Expose for highlights and.....?

[madmanchan]

This is why setting WB in raw conversion should be your first step. It will impact how you adjust exposure, among other things. It's also why WB is listed as the first adjustment in the CR plug-in and in LR as well.

[larsrc]

Shooting black and white negatives, I learned a good exposure technique, meter the shadows and stop down 2. This would ensure good shadow detail and leave highlight adjustment for printing when it could most often be handled.
Now, with digital, the larger horror is blown highlights. Has anyone worked out a scheme such as metering the highlights and opening up 1, 2, how many stops?
Thanks,
Bill
http://www.montanahi-line.net/
[a href=\"index.php?act=findpost&pid=207760\"][{POST_SNAPBACK}][/a]

It's a viable technique. The exact number of stops will vary depending on the camera and the colors involved in the highlights. I took a series of progressively more overexposed images of various single-colored items to see at what point RAW was blown. With the 350D, between 2.5 and 3 stops is allowable before unfixable color changes appear. Thus you could use the technique of metering for the brightest part *that you don't want blown* and stopping down 2-3 stops depending on what you've found works for your camera and the colour of the highlights (I found less latitude in blues, unfortunately). While bracketing will work in some cases, this method will allow for a good preset exposure for those pictures that you don't get a second chance at. Different situations have different best methods.

-Lars

[Tony Beach]

No it's not.  That's how white balance works.  All raw converters will do the same thing.
[a href=\"index.php?act=findpost&pid=209876\"][{POST_SNAPBACK}][/a]

Not at all.  NX can be set to apply WB gain from scratch, and I can change WB with Capture One LE with no overexposure penalty.

[Tony Beach]

I took a series of progressively more overexposed images of various single-colored items to see at what point RAW was blown. With the 350D, between 2.5 and 3 stops is allowable before unfixable color changes appear. Thus you could use the technique of metering for the brightest part *that you don't want blown* and stopping down 2-3 stops depending on what you've found works for your camera and the colour of the highlights...

-Lars
[a href=\"index.php?act=findpost&pid=209896\"][{POST_SNAPBACK}][/a]

I think you meant open it up 2 or 3 stops.  The actual number varies depending on the light, the color, and the camera.

[bernie west]

Not at all.  NX can be set to apply WB gain from scratch, and I can change WB with Capture One LE with no overexposure penalty.
[a href=\"index.php?act=findpost&pid=210077\"][{POST_SNAPBACK}][/a]

That's interesting.  I'm not sure of the process you are describing with NX, can you expand?

To clarify the Capture One thing - you are saying that you can set exposure to just short of clipping and when you change white balance it won't ever show clipping as a result?  That's an interesting feature, and one I have often thought from an academic point of view would be useful.  So I am envisaging a case where say the blue channel was just shy of clipping and the R and G are not clipping.  Then, change white balance such that it increases the blue multiplier, which would ordinarily cause the blue channel to clip.  But in this case, instead of increasing the blue multiplier, it would actually decrease the R and G multipliers in the correct proportions.  Is this what is happening in Capture?

[dwdallam]

That's interesting.  I'm not sure of the process you are describing with NX, can you expand?

To clarify the Capture One thing - you are saying that you can set exposure to just short of clipping and when you change white balance it won't ever show clipping as a result?  That's an interesting feature, and one I have often thought from an academic point of view would be useful.  So I am envisaging a case where say the blue channel was just shy of clipping and the R and G are not clipping.  Then, change white balance such that it increases the blue multiplier, which would ordinarily cause the blue channel to clip.  But in this case, instead of increasing the blue multiplier, it would actually decrease the R and G multipliers in the correct proportions.  Is this what is happening in Capture?
[a href=\"index.php?act=findpost&pid=210086\"][{POST_SNAPBACK}][/a]

If that is the case, what is the trade off?

[bernie west]

If that is the case, what is the trade off?
[a href=\"index.php?act=findpost&pid=210098\"][{POST_SNAPBACK}][/a]

If you're working on the raw data, then there should be no trade off.  Remember, a linear raw file is very dark and green.  The colour channels have already undergone a lot of scaling upwards to get to the nice bright image you see on your screen.  Sending them a little of the way backwards, shouldn't have any better or worse effect than shoving them around in the first place.

[Tony Beach]

That's interesting.  I'm not sure of the process you are describing with NX, can you expand?[a href=\"index.php?act=findpost&pid=210086\"][{POST_SNAPBACK}][/a]

In NX you can select a WB target, then the program applies a multiplier to the red and blue channels (e.g., B=1.3, R=1.6), and you can be fine tune the amount of multiplication in each channel.

To clarify the Capture One thing - you are saying that you can set exposure to just short of clipping and when you change white balance it won't ever show clipping as a result?  That's an interesting feature, and one I have often thought from an academic point of view would be useful.  So I am envisaging a case where say the blue channel was just shy of clipping and the R and G are not clipping.  Then, change white balance such that it increases the blue multiplier, which would ordinarily cause the blue channel to clip.  But in this case, instead of increasing the blue multiplier, it would actually decrease the R and G multipliers in the correct proportions.  Is this what is happening in Capture?

No, you can set WB and then adjust exposure compensation so that the individual channel does not clip; essentially, this is the same as what you are suggesting.

[bernie west]

In NX you can select a WB target, then the program applies a multiplier to the red and blue channels (e.g., B=1.3, R=1.6), and you can be fine tune the amount of multiplication in each channel.
No, you can set WB and then adjust exposure compensation so that the individual channel does not clip; essentially, this is the same as what you are suggesting.
[a href=\"index.php?act=findpost&pid=210109\"][{POST_SNAPBACK}][/a]

I don't know that it is.  As far as I know the process you are describing for Capture One is exactly the same for ACR and all the others I have played with.  You can adjust the exposure slider any time you want and it will change the amount of clipping.  This is totally independant of white balance.

The idea I am talking about is similar to what Dave Coffin does with DCRaw when you can opt to have multipliers greater than or less than one.  I would like to take it a step further and have it so that the multipliers are scaled so that the most saturated channel can't clip.  If I could find the time to revisit my 'C' skills I would probably do it.

[bjanes]

It does in ACR. If you get your highlights just where you want them, then change the WB, sometimes the highlight blow again. Many times you can crank up recovery though and offset the change. If not, then you need to mess with your exposure again.
[{POST_SNAPBACK}][/a]

White balance is obtained by multiplying the red and blue channels by a coefficient greater than unity. Typical RGB values for the Canon 350D are 1.392498, 1.000000, 2.375114 for tungsten and 2.132483, 1.000000, 1.480864 for daylight. Normally the red and green channels are considerably to the left of the green, but if the image contains strong reds or greens, these channels can get blown when the multiplier is applied. The green channel would never be blown, since the multiplier is 1.0.

As [a href=\"http://www.guillermoluijk.com/tutorial/dcraw/index_en.htm]Guillermo Luijk [/url] explains in his post on DCRaw, this can be avoided by using multipliers that are all equal to or less than one. For example, in daylight balancing one could use 1.00000, 0.46894, and 0.694436; the ratios between the multipliers are unchanged but no channel would be blown. With these multipliers, the image would be too dark, and one would have to apply a scaling factor of 2.132483 (about 1 f/stop) to the image after white balance. This can be accomplished by use of the exposure control. The same principles apply to conversion with ACR.

If the channels are blown with ACR during white balance, one can use the exposure control to apply a uniform scaling factor to restore the integrity of the channels (i.e use a negative exposure value). Depending on the image, the exposure control would be better than the recovery control, which is nonlinear and affects mainly the highlights. If you want to tone down the highlights, then the recovery control would be preferable.

Bill

[joofa]

White balance is obtained by multiplying the red and blue channels by a coefficient greater than unity. Typical RGB values for the Canon 350D are 1.392498, 1.000000, 2.375114 for tungsten and 2.132483, 1.000000, 1.480864 for daylight.
[a href=\"index.php?act=findpost&pid=210127\"][{POST_SNAPBACK}][/a]

What you are saying is would correspond to a WB matrix where off-diagonal elements are zero.
In typical implementations there are inter-channel dependencies. Therefore, off-diagonal elements are not zero (typically they are negative) to properly offset the R,G, and B channels.

[bjanes]

What you are saying is would correspond to a WB matrix where off-diagonal elements are zero.
In typical implementations there are inter-channel dependencies. Therefore, off-diagonal elements are not zero (typically they are negative) to properly offset the R,G, and B channels.
[a href=\"index.php?act=findpost&pid=210257\"][{POST_SNAPBACK}][/a]

What you say may or may not be true--I don't know. However, the multiplication factors are what are used in DCRaw and Iris with good results. What do you make of that?

Bill

[madmanchan]

What you are saying is would correspond to a WB matrix where off-diagonal elements are zero.
In typical implementations there are inter-channel dependencies. Therefore, off-diagonal elements are not zero (typically they are negative) to properly offset the R,G, and B channels.
[a href=\"index.php?act=findpost&pid=210257\"][{POST_SNAPBACK}][/a]

Depends on how you think about it. For pure WB you just need a diagonal matrix (which performs a non-uniform scaling operation, which is what Bill described). This produces neutral grays, as desired, but also produces awful colors because camera color response doesn't match human visual response. That's where camera color profiles come into play. If you want to characterize a RGB camera using a simple 3x3 matrix, then in general the off-diagonal elements of the matrix would be non-zero. You can either think of the color mapping as two steps: first a WB diagonal matrix applied, followed by a color matrix ... or as a single step where the WB and color matrix have been pre-multiplied together.

[madmanchan]

Bill, the source code for dcraw also has the color matrices for several cameras. They are not necessary for white balance but they are needed to get respectable-looking color.

[lovell]

Two points and a question:

1. Most (all?) RAW processors have the ability to bring back "blown" highlights. Just yesterday I was shooting in bright daylight and couldn't get ETTR exposure to acceptable levels without risking severe shadow noise. LR "Recovery" allowed me to pull back almost all the data in the blown clouds. I haven't done tests to see how "good" the data is, ie. how much color shift there is, yet, or how much leeway there is.
2. Bracketing is generally faster than tweaking an exposure using a histogram. Not as accurate of course, but faster.

Question: it seems that some are implying that white balance affects RAW - which I thought was not the case. I just checked my Canon's manual and it specifically says RAW allows me to change WB in post without degrading image quality.
[a href=\"index.php?act=findpost&pid=209804\"][{POST_SNAPBACK}][/a]

If you did recover "blown highlights" from your raw image, then they were not blown to begin with.  Once the tonal value reaches 255, or in the case of 12 bit/channel 4,095, then it it blown and blown for good and forever.  No amount of post processing is going to recover  that tone, and how could it?  Perhaps the in-camera histo showed "blown" but it actually may not have been.  By the way, when shooting raw, WB and other settings don't effect raw directly, however WB will influence the exposure as shown on the histogram.  One doesn't have to get it perfect when shooting raw, but getting it close would show benefit.

As to WB and the Canon manual, you misunderstand my comments.  I never said shooting raw and with the wrong WB is detrimental to the image, nope never said that.  What I did say is that if you shoot with the wrong WB, it WILL EFFECT EXPOSURE, and that might mean an EV that is not beneficial to your particular picture.  For example, if you take the same exact shot 5 times with 5 different WB settings, the resulting 5 histograms will be different.  So your Canon manual and what I wrote are both correct.

Bracketing is not that much faster, and if it is faster and less accurate then why even do it?...if you practice, you can adjust exposure on the fly and very fast.  I've never blown a shot by applying EC.  Applying EC per the histogram is a lot more accurate, and a shot worth taking is a shot worth that accuracy, yea?

[lovell]

This has been posted already somewhere above. It is still nonsense.
[a href=\"index.php?act=findpost&pid=209821\"][{POST_SNAPBACK}][/a]

Let me be more accurate....WB will effect the histogram.  Take the same exact shot with the same exact exposure values but different WB settings and you'll get a different histogram for each WB setting used.

[larsrc]

I think you meant open it up 2 or 3 stops.  The actual number varies depending on the light, the color, and the camera.
[a href=\"index.php?act=findpost&pid=210078\"][{POST_SNAPBACK}][/a]

Yes, you're right, open up. I don't know that the exact number will really depend on the light (assuming raw), but certainly color and camera, maybe even lens.

-Lars

[bjanes]

Bill, the source code for dcraw also has the color matrices for several cameras. They are not necessary for white balance but they are needed to get respectable-looking color.
[a href=\"index.php?act=findpost&pid=210298\"][{POST_SNAPBACK}][/a]

Eric,

Yes, on looking at the source code for DCRaw, I came across a list of color matrices for various camera and noted that Dave Coffin thanked Adobe for supplying them. I presume Thomas Knoll uses these or something similar in Camera Raw. Are these matirx operations commutative (IOW, can they be applied in any order)?

Bill

[Panopeeper]

Let me be more accurate....WB will effect the histogram

Well, you need to be much more accurate, but before that you need to gain a better understanding the subject. You posted just above

What I did say is that if you shoot with the wrong WB, it WILL EFFECT EXPOSURE, and that might mean an EV that is not beneficial to your particular picture

NO, it WILL NOT affect the exposure, and that is good so.

[madmanchan]

Yes, on looking at the source code for DCRaw, I came across a list of color matrices for various camera and noted that Dave Coffin thanked Adobe for supplying them. I presume Thomas Knoll uses these or something similar in Camera Raw. Are these matirx operations commutative (IOW, can they be applied in any order)?
[a href=\"index.php?act=findpost&pid=210652\"][{POST_SNAPBACK}][/a]

White balance, then apply the camera matrix. Matrix multiplication isn't commutative (generally) so the order is important in this case.

Example: you take a picture of a gray card. The actual camera R, G, B values are, say, 0.4, 0.7, 0.3. So you do white balance by applying scale factors of 7/4, 1, 7/3. This can be written as a diagonal matrix W = {{7/4, 0, 0}, {0, 1, 0}, {0, 0, 7/3}}. Then apply the camera color matrix C. So the combined operation is XYZ = C * W * (camera RGB). Once you're in XYZ space you now have colorimetrically-defined values and you can convert to whatever color space you want to work in (e.g., ProPhoto RGB with linear gamma, in the case of CR).

The DNG 1.2 spec has a more complete description of this process in one of its appendices.