?

But yes - OK I will do it manually!

I see the histogram changing when I aadjust things in ACR but don't really understand what it's doing, probably simply because it's in colour!


Simon "who like Rick got the hang of film and so probably will eventually of digital" King

[PacAce]

Thanks - but sometimes the auto feature gets it right???
Do you mean that the auto feature doesn't egt it right or that it somehow messes things up in another way?

But yes - OK I will do it manually!

I see the histogram changing when I aadjust things in ACR but don't really understand what it's doing, probably simply because it's in colour!


Simon "who like Rick got the hang of film and so probably will eventually of digital" King

No, I didn't mean that. There are times when the Auto feature is useful, but only for normally shot images where you have a spread of tones from dark to light. However, when an image has been overexposed on purpose and you need to correct that, Auto probably isn't going to do that for you very well because it's going to try to spread the tones in the picture across the whole histogram. But when you correct an overexposed image, what you want to do move the light tones of the image (those that are on the right side of the histogram) towards the middle of the histogram and this is oppose to what Auto would do.

[I Simonius]

No, I didn't mean that. There are times when the Auto feature is useful, but only for normally shot images where you have a spread of tones from dark to light. However, when an image has been overexposed on purpose and you need to correct that, Auto probably isn't going to do that for you very well because it's going to try to spread the tones in the picture across the whole histogram. But when you correct an overexposed image, what you want to do move the light tones of the image (those that are on the right side of the histogram) towards the middle of the histogram and this is oppose to what Auto would do.

I understand what you are saying, in principle, but whether I would know how to differentiate when actually using ACR is another matter

If there is any chance you'd have the time (and patience) to post visual examples again it would be super appreciated

Either way thanks for your explanations!

[PacAce]

I understand what you are saying, in principle, but whether I would know how to differentiate when actually using ACR is another matter

If there is any chance you'd have the time (and patience) to post visual examples again it would be super appreciated

Either way thanks for your explanations!
Sorry I missed this one, Simon. Can you refresh my memory again? What kind of visual examples were you looking for?

[Spatch]

Just to ressurect this fantastic thread. There has been a lot of talk about the normalisation of images. So 3 questions:

1) When you view the histogram on the back of the camera this is actaually the normalised JPEG based on the RAW image. The normalisation is based on a curve. Is this curve modified by the picture style settings or parameters currently set?

2) As 1) above but if you are shooting JPEG, is the saved JPEG normalised by the same in camera curve.

3) When viewing the RAW in ACR and the sliders are adjusted to get the result you want, does opening the file into PS normalise the picture based on the slider settings??

Thanks for any input.

[I Simonius]

Sorry I missed this one, Simon. Can you refresh my memory again? What kind of visual examples were you looking for?

erm.... sorry Ive lost the plot too!

(individually and severally):lol:

[Statement]

Am I to understand this correctly--that you can shoot on a higher iso with a noisier body like the 350d, but overexpose, and it will result in a noise free image?

[Robert_Lay]

Good Heavens - No!
First, there is no such thing as a noise-free image. Noise is inherent in all sensors.

Overexposing is not a means of reducing noise - avoiding under-exposure, however, is a method for minimizing noise, everything else being equal.

Exposing to the right should be a methodology for getting as close to overexposure as possible without over-doing it.

[RAitch]

Mind you, IMO... getting the right exposure produces the best result rather than just cramming data into the right half.
There is no right or wrong answer with exposure... but the less editing you have to do after the better. Colour rendition can be impacted to some degree as well.

You're right... over-exposure is probably worse than noise. You can always apply some noise reduction if it bothers you... but you can't re-invent data after it's blown out.

[E-K]

Mind you, IMO... getting the right exposure produces the best result rather than just cramming data into the right half.
There is no right or wrong answer with exposure... but the less editing you have to do after the better. Colour rendition can be impacted to some degree as well.

ETTR is simply about maximising the available dynamic range of the sensor. The sensor has a limit to the number of photons it can capture (saturation). The closer you can get to it without going over, the more information you have to work with. Making use of this information does take some effort of course :).

The main issue I see with ETTR, other than the risk of saturation, is that you need to increase the exposure which may mean shooting at a higher ISO. Shooting at a higher ISO does decrease the maximum available dynamic range so you need to make sure you have a net gain.

e-k

[RAitch]

I know what ETTR is all about and why people do it. The fact is, I've taken a couple example shots of a "proper" exposure and a ETTR exposure in the same place of the same subject... in each case, the "proper" exposure produced the best results even after PP.

Things like sunlight halos around tall grass and the detail within that came out way better in a correct exposure. Even after PP, I couldn't create the same clarity of tint tone and colour in the ETTR shot.

The key is getting everything exposed correctly and limiting the amount of PP after the fact.

Again, IMO, getting an image out of camera with the correct exposure is more important than cramming the right side with as much data as you can fit and fixing it later... it just doesn't work out better... in theory, it's a great idea though.

I suppose the true ETTR people would also drop their contrast settings way down as well. Just a thought.

[E-K]

...The key is getting everything exposed correctly and limiting the amount of PP after the fact.

Well I think the key differs depending on who you are and what you are trying to do :) Some people enjoy PP and trying to get every last ounce they can out of the sensor.

Again, IMO, getting an image out of camera with the correct exposure is more important than cramming the right side with as much data as you can fit and fixing it later... it just doesn't work out better... in theory, it's a great idea though.

Personally, a number of my shots are already pushing saturation when properly exposed so ETTR isn't really an option.

PP an ETTR image is certainly going to be more challenging than PP one that is properly exposed.

e-k

[RAitch]

Trust me, I enjoy post editing.

[malla1962]

Interesting post.:D

[thetolleyman]

Why can't this be done "in camera". Aren't these just little computers anyway? I would hoped the R&D department at Canon could have figured this out years ago and corrected for it, either across the board or as an optional "in camera mode", you know, a "corrected-overexposed mode".

Am I way off base here?

[I Simonius]

It is a cold, windy spring day here so looking out of my attic window it came to me that I should experiment with 'exposing to the right.' After all, what else is a chap to do?

Two pictures below are from a 100% crop of two images taken within a few seconds from my attic window to the small ex fishing village which is about 4 miles away. The camera is on a tripod and the lens is a Sigma 100-300 f4 EX with a Sigma 1.4 TC at full stretch.
The top image was exposed at 1/160th and the botttom at 1/40th - both at f13.
Images were processed in RawShooter identically, except for exposure correction. There is no clipping in the longer exposure's histogram. Limited corrections were made in Photoshop to levels + some sharpening and sizing. Identical corrections were made although both images would require different processing to get the best results, I suspect.

Others will know far more than I about that!

From this simple experiment, the differences in these pictures are quite marked with the bottom (exposed to the right) image looking as if it has been through some noise reduction software (it hasn't).

I'd be interested in any guidance/comments here about how the bottom image should be processed for best effect or any general views on this technique.

http://www.mayfieldghouse.freeserve.co.uk/Webpics/Exposetest.jpg
your pic not showing...

[RAitch]

Why can't this be done "in camera". Aren't these just little computers anyway? I would hoped the R&D department at Canon could have figured this out years ago and corrected for it, either across the board or as an optional "in camera mode", you know, a "corrected-overexposed mode".

Am I way off base here?
Um, yeah.
The exposure is based on too many things... and personally, I don't want a camera/computer to decide how an image is exposed.

I pick my metering mode, EC adjustments, or pay attention to where light falls on the meter so I can pick the settings to get the effect that I want.

Also, if your camera is going to correct over exposure... it's either going to have to darken everything or heavily drop the contrast most of the time. You'll end up with a lot of black that way.

In some cases, an overexposed image is very acceptable... the point is you're in control.
There is no right or wrong exposure... the idea is to duplicate the photographer's vision through the tools. If that requires exposing to the right and PP editing it back to avoid some digital noise... great! If it requires setting the exposure bang on and dealing with noise after (if you want)... great! If it requires 3 images blended into an HDR... great! If it involves the camera being the brain and having me reduced to an index finger... screw that! ;)

For what you're talking about, the 630 is your better choice. That's hands-off P&S technology... take what you get.

[bbc14f]

Here are a couple of examples of what shooting to the right does for you for low-key or dark subjects and scenes. All shots were taken at ISO 1600.

In the 1st attached image, A was shot using the correct exposure. B was shot 1 Stop under exposed. And C was shot to the right so that it was really over exposed as you can see from the bottom left image. The exposure for C was then corrected during RAW conversion processing. The image on the bottom right is the corrected image of C.

The 2nd attached image shows a 100% crop of A (the correctly exposed image), B (the under exposed image) after it was corrected in PSCS for the correct exposure, and C (shot to the right and corrected). As you can see high ISO noise is marginal in A, really bad in B and practically non-existent in C. Check out how silky smooth the background of C is.
.

I really think this post does a lot to explain this concept to me however I just need some answers to wrap it all up.

First what metering mode did you use. Second what did you meter off of, specific spot on the Budda, purple, gray wall etc? I guess the exif data would really be great also.

I know this was done a while ago so I understand if you don't have the exact data, but any info would be greatly appreciated.

[PacAce]

I really think this post does a lot to explain this concept to me however I just need some answers to wrap it all up.

First what metering mode did you use. Second what did you meter off of, specific spot on the Budda, purple, gray wall etc? I guess the exif data would really be great also.

I know this was done a while ago so I understand if you don't have the exact data, but any info would be greatly appreciated.

If you are going to be exposing to the right, it really doesn't matter what metering mode you use because, basically, what you are going to be doing is looking at your histogram and adjusting the exposure (increasing it) until the histograms is as far right as possible without blowing out any essential details. But more so in my sample pictures because they were shot using a flash (ambient light was very low and really didn't affect the image at all). :)

Since I was using a flash, the camera was set to manual mode with ISO 1600, and I adjusted the flash FEC to get the exposure I wanted. For the "normal" picture (_E2U1536), FEC was set to -2/3 because the Buddah is very dark and a normal flash exposure would tend to overexpose the image.

The "shot to the right" image (_E2U1539)was shot with FEC +1 1/3. This gave me a histogram that went all the way to the right without touching or going over the right edge.

As an FYI, I just want to point out that the sample pictures I used to demo "shooting to the right" was just that, a demo. In most cases, if you are using a very high ISO and there is enough lattitude to expose to the right, especially for a low key subject (which is what works best for shooting to the right), then that also means that you can probably lower your ISO and shoot for the correct exposure and get the same or better results. So, in essence, you get the most out of shooting to the right when shooting a low key subject, using ISO 100 and shooting with a flash or strobe.

When shooting with ambient lighting, however, your shutter speed and/or aperture will usualy dictate what ISO you need to use to get the shot you are after. In that case, you probably want to use the lowest ISO that will give you the correct exposure using the selected shutter speed and aperture. To me, it makes no sense to bump up the ISO just so you can shoot to the right (and even risking blowing highlight details that aren't detectable on the histogram or LCD screen) because that is essentially equivalent to shooting at a lower ISO and exposing correctly.

:)

[bbc14f]

If you are going to be exposing to the right, it really doesn't matter what metering mode you use because, basically, what you are going to be doing is looking at your histogram and adjusting the exposure (increasing it) until the histograms is as far right as possible without blowing out any essential details. But more so in my sample pictures because they were shot using a flash (ambient light was very low and really didn't affect the image at all). :)

Since I was using a flash, the camera was set to manual mode with ISO 1600, and I adjusted the flash FEC to get the exposure I wanted. For the "normal" picture (_E2U1536), FEC was set to -2/3 because the Buddah is very dark and a normal flash exposure would tend to overexpose the image.

The "shot to the right" image (_E2U1539)was shot with FEC +1 1/3. This gave me a histogram that went all the way to the right without touching or going over the right edge.

As an FYI, I just want to point out that the sample pictures I used to demo "shooting to the right" was just that, a demo. In most cases, if you are using a very high ISO and there is enough lattitude to expose to the right, especially for a low key subject (which is what works best for shooting to the right), then that also means that you can probably lower your ISO and shoot for the correct exposure and get the same or better results. So, in essence, you get the most out of shooting to the right when shooting a low key subject, using ISO 100 and shooting with a flash or strobe.

When shooting with ambient lighting, however, your shutter speed and/or aperture will usualy dictate what ISO you need to use to get the shot you are after. In that case, you probably want to use the lowest ISO that will give you the correct exposure using the selected shutter speed and aperture. To me, it makes no sense to bump up the ISO just so you can shoot to the right (and even risking blowing highlight details that aren't detectable on the histogram or LCD screen) because that is essentially equivalent to shooting at a lower ISO and exposing correctly.

:)

Great thats what I thought about the metering concerns. Now I have one more dilema.
I pretty much know what to do to get the exposure to the right side of the histogram; adjust aperture and shutter speed, etc. But how do I affect the hight of the histogram. This usually only occurs when there aren't that many colors in the shot and I was thinking that was the reason but after reading through this post I'm thinking I might be able to adjust this by tweaking my exposure. I'm guessing a wider aperture to let more light in..??? Could that be correct?

[PacAce]

Great thats what I thought about the metering concerns. Now I have one more dilema.
I pretty much know what to do to get the exposure to the right side of the histogram; adjust aperture and shutter speed, etc. But how do I affect the hight of the histogram. This usually only occurs when there aren't that many colors in the shot and I was thinking that was the reason but after reading through this post I'm thinking I might be able to adjust this by tweaking my exposure. I'm guessing a wider aperture to let more light in..??? Could that be correct?

I just answered that very same question a day ago so instead of me retyping it, let me just point you to that thread:

http://www.photography-on-the.net/forum/showthread.php?t=251685

:)

[bbc14f]

I just answered that very same question a day ago so instead of me retyping it, let me just point you to that thread:

http://www.photography-on-the.net/forum/showthread.php?t=251685

:)

Thanks a billion you've been a great help!

[Titus213]

Now that was a good read. Chock full of useful information - thank you all.

[Robert_Lay]

...
I pretty much know what to do to get the exposure to the right side of the histogram; adjust aperture and shutter speed, etc. But how do I affect the hight of the histogram.

I don't think you really intend to affect the height of the histogram, because the histogram (in PSCS) is always "full scale". That is, whatever tonal value has the highest population will be re-scaled to reach the top of the histogram.

I think that what you are really asking is how to change the "shape" or contour of the histogram curve. The shape can be changed (in simplistic terms) by a re-distribution of tonal values. But that re-distribution can be brought about by any number of different tools that distort the relationships and values between individual pixels.

One thing that I do not believe can be done at exposure time (without filters) is to change the shape of the histogram with aperture and shutter (i.e., by just changing the exposure).

[rdenney]

One thing that I do not believe can be done at exposure time (without filters) is to change the shape of the histogram with aperture and shutter (i.e., by just changing the exposure).

That is one of the main differences between digital and film photography. Digital sensors have much more linear response, which means that as you move the histogram left and right with exposure, its shape doesn't change.

But film has a characteristic S-shaped response curve, and that will change tonal relationships with different exposures. That's why exposure to achieve specific tonal results (using something like the Zone System) required so much testing and commitment. And that's why we worried about 1/4-stop accuracy when shooting narrow transparency films like Velvia. With digital, we can review the histogram on the ground, and that's a huge benefit. I used to use Polaroid in my view camera to check relationships, but that requiree a complex mental translation between Polaroid and regular film.

That's also why I find myself adjusting images out of my Canon with an S-chaped tone curve--it restores some of the characteristic film look that is still programmed into my aesthetic sensibilities.

Rick "who thinks film photographers should appreciate digital's linearity and convenience most" Denney

[Robert_Lay]

That is one of the main differences between digital and film photography. Digital sensors have much more linear response, which means that as you move the histogram left and right with exposure, its shape doesn't change.

But film has a characteristic S-shaped response curve, and that will change tonal relationships with different exposures. That's why exposure to achieve specific tonal results (using something like the Zone System) required so much testing and commitment. And that's why we worried about 1/4-stop accuracy when shooting narrow transparency films like Velvia. With digital, we can review the histogram on the ground, and that's a huge benefit. I used to use Polaroid in my view camera to check relationships, but that requiree a complex mental translation between Polaroid and regular film.

That's also why I find myself adjusting images out of my Canon with an S-chaped tone curve--it restores some of the characteristic film look that is still programmed into my aesthetic sensibilities.

Rick "who thinks film photographers should appreciate digital's linearity and convenience most" Denney
Now, if they would only allow us the ability to tweak the demosaicing algorithms that are used to convert the linear response of the sensor to its logarithmic output! Wouldn't that give us a tool to play with? It would be like having the characteristics of the film and the developer all at your fingertips.

[sjafari]

I am definately going to have to try shooting like this in the future. Thanks everybody for the awesome tip and explanations!

[PacAce]

Now, if they would only allow us the ability to tweak the demosaicing algorithms that are used to convert the linear response of the sensor to its logarithmic output! Wouldn't that give us a tool to play with? It would be like having the characteristics of the film and the developer all at your fingertips.
But such a tool already is available to us in the form of DPP, ACR, PS, etc, especially if you shoot raw, although I'm not sure what the demosaicing algorith has to do with the tone curve.

When you shoot JPEG, the S-curve tone is already applied for you in-camera. The shape of that curse will of course depend on the parameter settings you have selected, especially if the camera is a more recent model that incorporated picture styles.

If you shot raw, however, there are several other options available to you. You can process the raw file as neutrally as you can with the raw converter and then apply your own tone curve to it in the raw converter or after exporting to an image editor like Photoshop. BTW, this option is also applicable when shooting JPEG.

Or, you can convert the raw file to linear and then apply your own tone curve in post processing.

[tzalman]

When you shoot JPEG, the S-curve tone is already applied for you in-camera. The shape of that curse will of course depend on the parameter settings you have selected...

I understand that you are not fond of jpg, Leo, but that seems a bit harsh.:lol:

[tzalman]

But such a tool already is available to us in the form of DPP, ACR, PS, etc, especially if you shoot raw, although I'm not sure what the demosaicing algorith has to do with the tone curve.


Several years ago when I had a Minolta camera, I used a free RAW converter which did in fact allow user interface with the Variable Number of Varients algorithm. Since the demosaicer is deciding on two new color/tonal values for each pixel the parameters by which the interpolation is made can indeed affect tonal characteristics, but moreso it affects smoothness, noise, detail extraction and sharpness. I suspect that converters that allow control of these characteristics are in fact doing it by altering the algorithm.

[PacAce]

I understand that you are not fond of jpg, Leo, but that seems a bit harsh.:lol:

:confused: Not sure I get the punch line here!? :confused: But then, I've always been a bit slow execpt with the most obvious of punch lines. :lol:

[PacAce]

Several years ago when I had a Minolta camera, I used a free RAW converter which did in fact allow user interface with the Variable Number of Varients algorithm. Since the demosaicer is deciding on two new color/tonal values for each pixel the parameters by which the interpolation is made can indeed affect tonal characteristics, but moreso it affects smoothness, noise, detail extraction and sharpness. I suspect that converters that allow control of these characteristics are in fact doing it by altering the algorithm.

Oh, OK, I see now. (I was under the impression that the demosaic algorithm just interpolated pixels to produce the missing colors values and then passed the "reconstructed" image on to the image processing algorithm. But maybe I was looking at it too much at the macro level.) :)

[E-K]

Several years ago when I had a Minolta camera, I used a free RAW converter which did in fact allow user interface with the Variable Number of Varients algorithm. Since the demosaicer is deciding on two new color/tonal values for each pixel the parameters by which the interpolation is made can indeed affect tonal characteristics, but moreso it affects smoothness, noise, detail extraction and sharpness. I suspect that converters that allow control of these characteristics are in fact doing it by altering the algorithm.

Out of curiousity are you talking about dcRAW? In any case, I think you will find that converters allowing you to control the noise and sharpness do so after the demosaicing algorithm has been applied.

I do agree it's possible, I just think implementation wise it's much easier to do after interpolation. Some demosaicing algorithms perform better than others though. For example, I've been experimenting with dcRAW (UFRaw) again and I am pleased with what the AHD algorithm can do as far as keeping the detail without visible artefacts...well at least compared to the VNG algorithm ;)

e-k

[tzalman]

The converter was called RawDeal and was authored by an Australian who called himself Vector. It offered a choice of algorithms and if you selected VNG you could open a sub-dialogue where 3 parameters could be set. I think I remember that smoothness and radius where two of them, but I don't remember the third.
It's possible that some of his code was taken from dcRAW.

[E-K]

The converter was called RawDeal and was authored by an Australian who called himself Vector. It offered a choice of algorithms and if you selected VNG you could open a sub-dialogue where 3 parameters could be set. I think I remember that smoothness and radius where two of them, but I don't remember the third.
It's possible that some of his code was taken from dcRAW.

Thanks. It was the settings for the VNG Modified algorithm. There are four parameters according to the documentation - smoothness, sensitivity, K1 and K2. K1 and K2 are coefficients in the standard VNG algorithm and I believe have a standard value of 1.5 and 0.5 respectively. K1 and K2 are used in determining the threshold for selecting gradients.

The other two I'm not really clear on what they do but it sounds like it is specific to that modified algorithm.

e-k

[wannabegood]

Bibble Labs has RAW conversion that is , at least to me, much easier and more thorough than photoshop. The conversation here is very enlightening as to histograms, albeit deeper than this homeboy needs to go. Play with the camera. Find what works. Go with it! :) Sometimes it seems I can overthink something and kill it, instead of artfully creating it! Does that make sense?
Who was it that said in this thread they're having trouble wrapping their single grey cell around all this? I agree, wholeheartedly. Good luck deciphering, happy shooting!

[S.Horton]

Wow.

Link saved.

Most impressive, and greatly appreciated, Leo!

[Robert_Lay]

Both Leo and Rick Denney have already commented on the issue of linearity in regard to digital camera response curves, but I want to add some additional observations that relate specifically to "exposing to the right".

A good reason for being especially careful in "exposing to the right" is the fact that the digital cameras seem to have an unusually abrupt shoulder in the response curve in comparison with most negative films. In other words, there is very, very little head-room or "cushion" in the shoulder at exposure levels from Zone VIII + upwards, whereas with film the shoulder is still quite soft right up to Zone X or XI.

As a consequence, "exposing to the right" is quite risky unless you are shooting RAW. In other words, if you do get too close to the abrupt shoulder in the response curve, it will be difficult to recover the blown out highlights unless you are shooting RAW.

[E-K]

Even when shooting RAW you cannot technically recover a blown highlight. You can algorithmically attempt to guess at the correct values for clipped channels but it will always be a guess.

e-k

[RAitch]

Since RAW captures more data, you can quite easily recover what your camera would convert to JPEG as "blown highlights." If you're just over that edge, there is absolutely no problem. If you're really overexposed, that's a different story.

Remember, RAW is a 16-bit format that allows you to store the (roughly) 12-bits of image data. JPEG only uses 8-bits of data.
This extra data allows you to simulate a larger dynamic range in a single image. I often use the same RAW file with 2 exposure values to help remove highlights or get the best out of every area in the image.

I try to expose correctly (for what I want) in RAW... instead of cornering myself into the "rule" of exposing to the right.

[Robert_Lay]

Even when shooting RAW you cannot technically recover a blown highlight. You can algorithmically attempt to guess at the correct values for clipped channels but it will always be a guess.

e-k

Sorry, but I believe that is incorrect. My measurements, which have been confirmed many times over in practice, make it clear that the information is not lost - it is simply stored in such a way as to be of value only in the RAW shooting mode. The only issue is "how much" is stored. My measurements indicated that it is approximately one full stop on the Powershot G5. I have not yet measured the Rebel XTi.

If you read the following article, you will find some specifics on this topic.
Quick Tour of RAW Processing as downloadable PDF:
http://www.zaffora.com/W9DMK/QuickTourOfRAWProcessing.pdf

[PhotosGuy]

Just a reminder & I may have said this earlier, but every highlight may not be one that you need to keep, so use that "single grey cell" to decide what is most important in each shot. With cars, I prefer to start with a white "white", & let overly bright reflections blow out.

Need an exposure crutch? (http://photography-on-the.net/forum/showthread.php?t=89123)

[E-K]

Sorry, but I believe that is incorrect. My measurements, which have been confirmed many times over in practice, make it clear that the information is not lost - it is simply stored in such a way as to be of value only in the RAW shooting mode. The only issue is "how much" is stored. My measurements indicated that it is approximately one full stop on the Powershot G5. I have not yet measured the Rebel XTi.

If you read the following article, you will find some specifics on this topic.
Quick Tour of RAW Processing as downloadable PDF:
http://www.zaffora.com/W9DMK/QuickTourOfRAWProcessing.pdf

Hi Robert,

I think we're just talking about two different things. When I talk about clipping I mean that one or more channels are saturated in the sensor data. There is simply no way to recover this data since it has been pegged at the maximum value that can be digitised. So assume a pixel is completely saturated such that all channels are pegged at 0xFFF. I cannot reduce the exposure and recover the highlights in this case.

The document you reference seems to refer to this as hard clipping (although I could not find a concise definition for either hard or soft clipping in the document). They imply soft clipping is the result of someone increasing exposure in the RAW conversion. In this case you can compress the data to maintain the highlights rather than having it just become pure white.

RAW for sure has more lattitude than JPEG and I would not use JPEG for ETTR but if you end up saturating a sensor well when shooting ETTR then there is not much you can do to recover the detail in those areas - they will simply come out gray (or an odd colour if only one or two channels reached saturation) when you correct the exposure down.

So to summarise, the hard shoulder of digital is due to the physical nature of the sensor - it's linear up until saturation. JPEG dumps 4 bits of information and applies a curve in camera which could result in additional "soft clipping" which is subsequently not recoverable. RAW maintains the full 12 bits and depending on your RAW convertor you can adjust this data however you see fit (i.e. you can minimise any form of soft clipping). Saturated channels are not recoverable no matter whether you shoot RAW or JPEG - okay theoretically you can guess at some of values based on an algorithm which looks at nearest neighbour unsaturated pixels to determine the relative intensities but I don't know of any software which does this.

e-k

[E-K]

Since RAW captures more data, you can quite easily recover what your camera would convert to JPEG as "blown highlights." If you're just over that edge, there is absolutely no problem. If you're really overexposed, that's a different story.
...


Correct. As I responded to Robert I think we are just talking about two different things. I rarely shoot JPEG now so to me when a highlight is blown that is equivalent to clipping in one or more channels and this is not recoverable (or at least not easily recoverable without technology from CSI :)).

I would agree though that if you shot both JPEG and RAW then you should be able to recover some of the highlights that were lost in the JPEG.

ETTR and JPEG just seems nonsensical to me. You are doing it to capture the maximum amount of information and then throw 4 bits of it away? :)

As you though I tend to expose correctly rather than to the right mainly because in most of my shots I already have some channel saturation (or I'm very close).

e-k

[Geo]

Very interesting and instructional.

[Robert_Lay]

Hi Robert,

I think we're just talking about two different things. When I talk about clipping I mean that one or more channels are saturated in the sensor data. There is simply no way to recover this data since it has been pegged at the maximum value that can be digitised. So assume a pixel is completely saturated such that all channels are pegged at 0xFFF. I cannot reduce the exposure and recover the highlights in this case.


I agree. That's exactly why I created the term "hard clipping" in that article - to distinguish between the loss of detail information in the electronics and the "hiding" of information in the RAW format.


The document you reference seems to refer to this as hard clipping (although I could not find a concise definition for either hard or soft clipping in the document). They imply soft clipping is the result of someone increasing exposure in the RAW conversion. In this case you can compress the data to maintain the highlights rather than having it just become pure white.

One reason for a lack of a concise definition is that I created that terminology strictly for purposes of that paper just so I could distinguish between what was being done in the hardware and what was being done in the demosaicing algorithms and/or in the conversion to JPG.

RAW for sure has more lattitude than JPEG and I would not use JPEG for ETTR but if you end up saturating a sensor well when shooting ETTR then there is not much you can do to recover the detail in those areas - they will simply come out gray (or an odd colour if only one or two channels reached saturation) when you correct the exposure down.

So to summarise, the hard shoulder of digital is due to the physical nature of the sensor - it's linear up until saturation. JPEG dumps 4 bits of information and applies a curve in camera which could result in additional "soft clipping" which is subsequently not recoverable. RAW maintains the full 12 bits and depending on your RAW convertor you can adjust this data however you see fit (i.e. you can minimise any form of soft clipping). Saturated channels are not recoverable no matter whether you shoot RAW or JPEG - okay theoretically you can guess at some of values based on an algorithm which looks at nearest neighbour unsaturated pixels to determine the relative intensities but I don't know of any software which does this.

e-k

Thanks! Your comments are sincerely appreciated.

[RAitch]

I guess what I was trying to say is to blow out a channel in RAW... you have to be pretty far over that right edge. With JPEG, it's zero-tolerance... but there is quite a bit of room to play with in RAW.
Why shoot RAW+JPEG?? That confuses me... if you're shooting RAW, what the heck are you going to do with the JPEG? I don't know.
I wouldn't use the RAW file to recover data in the JPEG (if that's what you were implying)... just stick with the RAW file... it's all you need (that, AND a VISA).

I guess the main point to remember is that when shooting RAW, the point at which channels start getting clipped is much farther right compared to JPEG results.
Even RAW files that are blinking (blown) on your display are easily recovered with a brightness or exposure adjustment in your RAW converter.

As mentioned before, I often take multiple exposures from the same RAW file and blend them together (to remove hot spots).
If I take an outdoor portrait at high-noon and there is mixed sunlight and shadow on somebody's face, I'll take a normal exposure (with probably some blowouts in PS) then load the same RAW file again, drop the exposure/brightness values down a bit, and load another copy (not adjusting RAW conversion settings) into PS.
Then, with layer masks, I'll blend these images together to bring back detail in those blown out areas.
They're only blown out (in PS) because the RAW file is converted to a working JPEG (much like taking JPEGs in camera).

ETTR can be bad advice... when used with words like "always." If you always ETTR, you're not getting the best exposure of your image. Things will suffer (like colour saturation) that you will not be able to recover in post processing. Images always have better results with starting from the best exposure... rarely that's from ETTR.
I love playing with PS (viewing my gallery would confirm that) but I don't always expose a particular way. Much like Frank, I like to pick what's most important in an image, and expose that correctly.

If you're shooting JPEG, you can also play with your contrast settings to help with your clipping. For example, in bright sunlight, contrast is very high. It's near impossible to fit dark shadows and bright light in the same range. To help you out, drop your in camera contrast WAY down. That'll bring your histogram tighter together and reduce the difference between white and black.
In the same thinking, if you're in a snow storm (where everything looks like a bright shade of grey) then you can crank up the contrast to spread your histogram out and provide you with more distinguishing contrast.
If you don't, you'll have a narrow little bump in your histogram because everything falls within the same range. Expanding this helps you isolate shades in PP... otherwise your entire image will look very flat.

Interesting discussion...

[E-K]

I guess what I was trying to say is to blow out a channel in RAW... you have to be pretty far over that right edge. With JPEG, it's zero-tolerance... but there is quite a bit of room to play with in RAW.

I understand what you're saying. You gain about an extra stop lattitude with RAW over JPEG. My initial concern was over Robert's statement that:

As a consequence, "exposing to the right" is quite risky unless you are shooting RAW. In other words, if you do get too close to the abrupt shoulder in the response curve, it will be difficult to recover the blown out highlights unless you are shooting RAW.

Which when I first read it took to mean if you get to the saturation point of the sensor (abrupt shoulder) then it will be difficult to recover the highlights unless you're shooting RAW. My point was that if you reach sensor saturation you cannot recover the highlights whether it's JPEG or RAW.

Why shoot RAW+JPEG?? That confuses me... if you're shooting RAW, what the heck are you going to do with the JPEG? I don't know.
I wouldn't use the RAW file to recover data in the JPEG (if that's what you were implying)... just stick with the RAW file... it's all you need (that, AND a VISA).

It was for the point of comparison. That is to say, ASSUME you shot a scene with the same exposure in both RAW and JPEG, then you WOULD be able to recover more highlights from the RAW than the JPEG. I would never shoot JPEG if doing ETTR (which I said in the following paragraph of that post) because the point of ETTR is to maximise the amount of data and with JPEG you're already throwing 4 bits away.

I don't think we are disagreeing here - it was just a difference in usage of terminology and the object of discussion (i.e. when the highlights were blown where they blown in the RAW data or in the JPEG).

One other note is that the "blinkies" are based off of the internal JPEG (at least with the 350D) and so the reality is that you would need to back off the exposure until you don't get them and then make a judgement if you want to increase the exposure a little.

e-k

[Raymate]

This is rather interesting. Must try some test myself :)

[Robert_Lay]

Sorry, but I believe that is incorrect. My measurements, which have been confirmed many times over in practice, make it clear that the information is not lost - it is simply stored in such a way as to be of value only in the RAW shooting mode. The only issue is "how much" is stored. My measurements indicated that it is approximately one full stop on the Powershot G5. I have not yet measured the Rebel XTi.

If you read the following article, you will find some specifics on this topic.
Quick Tour of RAW Processing as downloadable PDF:
http://www.zaffora.com/W9DMK/QuickTourOfRAWProcessing.pdf

Since the above was posted, I have measured the Rebel XTi, and my tests indicate that the Rebel XTi also stores one full stop - the same as I measured in the G5.

[CanonLaw]

This has got to be the coolest thing I have learned in a long time!!!! Thank you!!! I have been doing this for years and never knew this! Thank you!!!! It is amazing!

[PeacePlanet]

Here are a couple of examples of what shooting to the right does for you for low-key or dark subjects and scenes. All shots were taken at ISO 1600.

In the 1st attached image, A was shot using the correct exposure. B was shot 1 Stop under exposed. And C was shot to the right so that it was really over exposed as you can see from the bottom left image. The exposure for C was then corrected during RAW conversion processing. The image on the bottom right is the corrected image of C.

The 2nd attached image shows a 100% crop of A (the correctly exposed image), B (the under exposed image) after it was corrected in PSCS for the correct exposure, and C (shot to the right and corrected). As you can see high ISO noise is marginal in A, really bad in B and practically non-existent in C. Check out how silky smooth the background of C is.
.

Is it correct assume that shoot to the right gives u better noise control ???

thanks

AL

[Robert_Lay]

Is it correct assume that shoot to the right gives u better noise control ???

thanks

AL

That topic has been discussed quite thoroughly some time ago.

As I remember, the consensus was as follows:
Expose to the right, even though you may have to boost the ISO to maximum in order to do so. Any noise problems created in so doing are easily mitigated with either of the two well known free noise reduction programs.

[RAitch]

I'd suggest finding the correct exposure and go with that.
You wouldn't want to expose to the right for a low key shot... then worry about bringing it all back down... to avoid digital noise (just an example).

[Robert_Lay]

I'd suggest finding the correct exposure and go with that.
You wouldn't want to expose to the right for a low key shot... then worry about bringing it all back down... to avoid digital noise (just an example).


Why not?

It seems that the major point of this entire thread is being ignored.

Any time you do not expose to the right, you are throwing away tons of information that could be applied to your final product.

[E-K]

Is it correct assume that shoot to the right gives u better noise control ???

thanks

AL

Yes and no. Exposing to the right is attempting to maximise your signal to noise ratio. Assuming that the same ISO and aperture is used (i.e. you are just increasing the exposure time) then the noise will be better when the ETTR exposure is corrected.

Increasing ISO so that you can expose to the right things gets a little fuzzy. Increasing ISO results (generally) in a corresponding decrease in the maximum possible S/N ratio and available dynamic range. In this case you have to determine whether the increase in S/N ratio by exposing to the right will offset the reduction in S/N because of an increase in ISO. It is likely in this case that you may see an improvement in the shadows but an overall reduction in quality.

I'd suggest finding the correct exposure and go with that.
You wouldn't want to expose to the right for a low key shot... then worry about bringing it all back down... to avoid digital noise (just an example).

I would think using ETTR for a low key shot would be a good thing since you improve noise in the shadows? I shoot RAW almost exclusively so taking an extra second to adjust the exposure scale is no big deal.

e-k

[RAitch]

If you want to shoot for Photoshop, go nuts. IMO Nothing beats an image that's properly exposed even taking into consideration digital noise and post adjustments. The properly exposed image will come out with better contrast and saturation in the end.
I love to post process... but always get the best results when an image is exposed properly rather than making drastic changes in post.

I always shoot RAW as well... but RAW shouldn't be an excuse for anything. Even though I shoot RAW, I still do CWB for example. The ability to adjust white balance in post shouldn't be an excuse to not do it when you're shooting in RAW.
For the same reasons, I wouldn't want to shoot in RAW with Photoshop exposure sliders in mind.

Even though you think you're ETTR, you may be blowing out some channels. If there's a colour cast, your average RGB exposure might be fine, but an individual channel might be blown. That's a big problem in JPEG if you're not using the RGB histogram to do your measurements. If you're using RAW, you might think you're still OK because it captures more dynamic range, but that's not the only problem.

Another main issue is contrast. Look at a cloudy sky for an example. If you expose that properly, the contrast in the mid tones will be exceptional. Colour saturation will be naturally amazing. If you ETTR on the cloudy sky and then try to drop that down to where it should be in post... you'll end up with a cloudy sky that lacks that contrast and colour (relatively). Sure, there won't be any noise in there... but it won't look it's best or you'll have to step up the PP which will introduce it's own problems (banding, saturation leakage, haloing, etc..)

My 30D does a great job with noise and in a few clicks, you can get rid of any with Neat Image or whatever you're using.

The fact that somebody would consider shooting a high key (ETTR) in order to produce a low key image is a little disturbing. I don't get it.
If you're worried about noise in the shadows THAT MUCH, learn about some anti-noise techniques or play around with a black layer on multiply with a dark area mask.


Why is everybody so paranoid about noise to the point where they'd consider doing such drastic measures? I'm lost I guess.
I've seen some photographers that use extreme ISOs and produce some really great work... and they don't shoot for the highlight data... they shoot for a correct exposure where everything will look it's best.

I guess that's my opinion... and it's obvious others don't agree. Oh well.

[E-K]

[edit: felt my previous post was a little more argumentative than originally intended ;)]

The way I see it ETTR is just another technique to reduce noise in situations where it can be applied. In order to take a successful ETTR shot you need to shoot RAW and ensure that you are not blowing any channels.

I know shooting high key to get a low key shot sounds weird but this is really a scenario in which ETTR would be beneficial. In addition it is something which is easily achievable in this case versus using Photoshop to selectively apply noise filtering.

I'm just not sure I follow why you believe there will be a decrease in contrast and colour for an ETTR shot relative to a properly exposed one? Can you expand on that a little?

RAW exposure development happens in linear space so the adjustment is a simple scaling of the RAW data. So if a proper exposure resulted in a value of (1000, 500, 400) then the +1 EC shot would result in a value of (2000, 1000, 800). Applying -1 EC in RAW conversion would make the value (1000, 500, 400) again -- in otherwords exactly what you had with the proper exposure (within the limits of any statistical deviation). However, after applying the -1 EC on the ETTR image I have also scaled the noise floor down.

All that being said I don't normally have the luxury of using ETTR in the environments that I typically shoot so it's somewhat moot for me :).

e-k

[Robert_Lay]

... So if a proper exposure resulted in a value of (1000, 500, 400) then the +1 EC shot would result in a value of (2000, 1000, 800). Applying -1 EC in RAW conversion would make the value (1000, 500, 400) again --
....


You could be right, but "linear" does not necessarily mean a slope of unity - it only requires proportionality.

[E-K]

You could be right, but "linear" does not necessarily mean a slope of unity - it only requires proportionality.

I'm not sure I follow where you're getting a slope of unity from but then my brain is a little toast right now ;).

All I'm saying is that the behaviour of CCD/CMOS sensor is linear (with the possible exception where you start to reach saturation).

The data number associated with a given photosite is proportional to the number of photons that strike it. So lets call this proportionality factor g (essentially the gain). For the sake of argument lets deal with one value out of the 3-tuple.

1000 = ng where n is the number of photons.

If I leave the shutter open twice as long (i.e. +1 EC) then twice as many photons strike the photosite -- i.e. 2n. Since we know ng = 1000 then (2n)g = 2000.

When you do a -1 EC in RAW conversion you are essentially saying you only meant to leave the shutter open for half as long so you need to divide the data number by 2.

e-k

[PacAce]

My take on ETTR is, if the camera is set to ISO 100 (or whatever the "normal" lowest ISO is on the camera and I don't mean the "L" setting like ISO 50 on the 1DmkII), then shoot as far right as you can without blowing out thte details. Compensating during raw conversion as E-K suggested would give the best image quality with the least amount of noise.

If the ISO is set to anything higher, then ETTR isn't really going to buy much and you'd be better off shooting at a lower ISO and exposing normally (not ETTR) as suggested by RAitch.

[Robert_Lay]

I'm not sure I follow where you're getting a slope of unity from but then my brain is a little toast right now ;).

All I'm saying is that the behaviour of CCD/CMOS sensor is linear (with the possible exception where you start to reach saturation).

The data number associated with a given photosite is proportional to the number of photons that strike it. So lets call this proportionality factor g (essentially the gain). For the sake of argument lets deal with one value out of the 3-tuple.

1000 = ng where n is the number of photons.

If I leave the shutter open twice as long (i.e. +1 EC) then twice as many photons strike the photosite -- i.e. 2n. Since we know ng = 1000 then (2n)g = 2000.

When you do a -1 EC in RAW conversion you are essentially saying you only meant to leave the shutter open for half as long so you need to divide the data number by 2.

e-k
I would like to think about that some more, but I think your "gain" term will, in practice, be a function of the ISO setting. However, trying to understand the official definition of ISO is a challenge in itself.