Color Balance and Clipping

[Henrik Paul]

This might sound to some of you as unnecessarily hi-fi or perhaps pixel-peeping. But I'm asking this just out of pure curiosity.

Assume that I'm shooting RAW (which I do), with an automatic color balance (which I do also). Now, even if I shoot in tungsten light, the resulting image should be adjusted as neutral-white. The same in incandescent light. I can even adjust the color balance in my favorite RAW-converter. But the real color of the light might have been heavily yellow or blue.

How can I be sure, that either of the red or blue channels aren't clipped, even if the histogram claims it not to be clipped (because of the adjustment in the color balance)? Is there a way for me to set the camera to record the "native color balance" (if there is such a thing), so that I, if I wanted to, could be absolutely sure that the channels aren't clipped?

Hopefully someone understood my point...

[32BT]

I understood your concern, which is valid, and the short answers are: You can't and no.

You might want to fix your white balance setting to cloudy daylight, or use the colortemp of your locale and/or the colortemp of your screen calibration. This should give you a fixed internal rendition used for the histogram. You can then "judge" the light in your image because the fixed white balance can become a reference. Tell tail signs of a problem are spikes at the end of the histogram data (not the far ends of the graph axes, but the far ends of the graph data). Unfortunately these are also produced by sharpening and smoothing settings. Which settings correspond best with your preferred final RAW conversion is unfortunately a matter of trial & error.

[John Sheehy]

Is there a way for me to set the camera to record the "native color balance" (if there is such a thing), so that I, if I wanted to, could be absolutely sure that the channels aren't clipped?
[a href=\"index.php?act=findpost&pid=84096\"][{POST_SNAPBACK}][/a]

Most digital cameras with Bayer CFAs are most sensitive to green light, and least sensitive to red, and blue falls in-between.

With my Canon DSLRs, the red channel is about a stop less sensitive than green, and the blue 1/4 - 1/2 stop less.

In incandescent light, this usually translates to equal recording of subject red and green, and blue 1.5 to 2 stops weaker.  In green foliage light, the red channel can be very weak and show exaggerated noise, especially if WB'ed for the subject.

To get the camera's histogram to closely resemble the native balance, you need to custom WB the camera with a certain magenta cast (in a certain light), instead of a white card in an arbitrary light.  The computer monitor could be used, you could adjust the color of a large rectangle on the screen until the RAW data is "white", and WB the camera off of this (review will have a cyan-ish green cast, though).  You also need to turn down the camera's contrast and saturation to get a histogram that is close to the RAW.  With my cameras, there is generally 1/3-1/2 stop more highlights in the RAW green channel than the point at which the review shows clipping.

Unfortunately, the histogram is still going to be green-weighted if it is a luminance histogram, but you have boosted the red and blue a bit with the custom WB, and the review image itself will reflect the WB fully.

[inspiration.in.print]

This might sound to some of you as unnecessarily hi-fi or perhaps pixel-peeping. But I'm asking this just out of pure curiosity.

Assume that I'm shooting RAW (which I do), with an automatic color balance (which I do also). Now, even if I shoot in tungsten light, the resulting image should be adjusted as neutral-white. The same in incandescent light. I can even adjust the color balance in my favorite RAW-converter. But the real color of the light might have been heavily yellow or blue.

How can I be sure, that either of the red or blue channels aren't clipped, even if the histogram claims it not to be clipped (because of the adjustment in the color balance)? Is there a way for me to set the camera to record the "native color balance" (if there is such a thing), so that I, if I wanted to, could be absolutely sure that the channels aren't clipped?

Hopefully someone understood my point...
[a href=\"index.php?act=findpost&pid=84096\"][{POST_SNAPBACK}][/a]

If you wont good white balance take a look at the Expodisc, It is not cheap but is the fastest and best way I have found.
If you are lucky you mite find one on Ebay

[bjanes]

This might sound to some of you as unnecessarily hi-fi or perhaps pixel-peeping. But I'm asking this just out of pure curiosity.

Assume that I'm shooting RAW (which I do), with an automatic color balance (which I do also). Now, even if I shoot in tungsten light, the resulting image should be adjusted as neutral-white. The same in incandescent light. I can even adjust the color balance in my favorite RAW-converter. But the real color of the light might have been heavily yellow or blue.

How can I be sure, that either of the red or blue channels aren't clipped, even if the histogram claims it not to be clipped (because of the adjustment in the color balance)? Is there a way for me to set the camera to record the "native color balance" (if there is such a thing), so that I, if I wanted to, could be absolutely sure that the channels aren't clipped?

Hopefully someone understood my point...
[{POST_SNAPBACK}][/a]

Henrik,

I think I may understand your point and I will take a stab at answering your question. You don't say what camera you use, but with Nikon DSLRs one can upload a unity white balance into the camera, which sets the red and blue multipliers to 1.0. If you look at the RGB histogram on the camera, they will reflect the RGB values recorded by the sensor and will give you an accurate representation of the channels. As John Sheehy pointed out earlier, the preview would appear quite green.

To achieve WB with a Nikon D200 under daylight conditions the blue channel is multiplied by 1.4 and the red by 1.8. If you are taking a photograph of a red flower, the red channel might record a pixel value of 150 (in 8 bit values) in the raw file, and the red channel would not be blown, but to achieve white balance the red channel would be multiplied by 1.8 and the resulting value of 270 would be clipped in the 8 bit white balanced rendering. However, this is no problem, since you can decrease the exposure adjustment in the raw converter to bring the red channel to 255 or wherever you want it (not taking the gamma correction into account). In practice, if the red channel is blown in a narrow color space such as sRGB, you should try to convert into a wider space, such as ProPhotoRGB rather than decreasing exposure.

Here are the multipliers for other Nikon cameras and color temperatures. As you can see, none of the tabulated color temperatures correspond to unity WB. As John pointed out the Canon cameras have similar multipliers, but these vary with the sensor.

[a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm[/url]

Bill

[Henrik Paul]

You don't say what camera you use, but with Nikon DSLRs one can upload a unity white balance into the camera, which sets the red and blue multipliers to 1.0.

Yes, I'm actually using a Nikon D200, so this solution sounds like worth a try. Although, what I didn't know is that the channels are always multiplied to compensate, never divided (at least on Nikon cameras, as the table shows). This means, on the preview histogram, if the R/B channels don't look clipped, they really aren't clipped, but if a channel looks clipped, it might or might not be actually clipped.

This solution makes much more sense, naturally, and therefore my original concern isn't actually a concern. I guess the engineers aren't paid for nothing

Thank you, Bill.

[32BT]

Yes, I'm actually using a Nikon D200, so this solution sounds like worth a try. Although, what I didn't know is that the channels are always multiplied to compensate, never divided (at least on Nikon cameras, as the table shows).

I completely forgot about those tables, even though Bill has referenced them on previous occasion.

What do you mean by "divided", I do notice some values less than 1.0...

[Blendenteufel]

Probably a practical approach is to keep bracketing your shots (maybe in 1/3 stop steps or even 1 full stop each appart).

In the end not only the WB setting but also the applied (contrast) curve, saturation and sharpening will all add their bit to the histogram. So to have the biggest latitude at conversion you would benefit from different exposure because you can pick the one with the highest shadow detail and intact highlights.

[32BT]

Probably a practical approach is to keep bracketing your shots (maybe in 1/3 stop steps or even 1 full stop each appart).

In the end not only the WB setting but also the applied (contrast) curve, saturation and sharpening will all add their bit to the histogram. So to have the biggest latitude at conversion you would benefit from different exposure because you can pick the one with the highest shadow detail and intact highlights.
[a href=\"index.php?act=findpost&pid=84288\"][{POST_SNAPBACK}][/a]

Agree, and the interesting question is: what does it mean to have "intact highlights" at the extremes of the WB scale if you don't want clipped channels?

[Henrik Paul]

What do you mean by "divided", I do notice some values less than 1.0...
[a href=\"index.php?act=findpost&pid=84286\"][{POST_SNAPBACK}][/a]

Would you look at that... You are absolutely correct, I must've missed those. Fortunately(?) they are only on the D2H, and none for the D200, so it's not a concern for me at least.

As a sidetracking note: Looking at the tables, I see huge differences between the same wb/environment settings between the cameras. For example 3.46 and 0.93 (a difference of 2.53) for the D2H, while there's only2.44 and 1.32 (1.12 difference) for the D200, in the same row of the table. This seems to be a perfect example that shows how different the colors on bayer filters are on cameras, even coming from the same brand.

[32BT]

This seems to be a perfect example that shows how different the colors on bayer filters are on cameras, even coming from the same brand.
[a href=\"index.php?act=findpost&pid=84292\"][{POST_SNAPBACK}][/a]

And/or the IR filter difference, perhaps even AA filter differences...

[bjanes]

Yes, I'm actually using a Nikon D200, so this solution sounds like worth a try. Although, what I didn't know is that the channels are always multiplied to compensate, never divided (at least on Nikon cameras, as the table shows). This means, on the preview histogram, if the R/B channels don't look clipped, they really aren't clipped, but if a channel looks clipped, it might or might not be actually clipped.

This solution makes much more sense, naturally, and therefore my original concern isn't actually a concern. I guess the engineers aren't paid for nothing

Thank you, Bill.
[a href=\"index.php?act=findpost&pid=84283\"][{POST_SNAPBACK}][/a]

Henrik,

Keep in mind that a multiplier of less than one is equivalent to division. The exposure control in Adobe Camera Raw is essentially a fractional multiplier. Here are some examples of a photo taken with a D70 of a highly saturated red flower. The colors are posterized since the images are uploaded as GIFs so that the type in the dialog boxes is sharp.

Here is a DCRaw conversion (a freeware raw converter) showing the raw file with no gamma or white balance correction. The red channel is slightly clipped and the red and blue channels are to the left, since there is no WB multiplier:

DCRaw
[attachment=1168:attachment]

Here is an ACR preview with default parameters. The red channel is clipped in sRGB.

ACR sRGB
[attachment=1169:attachment]

Here is a the same preview with ProPhotoRGB. The red is not clipped.
ACR ProPhotoRGB
[attachment=1170:attachment]

Here is an ACR preview with reduced exposure compensation to bring the red channel below clipping.

ACR sRGB Reduced Exposure
[attachment=1171:attachment]

The camera histogram shows the JPEG conversion into the selected color space. Unfortunately, the D70 does not have a ProPhotoRGB setting.

I hope this helps.

Bill

[John Sheehy]

never mind