Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: Zone System, light metering and neutral gray  (Read 12472 times)
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« on: September 28, 2007, 09:02:17 PM »
ReplyReply

In another forum we started a discussion about how the Zone System fits into the DR of our camera sensors. Ansel Adam's Zone System consists of 11 zones named 0 to X, where 0 is supposed to be pure black, X is pure white (no information in any of them thus), and zones I through IX are the ones containing information with 1 f-stop apart one to another, being zone V middle gray.

At the same time, camera exposure metering systems are supposed to translate the luminance of the measured area (no matter which measuring method is used) to a middle gray tone, i.e. to zone V.

Is this true? to which real linear level are measured areas translated to? and once the RAW file is developed, gamma corrected, and opened in Photoshop, what level in the range 0..255 means neutral gray? the 128 level?

Let's find out...

Me and another guy have run the same test on a Canon 350D and on his Fuji S3 Pro. The test consisted in just one shot: using a tele (300mm) over a uniformly lighted white surface, deliberately defocused to obtain the most regularly distributed image possible.

Under these circumstances, the final luminance produced by the camera should be the middle gray we are looking for on each camera, i.e. that gray tone to which the light metered areas are translated by the camera metering system.


LINEAR HISTOGRAM PRIOR TO WHITE BALANCE

Developing the image without applying any white balance (DCRAW -r 1 1 1 1 option) we can see what's the gap between the middle gray and the highlights right before starting to blow them up. This gap is how much we could overexpose the metered area before blowing it.





Now we will repeat the test for the Fuji S3 Pro, which has a very particular sensor, the Fuji super CCD which is actually two sensors in one (S captors: the regular ones, R captors: smaller and less sensitive photosites to prevent highlights from being blown in high DR scenes). So two images can be obtained from a unique RAW file.

LINEAR HISTOGRAM PRIOR TO WHITE BALANCE



Very interesing figures can be obtained from these histograms:

1. Under the same real exposure, the relative exposure of the S sensor over the R sensor happens to be 3.6 f-stops (just count the divisions). I have calculated accurately this figure over 3 different RAW files, so I think is a standard design value of how the Fuji Super CCD works.

2. Now, the security gap before starting to blow the highlights in the S sensor (not in the R sensor) is only about 2.3 f-stops. This means that the middle gray areas of the scene are allocated by the Fuji metering system over 2 f-stops higher than the 350D does.

3. This reduced security gap does not mean we cannot capture properly the 4 needed f-stops for zones VI, VII, VIII and IX, since the R photosites will do that job as soon as the information from the S photosites blowns up. In fact, the security gap for R captors is: 2.3+3.6=5.9 f-stops! almost 2 f-stops more than the Canon. That's is why it is so difficult to blow the highlights in a Fuji camera if high DR mode is activated.


NEUTRAL GRAY LEVEL AFTER GAMMA CORRECTION

Let's find out which value in the 0..255 range is the neutral gray located. Again we perform a totally neutral development on ACR getting the following image:

 .  

as we can see it's quite brighter than the gray produced by the 350D. With the colour checker we measure it to be about 112 (I measure the middle area of the image since my friend didn't manage to have a very uniformly lighted surface; that's why these histograms are a bit wider than desired. However as he spot mettered in the centre of the scene, that is the neutral gray (zone V) for the Fuji, and yields a value of 112, quite closer to 128 than that of the Canon.


Conclusions:
- Both cameras can register properly at least 4 f-stops higher than the metered area, which is needed to allocate zones VI, VII, VIII and IX of the Zone System
- Not all cameras provide the same gamma corrected value once in the range 0..255 for their neutral gray, and consequently it is wrong to say 128 is the neutral gray level in Photoshop. This will depend on the way the camera was designed (Canon 350D produces level ~74, and Fuji S3 in high DR mode produces level ~112).
- Of course a simple curve could readjust all levels making the true neutral gray (zone V) match level 128; but the actual starting point is a different value.

Agree? disagree?
« Last Edit: September 29, 2007, 11:12:05 AM by GLuijk » Logged

Jonathan Wienke
Sr. Member
****
Offline Offline

Posts: 5759



WWW
« Reply #1 on: September 28, 2007, 11:57:28 PM »
ReplyReply

The gamma curve inherent in RGB editing spaces means that there is not a fixed number of levels per f-stop. Nor is there a specific number of stops of DR between the white point and black point. So assuming that level 128 is 5 stops from 255 is incorrect. Since different RGB spaces have different gamma curves, the results will change depending on which space you convert to. Repeat your experiment, converting to sRGB, Adobe RGB 1998, and ProPhoto, and you'll get different results for each one. The only time there is a consistent level-to f-stop relationship is with linear RGB before gamma adjustment, when no clipping is present.
Logged

Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« Reply #2 on: September 29, 2007, 03:20:17 AM »
ReplyReply

Quote
The gamma curve inherent in RGB editing spaces means that there is not a fixed number of levels per f-stop. Nor is there a specific number of stops of DR between the white point and black point. So assuming that level 128 is 5 stops from 255 is incorrect. Since different RGB spaces have different gamma curves, the results will change depending on which space you convert to. Repeat your experiment, converting to sRGB, Adobe RGB 1998, and ProPhoto, and you'll get different results for each one. The only time there is a consistent level-to f-stop relationship is with linear RGB before gamma adjustment, when no clipping is present.
[a href=\"index.php?act=findpost&pid=142567\"][{POST_SNAPBACK}][/a]

Good point. I did the test converting in ACR to sRGB (gamma=2.2). I repeat it here for the Canon 350D with Prophoto (gamma=1.Cool and the middle gray moves from 74 (sRGB) to 57 (ProPhoto):

sRGB middle gray (level 74) vs ProPhoto middle gray (level 57)
 .  



Thx Jonathan.
« Last Edit: September 29, 2007, 03:26:41 AM by GLuijk » Logged

bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #3 on: September 29, 2007, 08:41:34 AM »
ReplyReply

Quote
The gamma curve inherent in RGB editing spaces means that there is not a fixed number of levels per f-stop. Nor is there a specific number of stops of DR between the white point and black point. So assuming that level 128 is 5 stops from 255 is incorrect. Since different RGB spaces have different gamma curves, the results will change depending on which space you convert to. Repeat your experiment, converting to sRGB, Adobe RGB 1998, and ProPhoto, and you'll get different results for each one. The only time there is a consistent level-to f-stop relationship is with linear RGB before gamma adjustment, when no clipping is present.
[{POST_SNAPBACK}][/a]

In a gamma encoded image, the zones are still present, but they are remapped in a non-linear fashion. This may be seen by photographing a step wedge whose steps vary by a density of 0.1 (1/3 f/stop) and plotting the results. I use Imatest to automate the process.

At default settings most raw converters (and their in camera equivalents) roll of the shadows and apply an S-curve to the values. These effects are shown below. On the X-axis each 0.3 density units represents one f/stop. Shown for comparison is a linear conversion from DCRaw.



Here is the ACR rendering into a gamma 2.2 space with default values, except for setting the black point to 0. It shows the S-curve well. Note that the gamma is 2.01 and not 2.2. Again, each 0.3 density change in exposure represents one f/stop. There is compression of the highlights and shadows.



And here is the ACR rendering with the curve set to linear, contrast = 0, and brightness = 0. This linearizes the tonal response curve, but gamma is still applied. The steps are uniformly distributed among the zones with no compression of the highlights and shadows. The deep shadows are lightened by flare light, and the curve is not linear in this area.



[a href=\"http://www.normankoren.com/digital_cameras.html]Norman Koren[/url] gives information about the ISO standard for digital camera ISO. If an 18% gray card is exposed to the ISO standard, the pixel level in a gamma 2.2 space should be 114, which is close to what Guillermo obtained. According to tests by DPReview, Canon exposures tend to be slightly hot, and Guillermo's value of 128 rather than 114 for this camera is typical for Canons.

Personally, I think Guillermo's experiment is very well done and beautifully illustrated (as usual for him).

Bill
« Last Edit: September 29, 2007, 09:25:53 AM by bjanes » Logged
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« Reply #4 on: September 29, 2007, 11:08:14 AM »
ReplyReply

Quote
Norman KorenIf an 18% gray card is exposed to the ISO standard, the pixel level in a gamma 2.2 space should be 114, which is close to what Guillermo obtained. According to tests by DPReview, Canon exposures tend to be slightly hot, and Guillermo's value of 128 rather than 114 for this camera is typical for Canons.

Bill, but we must consider something: the fact that the Fuji in gamma 2.2 yields 112 (very close to 114) for middle gray, is because it was designed to allocate the metered area of the image about 2.3 f-stops before starting to blow highlights. And Fuji can afford this since the R sensor will provide the highlight information for those areas beyond the metered area plus 2.3 f-stops which are blown in the S sensor.

If the Canon sensor would not allow at least 4 f-stops beyond the middle grey in the linear RAW data, people would quickly come to complain for easily getting their lights blown, so I think a value around 74 for the middle gray in gamma corrected range must be the rule for any single-sensor camera, not only the Canon. Even for the Fuji if the R sensor is not being used. Don't you think?

Well I already asked for a .RAF file to test this, will come back here as soon I get the result.



Quote
Personally, I think Guillermo's experiment is very well done and beautifully illustrated (as usual for him).

wow, thx!
Logged

bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #5 on: September 29, 2007, 11:19:56 AM »
ReplyReply

Quote
The gamma curve inherent in RGB editing spaces means that there is not a fixed number of levels per f-stop. Nor is there a specific number of stops of DR between the white point and black point. So assuming that level 128 is 5 stops from 255 is incorrect. Since different RGB spaces have different gamma curves, the results will change depending on which space you convert to. Repeat your experiment, converting to sRGB, Adobe RGB 1998, and ProPhoto, and you'll get different results for each one. The only time there is a consistent level-to f-stop relationship is with linear RGB before gamma adjustment, when no clipping is present.
[{POST_SNAPBACK}][/a]

There is not a fixed number of levels with a linear raw file either; the brightest f/stop contains half the levels. With gamma encoding, more levels are in the shadows, but this area is still poor in levels. This is shown by [a href=\"http://www.normankoren.com/digital_tonality.html]Norman Koren[/url] in a table on his web site.

For additional comments, see my previous reply.
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #6 on: September 29, 2007, 11:34:02 AM »
ReplyReply

Quote
Bill, but we must consider something: the fact that the Fuji in gamma 2.2 yields 112 (very close to 114) for middle gray, is because it was designed to allocate the metered area of the image about 2.3 f-stops before starting to blow highlights. And Fuji can afford this since the R sensor will provide the highlight information for those areas beyond the metered area plus 2.3 f-stops which are blown in the S sensor.

If the Canon sensor would not allow at least 4 f-stops beyond the middle grey in the linear RAW data, people would quickly come to complain for easily getting their lights blown, so I think a value around 74 for the middle gray in gamma corrected range must be the rule for any single-sensor camera, not only the Canon. Even for the Fuji if the R sensor is not being used. Don't you think?
[{POST_SNAPBACK}][/a]

Guillermo, I think a gamma 2.2 encoded value of 74 for the Canon is suspect; I mis-read your post initially. Light meters are calibrated for middle gray, but when exposing with raw, I think that one should forget about midtones, and place the exposure so that the highlights are just short of clipping. Since the response is linear, I don't think that the mid-tones have any special significance.

Here is a nice table of expected pixel values in raw and gamma encoded spaces complied by [a href=\"http://www.pochtar.com/gamut_view/gamma.htm]Julia Borg[/url]. She is using the Kodak Q14 scale where M has a density of 0.75  and a reflectance of 17.78%.

Bill
Logged
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« Reply #7 on: September 29, 2007, 12:06:01 PM »
ReplyReply

Quote
Guillermo, I think a gamma 2.2 encoded value of 74 for the Canon is suspect; I mis-read your post initially. Light meters are calibrated for middle gray, but when exposing with raw, I think that one should forget about midtones, and place the exposure so that the highlights are just short of clipping. Since the response is linear, I don't think that the mid-tones have any special significance.

Here is a nice table of expected pixel values in raw and gamma encoded spaces complied by Julia Borg. She is using the Kodak Q14 scale where M has a density of 0.75 and a reflectance of 17.78%.

Bill
[{POST_SNAPBACK}][/a]

mmm I must admit I feel lost when attending photographic considerations. I will try to explain why I think the 74 figure makes sense:

This is the linear non-white balanced histogram (DCRAW -r 1 1 1 1) of my Canon shot metering on a lighted white wall which should be made by the camera middle gray:



You can see there is a gap between the max of the RG channels and saturation. We could have overexposed by 4 f-stops and not blowing, ok?
Once WB is applied, since the B channel is much less powerful (for being almost 2 stops lower than the RG channels) we can approximate the final balanced linear histogram will still remain approximately over 4 stops below saturation. So let's do a simple calculation to translate a -4EV linear value into gamma corrected:

-4=log(x)/log(2) -> x=0.0625

Now we apply gamma:

0.0625^(1/2.2)=0,2836

And convert to 0..255 range:

0,2836*255=72,3 makes sense.


Let's do the same for the Fuji histogram, which was -2.3EV below saturation:

-2.3=log(x)/log(2) -> x=0,2031
0,2031^(1/2.2)=0,4845
0,4845*255=123,5 again makes sense at least not so far from 112 (with -2.2EV we would have got exactly 128)

The 2.3 figure is very approximate, as soon as we change it into -2.5EV which could even be more realistic, the final value becomes 116. It's very sensitive for the gamma expansion.


Maybe I am missing something or being too simplistic (the sRGB conversion should move a bit the levels for example, but I don't think too much).


I have uploaded here my Canon RAW file (the one yielding 74) in case you could check:
[a href=\"http://www.guillermoluijk.com/download/middlegray.cr2]RAW file[/url]
XMP development parameters used

Regards.
« Last Edit: September 29, 2007, 12:14:45 PM by GLuijk » Logged

Jonathan Wienke
Sr. Member
****
Offline Offline

Posts: 5759



WWW
« Reply #8 on: September 29, 2007, 03:40:09 PM »
ReplyReply

Quote
There is not a fixed number of levels with a linear raw file either; the brightest f/stop contains half the levels.

Not linear fixed, but logarithmic base-2 fixed. You can look at linear RAW data and easily tell how many stops from clipping you are.
Logged

bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #9 on: September 29, 2007, 03:59:11 PM »
ReplyReply

Quote
Not linear fixed, but logarithmic base-2 fixed. You can look at linear RAW data and easily tell how many stops from clipping you are.
[a href=\"index.php?act=findpost&pid=142755\"][{POST_SNAPBACK}][/a]

That is obvious. If you use log base 10 as I did my graphs, then -0.3 = 1 stop below clipping, -0.6 = 2 stops below clipping, etc. I don't get your point.
Logged
Jonathan Wienke
Sr. Member
****
Offline Offline

Posts: 5759



WWW
« Reply #10 on: September 29, 2007, 11:35:37 PM »
ReplyReply

Quote
That is obvious. If you use log base 10 as I did my graphs, then -0.3 = 1 stop below clipping, -0.6 = 2 stops below clipping, etc. I don't get your point.

Figuring out how many stops from clipping is a lot harder once you've converted to a standard editing space. Depending on the particular editing space, you'll have more levels per stop in the highlights than the shadows or vice versa, depending on whether it's gamma 2.2 or 1.8, and then there's the added complexity of the tone curve that the RAW converter wants to toss into the mix that deviates from mathematical accuracy for the sake of visual appeal. It's a lot easier to say clipped = zone X, and then progressively divide by two to define the transition to each next lower zone than it is to unravel the editing space gamma curve (which isn't even an actual gamma curve in the case of sRGB, which has a toe and shoulder built in to the luminance curve), hope you have either completely disabled the RAW converter tone curve or properly compensated for it (do you KNOW that the settings mean exactly what you think they do or are you guessing?), not to mention the issue of black point compensation when converting from the linear RGB to the editing space in the RAW converter...

It's just a lot easier before all the other factors muddy the water.
« Last Edit: September 29, 2007, 11:38:47 PM by Jonathan Wienke » Logged

Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« Reply #11 on: September 30, 2007, 05:58:09 AM »
ReplyReply

Please guys, could you tell me something about my figure calculations? do they make sense to you?

I can't sleep at night    
Logged

bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #12 on: September 30, 2007, 08:02:17 AM »
ReplyReply

Quote
Please guys, could you tell me something about my figure calculations? do they make sense to you?

I can't sleep at night   
[a href=\"index.php?act=findpost&pid=142882\"][{POST_SNAPBACK}][/a]

Guillermo,

I don't see anything wrong with your math. I did my own calculations using the green channel of the raw file. For white balance the blue and red channels would be equalized to the green, so I didn't bother with them.

From a DCRaw conversion, I get a raw pixel value for green of 125. The conversion to gamma 2.2 gives a pixel value of 52.2. The raw pixel value normalizes to 0.030525 (125/4095). Then I convert to gamma 2.2: 0.030525^(1/2.2) = 0.20472. Then converting to 0..255: 0.20472 * 255 = 52.2. This is close to what you calculated.

However, if the camera meter is calibrated to the ISO standard, the raw value should be  18/106 of full scale (e.g., a pixel level of 696 at the output of a 12-bit A-to-D converter).  In a gamma 2.2 space, this corresponds to a pixel value of 114. The ACR conversion gave a value of 74. I conclude that the image was under-exposed and I would recommend double checking the exposure.

Bill
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #13 on: September 30, 2007, 08:31:30 AM »
ReplyReply

Quote
Figuring out how many stops from clipping is a lot harder once you've converted to a standard editing space. Depending on the particular editing space, you'll have more levels per stop in the highlights than the shadows or vice versa, depending on whether it's gamma 2.2 or 1.8, and then there's the added complexity of the tone curve that the RAW converter wants to toss into the mix that deviates from mathematical accuracy for the sake of visual appeal. It's a lot easier to say clipped = zone X, and then progressively divide by two to define the transition to each next lower zone than it is to unravel the editing space gamma curve (which isn't even an actual gamma curve in the case of sRGB, which has a toe and shoulder built in to the luminance curve), hope you have either completely disabled the RAW converter tone curve or properly compensated for it (do you KNOW that the settings mean exactly what you think they do or are you guessing?), not to mention the issue of black point compensation when converting from the linear RGB to the editing space in the RAW converter...

It's just a lot easier before all the other factors muddy the water.
[{POST_SNAPBACK}][/a]

The sRGB curve has a linear portion in the shadows, but there is no shoulder for the highlights. Actually, sRGB is not a single function, but a grafting together of two functions: a linear function for the shadows (R <= 0.0405) and a power function for the remaining values, using an exponent of 2.5. This averages out to a gamma of 2.2 and this is used for simplified sRGB. With values above middle gray, there is little difference between sRGB and simplified sRGB. This is explained on Bruce Lindbloom's site.

A shoulder would be introduced in the rendering into the gamma 2.2 space if the raw converter was set to do so. With ACR you can linearize the TRC by setting brightness and contrast to zero. In my Imitest plots, there is no guesswork. The actual curve is plotted. Didn't you read my previous posts?

It is not that difficult to see how far you are from clipping if you have linearized the tone curve. You can do the calculations yourself or refer to a [a href=\"http://www.pochtar.com/gamut_view/gamma.htm]table[/url].

Finally, gamma does not affect extreme highlights much at all. It affects the mid-tones most as shown by Julia Borg's table.

Bill
Logged
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« Reply #14 on: September 30, 2007, 08:50:51 AM »
ReplyReply

Quote
Guillermo,

I don't see anything wrong with your math. I did my own calculations using the green channel of the raw file. For white balance the blue and red channels would be equalized to the green, so I didn't bother with them.

From a DCRaw conversion, I get a raw pixel value for green of 125. The conversion to gamma 2.2 gives a pixel value of 52.2. The raw pixel value normalizes to 0.030525 (125/4095). Then I convert to gamma 2.2: 0.030525^(1/2.2) = 0.20472. Then converting to 0..255: 0.20472 * 255 = 52.2. This is close to what you calculated.

However, if the camera meter is calibrated to the ISO standard, the raw value should be 18/106 of full scale (e.g., a pixel level of 696 at the output of a 12-bit A-to-D converter). In a gamma 2.2 space, this corresponds to a pixel value of 114. The ACR conversion gave a value of 74. I conclude that the image was under-exposed and I would recommend double checking the exposure.

Bill
[a href=\"index.php?act=findpost&pid=142892\"][{POST_SNAPBACK}][/a]

Thank you Bill.

Two important things:


1. Are you sure equalizing R and B channels to the G is the right way to do a proper white balance? IMO white balance should never change the overall exposure of the RAW data, so being consistent with this idea, all three RGB channels have to be rearranged in such a way that the final situation is that all of them remain aligned in the histogram, but the overall exposure (image brightness) remains the same. So setting the G as a reference wouldn't be right.
That is why I didn't want to use DCRAW for the white balance and used ACR instead. If you open my RAW file in ACR and tune the WB to obtain a pure gray (the 3 channels perfectly aligned) you will see that the G channel actually moves from its position, and I think it has to be that way. Making the G channel static, will lead you to a histogram corresponding to an image with an exposure alteration.

Of course since G weights more in any luminance model than the other 2 channels, making it static can yield quite accurate results but not the strictly right ones.

2. I understand what you say about the ISO standard. But I think this cannot be applied straight to the way digital cameras allocate the captured data into their linear dynamic range since they are firmly limited in the right end of the DR by saturation of the sensor.
I am quite sure I did a right exposure with my camera since I metered in all 3 modes of it. Ok, I may have a malfunctioning unit, but I think this is not the case and the reason is the camera must allow at least 4 complete f-stops above what it considers middle gray (zone V), otherwhise zones VI, VII, VIII and IX could not properly be recorded.



So a camera with a metereing system that would guarantee that middle gray after a 2.2 gamma correction falling straight into 114, would simply blow any information contained in the highest zones (IX, probably VIII too) since it would necessarily have to allocate the middle gray much closer to the end of the RAW linear range than 4 stops. In fact it should allocate it 2.2 stops below the saturation limit to accomplish with the ISO gamma 2.2 middle gray requirements.

Taking your 696 level middle gray pixel, another area of the scene which would be 4 stops higher in luminance (zone IX), would need to be recorded with a linear sensor value of: 696*(2^4)=11136, far from the 12-bit limit of 4096.


What do you think?

This was not the case of the Fuji since Fuji's R captors will give you even more than those needed 4 stops; that's why I am awaiting with impatience that my friend who owns a Fuji provides me with a RAW shot in the 100% (nor expanded DR) mode, which will have only one exposure image and thus, if my hypothesis is not wrong, should allow definitively more stops from the middle gray till the end of its histogram.
« Last Edit: September 30, 2007, 09:20:35 AM by GLuijk » Logged

Jonathan Wienke
Sr. Member
****
Offline Offline

Posts: 5759



WWW
« Reply #15 on: September 30, 2007, 11:34:57 AM »
ReplyReply

Quote
A shoulder would be introduced in the rendering into the gamma 2.2 space if the raw converter was set to do so. With ACR you can linearize the TRC by setting brightness and contrast to zero. In my Imitest plots, there is no guesswork. The actual curve is plotted. Didn't you read my previous posts?

Yes. Guillermo is using dcRaw, not ACR, and you still haven't addressed the issue of black point compensation when the linear RGB is converted to the editing space. So even if you linearize the TRC, you're still not guaranteed a fixed number of levels per stop in the converted file. You can get a close approximation that will be consistent for a particular camera, but the closer you get to the noise floor the greater the variance will be, depending on the camera model and ISO setting. If the camera firmware does any adjustment to the black levels (for noise reduction purposes or to even out a voltage bias in the ADC) before outputting the RAW, then even the linear RAW values aren't going to be a completely accurate guide to the exposure interval from the clip point.

I'd say that Guillermo's calculations are close enough to be useful and accomplish the task he has in mind, but that there are some real-world factors (mostly the camera firmware) that are going to introduce some small errors.

I think it's also worth noting that this entire discussion assumes a camera with a 10-stop capture range, and that some cameras have less and others more than that. In addition, given the principle of expose to the right, it is pointless to say that a middle gray should always be metered to a particular value. If you're shooting in fog where there are no bright highlights, there's no reason not to increase exposure so that the brightest highlights fall into Zone IX even if that means the midtones are in Zone VII instead of Zone V, and the darkest tones fall into Zone IV.
« Last Edit: September 30, 2007, 11:42:52 AM by Jonathan Wienke » Logged

bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #16 on: September 30, 2007, 12:37:18 PM »
ReplyReply

Quote
Yes. Guillermo is using dcRaw, not ACR, and you still haven't addressed the issue of black point compensation when the linear RGB is converted to the editing space. So even if you linearize the TRC, you're still not guaranteed a fixed number of levels per stop in the converted file. You can get a close approximation that will be consistent for a particular camera, but the closer you get to the noise floor the greater the variance will be, depending on the camera model and ISO setting. If the camera firmware does any adjustment to the black levels (for noise reduction purposes or to even out a voltage bias in the ADC) before outputting the RAW, then even the linear RAW values aren't going to be a completely accurate guide to the exposure interval from the clip point.
[a href=\"index.php?act=findpost&pid=142926\"][{POST_SNAPBACK}][/a]

I don't think black point compensation is relevant to the discussion. Firstly, we are talking about midtones and highlights, where BPC makes little difference. Secondly, BPC is used when the darkest point of the destination profile is less than that of the source profile. As we discussed in a previous thread (where we cleared up some misunderstandings), a good example is printing to paper, where the D-max is limited. In this case, it is best to map the darkest point in the source to the darkest point in the destination, and make adjustments in the darker tones from there. It would not make sense to apply BPC when rendering the raw file into the working space, since the black point of the output device is not known at the time of the conversion. The raw black point is simply mapped to the working profile without any compensation.

As to a fixed number of levels, who says that a fixed number of levels is present or necessary. As Norman Koren shows in his table, the number of levels varies with bit depth and gamma. Each zone will contain a varying number of levels, and the exact number is not important if no visible banding is present.


Quote
In addition, given the principle of expose to the right, it is pointless to say that a middle gray should always be metered to a particular value. If you're shooting in fog where there are no bright highlights, there's no reason not to increase exposure so that the brightest highlights fall into Zone IX even if that means the midtones are in Zone VII instead of Zone V, and the darkest tones fall into Zone IV.
[a href=\"index.php?act=findpost&pid=142926\"][{POST_SNAPBACK}][/a]


I'm glad we can agree on something. You will note in a previous post to Guillermo, I said that with exposure to the right and raw files, the midtones are largely irrelevant. One should really meter for the highlights, and place them just short of clipping. Even if you are working with camera rendered JPEGs, darkening an image is safer than lightening, since the lighter zones are information rich (many levels) and you are pushing them down to zones with fewer levels.
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2772



« Reply #17 on: September 30, 2007, 01:58:05 PM »
ReplyReply

Quote from: GLuijk,Sep 30 2007, 07:50 AM

Two important things:
1. Are you sure equalizing R and B channels to the G is the right way to do a proper white balance?
[{POST_SNAPBACK}][/a]

No, I'm not sure that the above method is the best or most proper, but that is how it is done in DCRaw and in Nikon cameras. For a list of white balance multipliers see the post by [a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]Julia Borg[/url]. However, as you noted, green is predominant because of its weighting for luminance and because the sensor has twice many green elements than red or blue.

[a href=\"index.php?act=findpost&pid=142902\"][{POST_SNAPBACK}][/a]
[/quote]

2. I understand what you say about the ISO standard. But I think this cannot be applied straight to the way digital cameras allocate the captured data into their linear dynamic range since they are firmly limited in the right end of the DR by saturation of the sensor.
I am quite sure I did a right exposure with my camera since I metered in all 3 modes of it. Ok, I may have a malfunctioning unit, but I think this is not the case and the reason is the camera must allow at least 4 complete f-stops above what it considers middle gray (zone V), otherwhise zones VI, VII, VIII and IX could not properly be recorded.



So a camera with a metereing system that would guarantee that middle gray after a 2.2 gamma correction falling straight into 114, would simply blow any information contained in the highest zones (IX, probably VIII too) since it would necessarily have to allocate the middle gray much closer to the end of the RAW linear range than 4 stops. In fact it should allocate it 2.2 stops below the saturation limit to accomplish with the ISO gamma 2.2 middle gray requirements.

Taking your 696 level middle gray pixel, another area of the scene which would be 4 stops higher in luminance (zone IX), would need to be recorded with a linear sensor value of: 696*(2^4)=11136, far from the 12-bit limit of 4096.
What do you think?

This was not the case of the Fuji since Fuji's R captors will give you even more than those needed 4 stops; that's why I am awaiting with impatience that my friend who owns a Fuji provides me with a RAW shot in the 100% (nor expanded DR) mode, which will have only one exposure image and thus, if my hypothesis is not wrong, should allow definitively more stops from the middle gray till the end of its histogram.
[a href=\"index.php?act=findpost&pid=142902\"][{POST_SNAPBACK}][/a]
[/quote]

The ISO standard gives 7% headroom an Norman explained. More headroom for highlights can be given for more demanding situations, but then noise and tonality may suffer.

My Nikon D70 and D200 comply closely to the ISO standard. Here is an experiment with the D70. I took a Kodak 18% gray card and Q14 target and placed them on a background. Since the ISO standard for light meters pegged to 12% reflectance, rather then 18%, I took an exposure reading from the gray card and gave 1/2 stop more exposure than indicated. The gray was rendered at about 118 in aRGB and the steps in the Q14 target were more or less as they should be. I used the default ACR tone curve.

[attachment=3436:attachment]

If your metered exposure is rendered at 73 as shown below (simulated by adjusting exposure in ACR), your picture will appear dark as shown below. Your highlights might be protected, but the picture will be dark and when you lighten the image you will increase noise and may have banding in the shadows. What is confusing is that on DPreview, Phil noted that the Canon 350D gave 1/3 stop over nominal exposure.

[attachment=3435:attachment]

You might want to search the forum for postings by John Sheehy. He has done extensive experimentation with Canon cameras and has found that Canon cameras do allow more headroom than some other cameras, such as the Nikon; since they have less noise, perhaps they can afford to do so.
Logged
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1280



WWW
« Reply #18 on: September 30, 2007, 02:28:44 PM »
ReplyReply

Yes it is as you say Bill, the"middle gray" my 350D produced once gamma corrected is quite dark, I posted a couple or real images above (74 level is a quite dark gray). I don't mean this behaviour is better or worse than that of a camera producing 114, I simply want to find out if each camera behaves in a different way with respect to this.

However I didn't understand too well why you corrected your camera's metering by 1/2 stop. Aren't we finding out in which level the camera allocates any metering on a uniform surface? independently of the reflectance of the card used. You are using the built-in camera's meter, don't you? If so the camera doesn't know how much reflectance the card has, but actually doesn't need to know, it simply measures the amount of light entering its metering system to allocate that luminance as a middle gray in the final image.

What you say about John is consistent with the general belief that Canon cameras trend to underexpose (a lot of people take them to be "fixed").
« Last Edit: September 30, 2007, 02:32:10 PM by GLuijk » Logged

Jonathan Wienke
Sr. Member
****
Offline Offline

Posts: 5759



WWW
« Reply #19 on: September 30, 2007, 02:38:27 PM »
ReplyReply

Quote
I don't think black point compensation is relevant to the discussion.

Perhaps it is not relevant for the reasons you mention, but it is my understanding that BPC is always used when converting to one editing space to another (ProPhoto to sRGB, etc.) and that it was used when converting linear RAW to the destination space.
Logged

Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad