Finally, a raw histogram?

[bjanes]

In his recent review of new firmware for the Leica S2, Mark Dubovoy noted that the histogram is now based on the raw data and is not derived from a JPEG preview. Many have been requesting this for years, but thus far, only Leica has listened to its customers. Does the S2 also have RGB histograms based on the raw data and not referred to a relatively narrow space such as AdobeRGB or sRGB?

Regards,

Bill

[Doug Peterson]

Many have been requesting this for years, but thus far, only Leica has listened to its customers.

Not just Leica :-P. Phase/Leaf.

One of the benefits of not having in-camera JPGs is you can't base your histogram off of them :-).

[tho_mas]

Not just Leica :-P. Phase/Leaf.

however the histogram is Gamma corrected. Gamma 1.8 (with Phase)... as Capture One is based on Gamma 1.8.
I am finding a Gamma corrected histogram very practical anyway... actually for me it's not so important if the histo on the camera is Gamma corrected or not... it's much more important that the RAW software shows the same histo as the camera for the respective capture (and C1 shows the same histo, at least with Phase backs).

[BernardLanguillier]

How is White Balance handled in these "raw" histograms?

Cheers,
Bernard

[bjanes]

however the histogram is Gamma corrected. Gamma 1.8 (with Phase)... as Capture One is based on Gamma 1.8.
I am finding a Gamma corrected histogram very practical anyway... actually for me it's not so important if the histo on the camera is Gamma corrected or not... it's much more important that the RAW software shows the same histo as the camera for the respective capture (and C1 shows the same histo, at least with Phase backs).

Gamma encoding is desirable for all but purists, or else everything occurs on the left. Even better IMHO would be a log base 2 encoding, since we photographers tend to think f/stops. For proper ETTR exposure, it is imperative that there is no headroom safety factor for the highlights.

Regards,

Bill

[bjanes]

How is White Balance handled in these "raw" histograms?

Hopefully, WB would not be peformed and the raw channels are merely displayed. As long as no channel clipping takes place, one can white balance using multipliers less than 1.0 if needed to prevent clipping.

Regards,

Bill

[tho_mas]

How is White Balance handled in these "raw" histograms?

WB is reflected in the histogram.

For proper ETTR exposure, it is imperative that there is no headroom safety factor for the highlights.

there is no headroom... exposure warning is first shown on overexposure... so at RGB levels above 255.

we photographers tend to think f/stops

Caputre One provides an additional histogram to display exposure evaluation (as opposed to the histogram that shows the values of your adjusted capture incl. levels, curves, color edits, whatever...). The scale of the exposure evaluation also shows f-stops (see attachment). The histogram on the digibacks do not show these indications... however you can use Capture One's indications as reference. If you are really anal you could also draw the scale on the back's LCD with a permanent marker ;-)

Hopefully, WB would not be peformed and the raw channels are merely displayed. As long as no channel clipping takes place, one can white balance using multipliers less than 1.0 if needed to prevent clipping.

Sounds cumbersome. Why not just set the correct WB while shooting? Or at least a preset that comes close (tungsten, daylight etc.). This way the histogram is "WYSIWYG" (so to speak).

I can understand that it may be annoying for you that your camera doesn't show a useful histogram and that your RAW software doesn't show the same histogram as your camera. But these limitations do not apply to all systems... obviously.

[JimAscher]

I have a Sanho HyperDrive Album which has been the subject matter of other threads in this forum, including one I initiated myself.  One of its attributes is to enable one to examine one's RAW output in the field, by transferring a RAW image from the camera to the Album via the memory card.  The Album does contain a histogram.  The discussion in this thread has prompted me to query the Sanho technical staff to determine whether the histogram depicted in the Album for a RAW file is genuinely that of the RAW file and not of the accompanying JPEG image.  The reply I just received is: "If you decode the image file in RAW, the histogram represents what's in the RAW file."  I don't know how much use this is to anyone else, but I thought I'd pass it along.

[NikoJorj]

Why not just set the correct WB while shooting? Or at least a preset that comes close (tungsten, daylight etc.). This way the histogram is "WYSIWYG" (so to speak).

Because the the application of WB distorts the histogram (it amplifies some channels more than others), and it doesn't help to see if there is raw clipping (and where).
If a channel blows because of WB application but is not blown in the raw capture, it's easy to deal with in processing.
If the raw data is blown, you lost...

[tho_mas]

Because the the application of WB distorts the histogram (it amplifies some channels more than others), and it doesn't help to see if there is raw clipping (and where).
If a channel blows because of WB application but is not blown in the raw capture, it's easy to deal with in processing.
If the raw data is blown, you lost...

again, that might apply to your system as your RAW software doesn't show the same as the LCD on your camera. If you are working with a Phase back & C1 the clipping shown on the camera's histogram and C1's histogram is absolutely consistent (I guess it's the same with Leaf + Hasselblad). This is why it makes sense to adjust the WB when shooting.
Sure you can shoot for instance with a too cold WB to avoid color clipping in the red chanel... but when you alter the WB in the RAW software clipping in the red chanel will be introduced. So what did you gain?

[bjanes]

Hopefully, WB would not be peformed and the raw channels are merely displayed. As long as no channel clipping takes place, one can white balance using multipliers less than 1.0 if needed to prevent clipping.

Sounds cumbersome. Why not just set the correct WB while shooting? Or at least a preset that comes close (tungsten, daylight etc.). This way the histogram is "WYSIWYG" (so to speak).

I can understand that it may be annoying for you that your camera doesn't show a useful histogram and that your RAW software doesn't show the same histogram as your camera. But these limitations do not apply to all systems... obviously.

My camera (Nikon D3) does show a reasonable histogram and this histogram is very similar to that of ACR when one uses the necessary BaselineExposure correction of -0.5 in ACR. However, at times is is useful to look at the raw data without white balance. A good example is this yellow flower: the histogram with an ETTR exposure is shown in the ACR preview, when rendering into Adobe RGB, which was the space set in the camera and used for the JPEG preview and histograms. The histograms of the file as rendered into Photoshop by ACR are also show. The luminance histogram looks fine, but is heavily weighted towards the green so that the red and blue channels have little effect. The green in the RGB histogram is fine, but the red is clipped. Yellow, of course, contains red and green.

Camera histogram:


ACR histogram with Exposure -0.5 baseline correction:


Photoshop luminosity histogram:


PhotoshopRGB composite histogram:


Rawnalize histogram showing the raw data without any white balance:


Examination of the raw file in Rawnalize shows that the green and red channels are short of clipping. However, the WB multiplier for the green channel is 1.0 and that for the red channel is 1.7 as shown by Rawnalize. When white balance is applied, the red channel will be clipped, whereas it is intact in the raw file. One could decrease exposure in the camera so that the red is not clipped during normal white balance with a red multiplier of 1.7. However, a better approach is to use multipliers all less than one so that no white balance clipping can occur and one can get a better signal:noise. One can do this directly in DCRAW, but in ACR one would decrease the exposure.

To obtain a better evaluation of the RGB channels in the raw file, many Nikon users load a special White Balance into the camera, so that white multipliers of 1 are used for all channels (UniWB). A related ploy regarding WB is to place a magenta filter over the camera lens to hold back some of the green light and obtain a better balance between the channels. This can easily add a half stop to the DR. With older cameras with a limited DR, this was advantageous, but it is probably not worthwhile with current state of the art cameras.

Regards,

Bill

P.S.

In the example shown, it was not possible to eliminate red clipping by any reasonable decrease in exposure, becuase saturation clipping was occuring during rendering into the relatively narrow Adobe RGB space, which is the widest space available on this camera. One could eliminate such clipping through the use of ProPhoRGB.

[tho_mas]

My camera (Nikon D3) does show a reasonable histogram and this histogram is very similar to that of ACR when one uses the necessary BaselineExposure correction of -0.5 in ACR

sorry to disagree... the camera histogram shows clipping only in the red chanel and no clipping in the green chanel whereas ACR also shows heavy clipping in the green chanel.

Again, those workarounds might be great for you as you use tools that do not show consistant histograms.
Therefore it's also quite understandable that you try to find a workaround.

But on a Phase back such an issue simply does not exist. So I am looking at the "problem" from the other side...
You have to white balance the image at some point anyway... either when shooting or afterwards in the RAW software.
Attached 3 screenshots (from a Phase back). The 3 histos show exactly what the camera histo would show if you'd set the respective WB while shooting.
1.) white balanced (looking only at this crop you could even increase exposure by 1/3 or 2/3 stops).
2.) wrong WB 2000K - heavy clipping in the blue chanel.
3.) wrong WB 8000K - no clipping but due to the red chanel you wouldn't increase exposure here.

Now, as the white balanced histogram on the camera shows exactly the same as the histogram in the RAW software, it's pretty clear that white balancing while shooting is preferable with this system - you simply can see the real final outcome right on the camera LCD.
If you want so: it's consistent "out of the box" and this is why I couldn't care less about a "raw-raw"-histogram. The P1/C1-approach simply works... in terms of practicability.

[hjulenissen]

sorry to disagree... the camera histogram shows clipping only in the red chanel and no clipping in the green chanel whereas ACR also shows heavy clipping in the green chanel.

Again, those workarounds might be great for you as you use tools that do not show consistant histograms.
Therefore it's also quite understandable that you try to find a workaround.

But on a Phase back such an issue simply does not exist. So I am looking at the "problem" from the other side...
You have to white balance the image at some point anyway... either when shooting or afterwards in the RAW software.
Attached 3 screenshots (from a Phase back). The 3 histos show exactly what the camera histo would show if you'd set the respective WB while shooting.
1.) white balanced (looking only at this crop you could even increase exposure by 1/3 or 2/3 stops).
2.) wrong WB 2000K - heavy clipping in the blue chanel.
3.) wrong WB 8000K - no clipping but due to the red chanel you wouldn't increase exposure here.

Now, as the white balanced histogram on the camera shows exactly the same as the histogram in the RAW software, it's pretty clear that white balancing while shooting is preferable with this system - you simply can see the real final outcome right on the camera LCD.
If you want so: it's consistent "out of the box" and this is why I couldn't care less about a "raw-raw"-histogram. The P1/C1-approach simply works... in terms of practicability.

But arent you mixing source representation with target representation?

The camera sensor clips when the camera sensor clips, no matter WB. Sensor clipping is something we usually want to avoid or limit.

If any transforms or gains are applied to the image before a histogram is calculated it is harder to "calculate back" how close to sensor clipping we are.

If you have saved a good exposure to file, it is a trivial operation for offline editing programs to do calculations in floating point or increased integer precision so that WB can be done without clipping and with negligible loss of precision.

-h

[fdisilvestro]

Examination of the raw file in Rawnalize shows that the green and red channels are short of clipping. However, the WB multiplier for the green channel is 1.0 and that for the red channel is 1.7 as shown by Rawnalize. When white balance is applied, the red channel will be clipped, whereas it is intact in the raw file. One could decrease exposure in the camera so that the red is not clipped during normal white balance with a red multiplier of 1.7. However, a better approach is to use multipliers all less than one so that no white balance clipping can occur and one can get a better signal:noise. One can do this directly in DCRAW, but in ACR one would decrease the exposure.

P.S.
In the example shown, it was not possible to eliminate red clipping by any reasonable decrease in exposure, becuase saturation clipping was occuring during rendering into the relatively narrow Adobe RGB space, which is the widest space available on this camera. One could eliminate such clipping through the use of ProPhoRGB.

As you point out at the end of the message, the red clipping is not due to wrong exposure, but an out of color gamut in the AdobeRGB color space. IMO the best approach is to use the ProphotoRGB color space before any exposure correction.

The histogram from Rawnalize show luminosity values for each channel, so one can determine if there was saturation at the sensor level, but does not give information about saturation, since the data hasn't been converted yet to a color space.

In any case, reducing exposure in the raw converter should not add noise. Doing it at exposure time will reduce signal to noise ratio.

Just a comment about using multipliers lower than 1 for white balance: This works fine as long as there are no blown out channels, otherwise it could produce color cast in those areas

[tho_mas]

But arent you mixing source representation with target representation?

I don't know... possibly.

The camera sensor clips when the camera sensor clips, no matter WB. Sensor clipping is something we usually want to avoid or limit.

correct.

If any transforms or gains are applied to the image before a histogram is calculated it is harder to "calculate back" how close to sensor clipping we are.

but why would I want to "calcuate back"? As you said: when the sensor clips it clips (so we stop down). The situation that the sensor is actually not clipping but the histogram yet shows clipping (due to a conversion into a color space such like AdobeRGB or sRGB) simply does not exist on a Phase back. At least not as long as you white balance while shooting. You could produce "fake"-clipping in the histogram although there is no sensor-clipping when you set a completely wrong WB while shooting (as demonstrated above).
This is why I said "WYSIWYG"... when you set WB accurately while shooting (so that no further WB adjustment is needed in post) you can effectively see how far you can increase exposure when shooting.

[madmanchan]

The problem with WB being reflected in the in-camera histogram is that it can lead you to believe that a channel is clipped when in fact it is not.

The other problem is that there may be other color conversions in play (e.g., a transformation from the camera primaries to standard primaries, like sRGB / Adobe RGB).

Bill has a good example with the yellow flower. Many cameras (especially recent Canon models) have a much weaker red channel compared to the green channel. Even when photographing a red rose under bright outdoor conditions, the native green channel will clip first, before the red. Using white-balanced camera histograms, with possible additional color conversions on top of it, make it very difficult to predict how to optimize the exposure.

The fact that WB has to be applied at some point (e.g., during raw conversion) is completely separate from the idea of an in-camera histogram that helps the user optimize exposure (e.g., by maximizing signal-to-noise).

[tho_mas]

The problem with WB being reflected in the in-camera histogram is that it can lead you to believe that a channel is clipped when in fact it is not.

not on a Phase One back (I guess: not on any MFD back)... only if the WB set when shooting is way off.

The other problem is that there may be other color conversions in play (e.g., a transformation from the camera primaries to standard primaries, like sRGB / Adobe RGB).

not on a MFD back.

The fact that WB has to be applied at some point (e.g., during raw conversion) is completely separate from the idea of an in-camera histogram that helps the user optimize exposure (e.g., by maximizing signal-to-noise).

see my last post.

[fdisilvestro]

Tho_mas

Are you claiming that WB affects Raw data in MFD back? It is not metadata?

[tho_mas]

Tho_mas

Are you claiming that WB affects Raw data in MFD back? It is not metadata?

no. WB affects the histogram... not the actual RAW data.
but as the histogram is built off of the actual RAW data and is only altered by gamma (to be more precise: a gamma 1.8 default color space that corresponds to the default color space in Capture One) and WB it's exactly what you see in the RAW software. (Of course only if you set the correct settings in the RAW software.) Therefore a WB while shooting is not counterproductive ... quite the opposite: it helps to expose as much as possible without clipping (at a given WB).

[fdisilvestro]

While I agree that you method reduce the amount of time spent in post processing, the fact is that what you describe is not a "raw histogram"

Gamma does not define a color space and it affects the middle values but not the extreme values, so if you are after the clipping point, it will make no difference

A raw histogram shows only the luminance level in the channels, there is no info about color saturation. It is only when you convert to a color space when a non clipped sensor channel may result in a clipped channel because of saturation in the output color space. As long as the original luminosity value is not clipped, you can adjust in post processing (of course, it will be time consuming)