Camera White Balance and Post Processing

[bjanes]

accuracy
     n 1: the quality of nearness to the truth or the true value; "he
          was beginning to doubt the accuracy of his compass";
          "the lawyer questioned the truth of my account" [syn: truth]
          [ant: inaccuracy]
    * 2: (mathematics) the number of significant figures given in a
        number; "the atomic clock enabled scientists to measure
        time with much greater accuracy"

You don’t or you can’t? Pretty clear which is the case. Doesn’t matter. You continue to misuse and mangle a well defined word (accurate) despite my attempts, now you are telling us using a certain process in ACR is ‘standard practice’. I suppose you use that white bal card, click and never further adjust because that process produces perfect, accurate color?

If you understood the scientific definition of accuracy, you wouldn't have quoted a definition from a general dictionary. In science and statistics, accuracy and precision are strictly defined. See Wikipedia or a standard text. In Wikipedia, look at the target analogy. The number of decimal points has more to do with precision than accuracy.

In science and statistics, accuracy is the degree of closeness of measurements of a quantity to that quantity's actual (true) value. By doing as proper white balance, the appearance of the flower in the image is closer to what it was in the scene than if an inaccurate white balance had been used. Perfectly accurate color is not possible in current CFA sensors which do not follow the Luther-Ives conditions, but one strives to minimize the error. Use of eyeball white balance rather than measuring the white balance is absurd if one desires accuracy rather than a pleasing color for the flower. Pleasing can not be quantified, while color can (Delta E).

I agree there is no need to continue the discussion since you are not amenable to reason, as I have learned in previous discussions with you.

Regards,

Bill

[bretedge]

You change the WB as your first step in PP, but what is your white reference? If you were shooting in daylight, you could use daylight, but the color temperature of daylight varies through the day according to the position of the sun. According to measurements by a Kodak scientist, direct sunlight in the morning or afternoon has a CCT of 4300K whereas that at noon is 5400K. Daylight (a mixture of direct sun and skylight) is 6500K.

To illustrate, here is a shot of a flower taken at 9:37 a.m. on Sept 21 at a position of 42 N latitude using auto WB with the Nikon D3. The calculated zenith distance of the sun was 60 degrees. The camera looks at the actual colors in the image and not the color of the light. Since the image contains no neutral colors, the ACR WB is not accurate; ACR gives 4400K, which is renders the image as too blue.



Setting the color balance to daylight helps. ACR gives a WB of 5500K.



A more accurate WB can be taken from another image containing a white cloud as a reference point.



One can apply this WB setting to the flower image to obtain a more accurate rendering of the actual appearance of the flower.



For accurate colors, it is usually best to take a WB reading from a neutral card such as a WhiBal. One could take several readings throughout day day when shooting at different times. One could set the WB to render the most pleasing colors, but if it is important to have accurate color, a WB reference point is necessary.

Regards,

Bill

This is all fine and well but I adjust white balance not to achieve a technically perfect white balance according to the precise time of day and/or position of the sun, but to my personal preference.  I may want the scene to appear a bit warmer or cooler than it was in reality and I adjust white balance by "eyeing" it on my calibrated monitor.  I'm not a scientist - I'm a photographer. An artist.  As such, I take liberties with my art, one of which is setting a white balance that looks good to my eye. 

Some times, I'll even blend two images set at different white balances to achieve a cooler temp in some parts of the image and a warmer temp in others.  It boils down to personal preference, I guess.

One final point: Rarely do I ever use one of the white balance presets.  I prefer to move the slider in Lightroom until the white balance looks good.  I could care less what the number or setting is, so long as the final result is what I envision.

Good topic of conversation here, folks.

[Schewe]

Good topic of conversation here, folks.

Not for nothing but if you're gonna quote a post it would be useful to edit the quote to include the salient point (and not quote a post of multiple images).

There are a lot of ways to get your white balance optimal...it behooves you to know all of them including the use of the White Balance tool as well as manually adjusting the image visually.

[hjulenissen]

Pleasing can not be quantified, while color can (Delta E).

I believe that you can quantify subjective preferences and the impact of a given degradation by rounding up a representative group of people and doing a test. It is costly, time-consuming and prone to certain errors.

-h

[RFPhotography]

Setting white balance with the ACR white balance eyedropper is standard practice

Really?  I think I can probably count on 2 hands the number of times I've used that method of adjusting WB in ACR or LR.  OK, maybe two hands and 1 1/2 feet.

Your method of altering WB appears flawed.  Applying the 5900K setting from another image has produced a very unpleasant colour cast and made the image too warm.  Comparing the flower to the image you took the WB from, the reference image doesn't have nearly the warmth you've imparted to the flower shot. 

Oh, by the way, you need to clean your lens or sensor or both.  

You calculate DeltaE values on all your images, do you?
 

[digitaldog]

Really?  I think I can probably count on 2 hands the number of times I've used that method of adjusting WB in ACR or LR.  OK, maybe two hands and 1 1/2 feet.

So here’s an interesting reality check!

I shot a Passport under a Solux lamp. Measuring the illuminant with an EyeOne Pro I’m told the CCT value is 4176K. What does LR show for this value when the camera was set for Auto White Balance (As Shot)? 4100K (+7 for Tint). Not bad! Looks reasonable too.

What does the slider show after white balancing? 5001K with tint at +10. Too warm! And as some of you know, the Passport has a number of warm and cool white patches to use for seasoning to taste. Its quite possible to end up with something quite close to CCT 4100K (which I got simply by using Auto White Balance in the first place!).

Clicking on the same white patch in a different raw converter (Raw Developer), I get CCT 4232K.

OK Bill, which is accurate?

AWB looks great to me, the WB shot is too warm.

[Bryan Conner]

which is accurate?

I can answer this beyond any shadow of a doubt!!!!       YES is the correct answer!   

[RFPhotography]

Here's an example, Bill.  The first image was shot with AWB and the As Shot WB in LR is 4450, Tint -16.  A more 'accurate' white balance would have the wood more grey as in the second shot.  The 'accurate' WB is 5180, Tint -5.  While the second shot may be more 'accurate' it's definitely not as 'pleasing' as the first one - to me.  I like the blue hue to the wood in the first shot.  It's also more representative of the fact that the shot was taken in open shade.

[Tim Lookingbill]

Very good example demonstrating this issue, Bob.

Now boost saturation on the first one that's pleasing but inaccurate to get back the warmth of the pinky peach hue to the right. That's what gets confused in applying WB is that our eyes want to see an optical white instead of the overall warm filter like look of the "accurate" one underneath.

Film often exaggerated this effect playing cool against warm except by amplifying it with saturation instead of R=G=B or accurate looking WB. Violet blue whites next to warm pastels makes for pleasing color but not necessarily accurate.

[RFPhotography]

And if this were a 'keeper' Tim, I'd do that additional kind of work on it.  But this is just a throwaway.

[bjanes]

So here’s an interesting reality check!

I shot a Passport under a Solux lamp. Measuring the illuminant with an EyeOne Pro I’m told the CCT value is 4176K. What does LR show for this value when the camera was set for Auto White Balance (As Shot)? 4100K (+7 for Tint). Not bad! Looks reasonable too.

What does the slider show after white balancing? 5001K with tint at +10. Too warm! And as some of you know, the Passport has a number of warm and cool white patches to use for seasoning to taste. Its quite possible to end up with something quite close to CCT 4100K (which I got simply by using Auto White Balance in the first place!).

Clicking on the same white patch in a different raw converter (Raw Developer), I get CCT 4232K.

OK Bill, which is accurate?

AWB looks great to me, the WB shot is too warm.

I can't really determine which is more accurate merely from a screen shot re-displayed on my screen. Your custom WB with a temp of 5000K and a tint of +10 sounds a bit screwy. Rather than eyeballing the images, one should use an objective measurement. I took pictures of a ColorChecker with my D3, illuminated by 6 Solux bulbs. I rendered into sRGB with ACR using the AdobeStandard profile. Since the color checker contains a wide range of colors and neutrals, I expected auto WB to perform well, and it did.

Here is the As Shot image. It is the real image, not a screen shot. The Temperature reads 4000 and the Tint -11. The whites and grays are slightly off.



Here is the custom WB image, white balanced in ACR with the WB eyedropper on gray patch two. The Temp was 3900 with a Tint of -1. The neutral patches are more neutral.



One can check for accuracy with Imatest. The results are shown. The custom white balance shows better overall accuracy by a small margin. Note especially that the Delta C for the middle neutral patches is much better with the custom WB.

As shot WB:


Custom WB:


However, auto WB does not work well when the image contains a dominant strong color and no neutrals, as in my original flower shot. To simulate this condition, I took a close up of the color checker such that the frame was largely filled by the red patch. The black surrounding the patches constitutes a small amount of neutral. The As Shot WB as read by ACR was 3450 with a tint of -23. The color was way off as expected.



Here is the proper custom WB from the full shot of the checker with the erroneous Auto WB from the close up shown as in inset.



QED.

Regards,

Bill

[RFPhotography]

Bill, you're missing the entire point that people are trying to convey.  Either that or you're just ignoring it on purpose.  "Accurate" isn't always "best".

[bjanes]

Really?  I think I can probably count on 2 hands the number of times I've used that method of adjusting WB in ACR or LR.  OK, maybe two hands and 1 1/2 feet.

The number of times you have used the white balance tool does not concern me, since I have no way to evaluate your level of experience and technical proficiency. As Guillermo pointed out (post 32) the intent of white balancing varies. Sometimes technical accuracy is needed, while in other cases an artistic impression is required. What is your intent? If one were to conduct a survey of Micheal Tapes' (WhiBal) customers, the number of users using the white balance tool would be much higher than in your case. 

Your method of altering WB appears flawed.  Applying the 5900K setting from another image has produced a very unpleasant colour cast and made the image too warm.  Comparing the flower to the image you took the WB from, the reference image doesn't have nearly the warmth you've imparted to the flower shot. 

Really? As I stated in my post, my aim in the WB was for technical accuracy and not for artistic intent. Your subjective evaluations are not relevant, since you have no way to relate the color in the image to that in the scene. The white balance would ideally be accomplished by reading of a neutral in the scene, but no definite neutral is present. The best that can be done is to use the white balance from another scene that does include a neutral, in this a white cloud. See my reply to DigitalDog where auto white balance on a close up of the ColorChecker red patch was way off. By using the WB from a different image containing the entire chart, accurate WB can be obtained and the color of the red patch is much more accurate than with auto WB.

You calculate DeltaE values on all your images, do you?

Obviously not, but I do calculate DeltaE when calibrating my camera. DeltaE can not be calculated from most routine images, since reference colors are not available. I don't carry a spectrophotometer in the field, nor do most users, and your comment is gratutious. DeltaE includes luminance, hue, and chroma. DeltaC as shown in the Imatest color checker plots removes the luminance component. Further, an increase in chroma (saturation) is often desired, but shifts in hue are not. A faulty WB can produce hue shifts. The informed photographer is aware of such considerations. Are you?

Regards,

Bill

[bjanes]

Bill, you're missing the entire point that people are trying to convey.  Either that or you're just ignoring it on purpose.  "Accurate" isn't always "best".

Please refer to Guillermo's post 32. If you want to reproduce the color of a red Ferrari for a car brochure, accurate is best. The same applies to catalog reproductions of merchandise where the color of a garment must be accurately shown. The same applies to scientific work. If an artistic effect is desired, then by all means adjust the WB. Even here, an technically correct WB is often a good starting point. It is you who can't or won't read what I was saying regarding my intent for WB.

Regards,

Bill

[digitaldog]

Please refer to Guillermo's post 32. If you want to reproduce the color of a red Ferrari for a car brochure, accurate is best.

You’re confusing (again) ‘accurate’ with a color match. The numbers could (most likely will) be totally different from the measured (spectral to lab to RGB values) of the Ferrari and that which you end up with visually matching in on output referred device.

I can't really determine which is more accurate merely from a screen shot re-displayed on my screen.

Then there is something seriously wrong with your display! The WB image is way too warm while the AWB is pretty much a spot on visual match.

Bill, you're missing the entire point that people are trying to convey.  Either that or you're just ignoring it on purpose.  "Accurate" isn't always "best".

Worse, he’s making up a term (accurate) without any way to backup what that means, when its invoked, and how its calculated. If it looks right, its “accurate”. Why not just say, it looks right?

[bjanes]

Then there is something seriously wrong with your display! The WB image is way too warm while the AWB is pretty much a spot on visual match. At first, I couldn't believe that his neutral values could be off that much and was looking for an explanation. Apparently faulty technique.

There's nothing wrong with my display. See the calibration results. However, rather than posting a screen shot, it would be much better to post the actual photo.



There is something seriously wrong with your white balance. If you used the WB dropper, the RBG values in the neutral patches should be equal, while in fact they are far from equal. You should repeat your experiment.

Worse, he’s making up a term (accurate) without any way to backup what that means, when its invoked, and how its calculated. If it looks right, its “accurate”. Why not just say, it looks right?

Accurate (scientific definition from Wikipedia): "In the fields of science, engineering, industry and statistics, the accuracy[1] of a measurement system is the degree of closeness of measurements of a quantity to that quantity's actual (true) value."

The accuracy of a color measurement can be expressed as DeltaE or DeltaC as shown in my Imatest examples. Without any reference points an image that looks right could be far from the actual color values in the image.

This is not my first exchange with the DigitalDog. See "Does a raw file have a color space?". In that exchange he refuses to reply to logic, throws in red herrings, changes the topic, does not remain on task, and even declines to listen to the opinions of his own experts (Thomas Knoll and others).

What about the auto WB in the close up of the red patch? Is that accurate? Does it look right? One could compare the rendered color to the actual color of the patch and calculate a DeltaC.

I give up with the DigitalDog. This emperor has no clothes!

Regards,

Bill

[digitaldog]

There is something seriously wrong with your white balance. If you used the WB dropper, the RBG values in the neutral patches should be equal, while in fact they are far from equal. You should repeat your experiment.

Nope. Not so. As you can see, I clicked on a white patch! In fact, the RGB values change based on which white I WB on. Some more than others. Now if you were looking closely, you’d see I WB on the first white patch of the Macbeth. The 2nd is a better move. The difference in the rendering is far from subtle. Yet WB I did. BUT, in your own examples here, you WB on a cloud to produce an “accurate” WB (again, this is nonsense description). Since you seem to feel one can WB on any white, neutral or not, you will see as I did in this example, WB is not ‘accurate’ and which white you WB plays a big role in getting a preferred (pleasing) rendering. The white patch as measured in that Passport IS neutral. And yet, WB on that one patch, the result is a warm rendering which is not ‘accurate’ but more important, not pleasing to me as the image creator looking at the scene and the capture on my display. Had I shot this under candlelight, I may have preferred this warm rendering. If I wished to express the image shot under Solux as candlelight, this image would be fine.

WB is subjective. Why do you suppose a company like X-Rite put a range of white’s with warmth and coolness into the Passport? Accuracy is term that doesn’t belong in the conversation. That’s the bottom line. Your use of the term ‘accurate’ is misleading. As I said here, you may say you WB and its accurate and I can disagree. You can say its close, I can say its closer or not closer and its all subjective. You’ve as yet supplied no methodology of measuring scene colorimetry and providing any proof that its accurate to the output referred data you end up with. It matches what you believe is the original, great. Calling that accurate instead of describing this as subjective with no way to back up an accuracy metric should be dismissed.

[RFPhotography]

No, Bill.  I have read what you've written.  I simply reject the premise for the most part.

As I stated in an earlier comment (which you may not have read) there are times when an 'accurate' white balance is desired.  And some of the instances you mentioned are examples.  But I reject the premise that simply correcting WB to the same temperature as the lighting source will render the colours as most 'accurate' in the image in every situation.  Might in some, but not in others.  I do some copy work for artists.  I use an XRite Passport as a tool to try and make sure my starting point is as close as it can be.  But inevitably I end up adjusting from that starting point because the colours in the image may still not be 'accurate' even after adjusting to the 'proper' white balance.  That won't absolutely get you a colour match between the original object and the end image.  It simply won't.  I agree that 'accuracy' (i.e., a match) in colour reproduction is important in the instances noted.  I reject the idea that WB is the only or single way to achieve that.

In the image you keep coming back to of taking a WB reading off one image to apply it to another, I'll continue to say that methodology is flawed.  The lighting in that scene with the clouds is very different from the lighting you end up with in the flower image.  You like it, fine.  I think it's far too warm and unless you measured the colours of the flower there's absolutely no way you can tell me with any degree of certainty that the colours in the flower are 100% 'accurate'.  My bet is they're not.

[Tim Lookingbill]

From what I've seen from Andrew's posted WB test image compared to Bjane's is that camera's clearly have different quality and characteristic filtering (infrared?) systems that influence how a Raw converter finally interprets the appearance of each camera's AutoWB under any given light source.

Bjanes makes valid points about that. His camera's AutoWB sees the Solux lit neutrals as too red and so bumps up the green while Andrews sees it (assumptions based on observation) as too bluegreen and bumps up yellow orange/amber.

It's not been mentioned that image reproduction is nothing but a magic trick that fools the optics of human vision into seeing something that resembles reality which has certain emotional aspects that induce us to trip the shutter.

It's much like ventriloquism tricks the human auditory system by having the ventriloquist attenuate certain directional audio frequencies humans use to determine source and direction of a particular sound. (Capital Record's audio engineers put this knowledge to good use in coming up with their famous stereo imaged "echo chamber"). Measuring these mimicked frequencies with a machine won't necessarily deliver identical wave pattern formations but the actual sound was enough to fool the listener's ear due to their position to the ventriloquist.

A spectrophotometer only reads energy waves of light and assigns a number no matter what type of light source. There's no attempt or consideration in attenuating these spectral frequencies into fooling the eye. This is what Andrew refers to as "scene referred". It's accurate according to a spectrophotometer but it doesn't take into consideration human optic's persistent adapting to the surround effect of that scene that a spectrophotometer has no clue about.

So when we capture "scene referred" color that may be "accurate" according to a machine, when we view this out of its surround that our eyes adjusted for at the time we tripped the shutter and now view in our dark warm lit studio, the "accurate" color looks butt ugly on our calibrated display which makes it even more uglier from the display's attempt at artificially mimicking D65/6500K daylight.

[bjanes]

But I reject the premise that simply correcting WB to the same temperature as the lighting source will render the colours as most 'accurate' in the image in every situation.  Might in some, but not in others.  I do some copy work for artists.  I use an XRite Passport as a tool to try and make sure my starting point is as close as it can be.  But inevitably I end up adjusting from that starting point because the colours in the image may still not be 'accurate' even after adjusting to the 'proper' white balance.  That won't absolutely get you a colour match between the original object and the end image.  It simply won't.  I agree that 'accuracy' (i.e., a match) in colour reproduction is important in the instances noted.  I reject the idea that WB is the only or single way to achieve that.

I agree that simply setting the white point will not ensure an accurate reproduction of the image. However, it is a first start. I use the DNG profile editor to create a custom profile for my camera which improves the color accuracy as measured by Imatest ColorChecker. However, to get the most accurate results, I need to use a linear tone curve in ACR, and readily concur that a linear tone curve does not produce pleasing results.

In the image you keep coming back to of taking a WB reading off one image to apply it to another, I'll continue to say that methodology is flawed.  The lighting in that scene with the clouds is very different from the lighting you end up with in the flower image.  You like it, fine.  I think it's far too warm and unless you measured the colours of the flower there's absolutely no way you can tell me with any degree of certainty that the colours in the flower are 100% 'accurate'.  My bet is they're not.

Of course, the colors are not 100% accurate. Use of a custom profile helps, but since the CFA filters of any camera do not meet the Luther-Ives criteria, metameric error will occur and it can only be minimized by a profile. I agree that the use of a cloud is not the best way to measure white balance. However, my point was that if the image contains a predominant saturated color and no neutrals, automatic color balance will be thrown off. This was demonstrated with a close up of the red patch of the color checker and AutoWB. In this situatiion one can take the WB from another shot under identical conditions or even better, take a custom WB from a neutral target. With no accurate WB, you could adjust the image to produce a pleasing result, but without a reference point, you would have to rely on memory to get the color right. Color memory is not very reproducible.

I haven't done museum reproduction work, but I understand that the best results are obtained with a multishot back using apochromatic lenses and a custom ICC profile, perhaps in Capture One. What is your experience?

Regards,

Bill