« One Brave Year | Main | What, Me Argue? »

Monday, 15 December 2008

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00df351e888f88340105366fcf26970c

Listed below are links to weblogs that reference White Balance, Part I: Accurate Color:

Comments

Nice article.

You and your readers might be interested to learn about a feature of the Olympus E3 and E30 which, as far as I know, is unique -- but which other camera manufacturers would be well advised to copy.

Most DSLRs analyze the image on the sensor to try to guess at the correct white balance, which can lead to errors in guessing the color of the ambient light, especially if the subject is itself colorful.

The Olympus E3 uses a much more direct method. On the body of the camera is a small, opaque, neutral colored panel. This opaque panel takes on the average color of the ambient light which is measured by a sensor behind this panel.

This is exactly the same principle as the "Expodisc" except that on the Olympus E3 it happens automatically without having to fiddle with custom white balance.

In my experience the E3 has more accurate color balance under difficult conditions, especially colored subjects, than other cameras.

Thanks for great photo blog.
- Eric Brandon

My question for part II is what variations from technically correct color rendition can lead to more pleasing pictures indoors in modest light levels. The blue channel is the noisiest -- should I boost it to the relative strength it would have at 5600K? And people don't seem as much concerned about accurate rendition of facial tone as interested in how the final result compares to some personal ideal.

Another question -- color is a three dimensional space, yet there are only two parameters available for color correction in camera, and only the fanciest raw file developers offer anything more. From using the Color-Checker card with Norm Koren's toolkit, I found that the correct match has to be done at a particular luminance level, Y, so that the comparison of the relative color coordinates, A and B, can be meaningful. Does setting color balance in too bright or too dark an area of a scene lead to under or over saturated colors, or is this purely a theoretical concern in practice?

regards,

scott

Beside documentary photography [incl. commercial, advertising and police] accurate colour - defined as 'the colour in the scene as would be agreed upon by most people being there and without major eye deficiencies' - is a necessary starting point to change the colour [selectively] for creative purposes. Curiously something advertising photographers know and have done for ages.

The death of filters is always discussed with digital capture, particularly for WB correction. But for extreme color shifts (like low Kelvin tungsten) we are forcing the blue channel into overdrive resulting in noise. Personally I haven't ventured there but there is still value in on-lens filtration. No?

Eric,

The E1, which delivered excellent AWB, had this direct incident light sensor, and if I'm not mistaken Nikon and some others have it in some models. It isn't the same as a custom white balance though (it's often better) because it uses the incident light information *in combination* with the same through the lens data used by cameras that lack this feature. BTW, don't wear a hat or cap with a big brim or you'll block the light and defeat the sensor.

Scott, I'll look into your first question. For the second, you don't have to be terribly precise, but the test target should be exposed to give "light gray" value: not up near white, but nowhere near as dark as an 18% gray card. In the illustration you'll see I've held it against a medium-dark background area and the camera on "A" placed it well.

Marek, filters can't "add" anything to the information that gets to the sensor. A blue filter doesn't add blue light, it just puts a partial block on all the other colors so blue represents a higher proportion of the (now attenuated) light that gets through--I can't think of how that would be better than letting all the light through and dealing with it in software. As I recall, a filter to convert daylight color film for tungsten light cost nearly two stops of light.

Anyone know if (why not?) the whitebal cards aren't available in UK|EU?

One thing I've noted when in a fluorescent environment is that I get better, more consistent results when my custom white balance is set at a shutter speed below a 30th of a second at the time of recording. There is a constant flicker in the spectral wave length which our eyes don’t always see which pulse at the 60th of a second level. This is more of an issue when in a smaller room lit by fewer lights as the blend of a larger room can balance out the spectral flicker.

Mike

"Beside documentary photography [incl. commercial, advertising and police] accurate colour - defined as 'the colour in the scene as would be agreed upon by most people being there and without major eye deficiencies'"

This comment hits the nail on the head with regards to white balance. Unless you have the object you photographed in front of you how can you tell if the color is correct?

From my own experience of product photography, I find camera sensors react differently to different colors. If I photograph three colored objects at the same time, for instance, red, green and purple. I find it almost impossible to match at least one of the colors regardless of white balance. It's like doing a puzzle, you'll get the red correct and the purple will be wrong.

Any thoughts on this.

Tim

A few responses to the above:

Carl, I think Marek is right about the use of filters. Assuming conditions permit it (i.e., you use a tripod with a static subject or there is enough light that you can still get reasonable shutter speeds with the filter on), using a filter should reduce the amount of noise in the image. While the filter isn't adding anything, by holding back some of the light (and therefore requiring an increase in the overall exposure), the light channels are more evenly exposed and you don't have to "boost" one channel at the expense of noise. Dante Stella has an excellent write-up on this, so I will just point to his site: http://www.dantestella.com/technical/lightbalance.html

***

It was very interesting to compare both of Carl's posts on this subject, since using a gray card vs. an ExpoDisc presents various advantages and disadvantages.

When using a gray card, you aren't really adjusting the white balance until afterwards in post-processing. This means that while you are taking pictures, your exposures could be slightly off, since the histograms displayed on the LCD will be based on whatever your camera's auto white balance feature selects and will likely be wrong by some margin. If you base your exposures on the histograms, you could be over- or underexposing your pictures by a bit. In the worst case, you could be blowing out a channel without realizing it. On the upside, the gray card is small, light, fast to use (since you just need to take a picture of the card) and works in all kinds of light.

With an ExpoDisc, you are actually setting the white balance at the time of capture. This means you need to put the ExpoDisc over the lens, take a white balance reading, then take off the ExpoDisc and keep taking pictures. Not a big deal, but taking a white balance reading takes longer than just taking a picture of a gray card. On the upside, your exposures will be right, because the histograms will be based on the "right" white balance. On the downside, I am sure that I read somewhere that ExpoDisc's don't work in certain lighting situations (perhaps under household tungsten lighting?) and they are also expensive (an original ExpoDisc in 77mm size costs $100, while an "ExpoCap" from the same company costs $60 for the 77mm size. By comparison a "WhiBal" (see ad at the top-left of this page) costs about $20 for the most commonly used size.). Can anyone refresh my memory on the situations where an ExpoDisk isn't 100% reliable and what the reason is?

***

Michael, is the same true of tungsten lights as well? I believe you need to pick a multiple of the frequentcy of the electricity where you are (it's 60 Hz in the US and 50 Hz in Europe) to get an optimal reading. So 1/30th or 1/60th (or slower multiples) should work in the US, and 1/25th and 1/50th (or slower multiples) should work in Europe. I seem to recall this being an issue with some handheld light meters as well - if their sampling period was too short, they would give inconsistent meter readings under artificial lighting.

***

Finally, for those on a budget, take a look at this: http://www.bhphotovideo.com/c/product/300868-REG/Porta_Brace_WBC_White_Balance_Card.html
This is more durable than some of the cheaper gray cards (which are just painted card stock), as it is made of flexible plastic. You can't tell from the B&H site, but the card is about 5x7 inches if I remember correctly and can be cut into smaller sizes or multiple cards as needed. For the record, gray cards are supposedly slightly more accurate than white cards.

Best regards,
Adam

@Dave Pawson
You can easily order WhiBals directly from Michael Tapes at RAWWorkflow. I've ordered from him several times and have never been disappointed. It is quite normal that small businesses do not have the ressources to cater for distributors everywhere in the world.

@Tim
Trouble is that people think that colour as a physical phenomenon is as solid as the number 2, which will hold everywhere in the universe. They conveniently forget it has to be perceived, processed and put back into words. What I see as red you might see as orange and Mike might see it as purple. We might then not even agree upon the label we give to the same colour.

And now add in the various output devices and environmental lighting [not scenic but on the photo/object].

Scott, to your second point - it's a dgrees of freedom thing. White balance is about getting the relative position of the 3 colours balance, so I only need to change 2 relative to the 3rd. Thus, a 2-paramenter colour shift is all that is required.

Nit pick: The degree symbol shouldn't be used with Kelvin temperatures (as in 5000° K). Because Kelvin temperatures are referenced to absolute zero they are as specified as units, not degrees, so the correct nomenclature would be either 5000 Kelvins or 5000K.

Trivia, I know :)

And a tip: You can use white mat board as a white balance card. Since there are *many* shades of white mat board available, a little experimentation will get you white balance cards that yield neutral, warm, cool and every white balance variation in between, if you like. My local art supply store gives me scraps for free.

Tungsten, especially household tungsten, is so red that, unfiltered, you have a choice between lots of noise in the blue channel, or blowing the highlights in the red channel. Obviously noise in the blue is better, but it's not GOOD. Filtering lets you bring the channels more into coincidence, so that all of them are decently exposed -- at the cost of needing an even higher ISO or a longer shutter speed or a wider aperture. It's all tradeoffs.

Thanks for the nitpick, Mark. I've made the change.

Mike J.

Adam..."where an expo-disc isn't 100% reliable...". What comes to mind is that the expo needs to be used like an incident meter...pointed toward the light illuminating the subject. That's sometimes inconvenient or impossible. The WhiBal works on reflected light. Also, you have to turn off autofocus to use the expo,and then remember to turn it back on!

John, the issue with auto focus is camera dependent. Nikons will fire when in Custom White Balance mode. But I in my use the WhiBal works better.

I have a question. When you took the photo of the WhiBal card above, did you have the camera set to auto white balance and then continue to take photos while set to Auto WB, or did you originally choose a WB preset that was close, and continue to shoot with the same preset for uniformity in processing later? Seems like Auto WB would keep changing as you move around and take photos, removing all advantage of being able to reference the shot of the grey card in post process. (I'm just trying to learn how to take better photos.)

From my own (brief) experience with digital colour images, first with grey cards, then with the ColorChecker chart and ACR calibrator scripts, I've come to the conclusion that it's difficult to obtain accurate colour but it's more important to get pleasing colour.

I spent days trying to get Adobe ACR colours looking like Canon DPP's because no matter how I played with white balance and tint controls I couldn't quite get the sickly yellow-green out of skin. ACR 4.5 introduced the concept of camera profiles, in essence the ability to map colours, greatly simplifying the process.

"Seems like Auto WB would keep changing as you move around and take photos, removing all advantage of being able to reference the shot of the grey card in post process."

I generally leave the camera in AWB because I like to see what the camera's software goes for: sometimes it's informative, and a lot of the time I like it. You don't lose the advantage of the test card. With any good RAW converter, after using the card to set WB of the test file, you can apply it to all appropriate other files together. There are several ways to do this. One way with PS is to set the test file's white balance, but change nothing else, then drop back into Bridge. Now select all the files in the same lighting conditions, and from the drop-down menus order Edit-->Develop Settings-->Previous Conversion. Bingo, all the files set to the new WB.

Tim:
"From my own experience of product photography, I find camera sensors react differently to different colors. If I photograph three colored objects at the same time, for instance, red, green and purple. I find it almost impossible to match at least one of the colors regardless of white balance. It's like doing a puzzle, you'll get the red correct and the purple will be wrong.

Any thoughts on this."

I would try creating ICC color profiles. Put an IT8 target in the subject position and take a sample shot, then go ahead with the product shots. Any change of lighting (or of camera settings, if not shooting RAW) requires another IT8 shot. Lights age, so regular updates of profiles should be done. Easiest is probably to do one for each shoot and keep the profile with the images. RAW processing must be identical for each target and all its associated images.

There are several software apps that will convert the IT8 shots into ICC profiles. The profile is a map of the differences between the color values from the film scanner or the camera sensor system and the 228 color and 24 gray blocks of known value on the target. Assigning these profiles to the matching images will show the correct colors in color profile aware apps, like PS. Converting the profile to a standard, such as aRGB or sRGB, will give the correct colors for users/uses that don't have your individual shoot profiles available (insofar as the gamuts of the chosen profile and display method support them).

The technology takes into account capture and reproduction processes, like film, that have different response curves for the different color channels. Digital generally doesn't require such detail, which is why simple, single neutral references are often used, but in your case, it appears that isn't sufficient.

I haven't tried this with digital, but it works wonders with film. I use IT8 targets from http://www.coloraid.de and VueScan to create and apply the profiles.

I have a small set of examples at http://www.moosemystic.net/Gallery/tech/Scan/VuesProf/

Holding an IT8 target up next to the screen image of the IT8 target shot clearly shows how much more accurate the profiles colors are to the original subject than the unmediated scan.

Where it will possibly not work is with subjects with unique responses to light, where the colors our eyes see are not simple reflection, but involve refraction, iridescence, metamerism, fluorescence, etc.

Moose

Could someone please fill me in on what the benefit of a specially designed white card might be for setting your white balance, instead of using an old Kodak 18% card I've got lying around (or even just a sheet of paper)?

Well shooting digital makes colour easier (I struggle to get good colour when scanning negative film) but sometimes it all gets too much and I persuade myself that the shot was intended as B&W anyway.
BTW my Olympus E300 had dreadful auto WB! so maybe their pro models needed that extra sensor.

Cheers, Robin

Richard,

While having the auto white balance jump around after the picture of the gray card is taken may affect the histograms (and thus the accuracy of the exposures), it shouldn't create any additional work in post-processing as far as white balance is concerned. In post-processing you don't increase or decrease the color temperature of all the pictures by a fixed amount, instead you set the color temperature (and tint) for the first picture using the eyedropper tool, then copy those white balance settings to all of the other pictures taken in the same light. So in that respect it shouldn't matter what the camera's auto white balance setting picked for each shot, as those settings will just be overwritten.

Best,
Adam

Richard - Assuming you're shooting RAW files, it doesn't matter what mode the camera is in. When you open the RAW file, the white balance needs to be set, so to set it you neutralize your picture of the WhiBal card. It doesn't matter what the camera actually recorded.

Now if you're shooting JPG, you may need the WhiBal in every shot as you turn or face different lighting circumstances. So unless you take the shot twice every time, you may still have issues, although they would probably be greatly lessened.


(PS - Mike...I may be re-answering something, so feel free to circular file this...)

Because I take a lot of pictures under compact fluorescent light (which has very little blue output,) I've done quite a bit of experimentation with custom white balance and balancing filters. Setting CWB with a WhiBal or Expodisc is the worst of all worlds, because it can mask clipping in the red channel and it amplifies blue channel noise to a unacceptable level. An 81A (blue filter) will preserve the DR of the camera, but the increased exposure that's needed will result in all three channels being fairly noisy. You have to have (or add) a LOT of light to see the benefits of the 81A. A CTO-gelled flash (the gel matches CF fluorescent bulbs well enough to make fine tuning of WB during PP as snap) balanced with a blue lens filter will yield a low-noise, high-DR image that otherwise wouldn't be possible.

Note: this alone is justification for getting yourself a 580 EX II instead of a 430 EX (or the Nikon equiv., of course) for Xmas...the bigger GN definitely helps when you start adding gels and filters to everything. ; )

Finally, a tip: For people looking for a cheap alternative to the WhiBal that actually works (unlike the vast majority of enthusiast DIY WB solutions) Sintra PVC in light grey is very neutral, much tougher than a cardboard gray card, and very cheap. Can be purchased online in smallish batches at a very good price.

My experience is to take into account too precise white balance with a pinch of indifference because in most scenes the neutrality of white balance can detract from the atmosphere one is trying to capture, even in product shots the numerous tweaks to produce something acceptable tend to lead away from the perceived desirability of getting colour balance correct. Afterall - and this is no criticism - In colour photography generally we tend to like to see a boost in saturation/shift of colour because that's more pleasing and we cannot say for sure what the colours were actually like at the time. I do like to explore these incredible capabilities in digital capture, but, with the variables in process of getting a print on view the range of tweaks that can be made often turn into a Quagmire of "is what I am doing the best I can do". So I go for a walk and try to forget half of what I think I know.

The missing part of the colour equation is this: the red, green and blue cones in your eyes -the levels of whose excitations give colour information to the brain - have very different spectral sensitivities to the RGB filters in your camera's Bayer array, which is one reason why Tim (featured comment) can't get all three products to be the right colour. Then again the RGB elements of the display device differ from both in their spectra.

Since biological spectral sensitivities differ measurably even between different "normal" individuals (you can measure them with various tests that ask an individual to compare say broad spectrum green light and monochromatic green light, adjusting the wavelength of the latter until the two sources appear to be the same colour) there's no way that a simple two-parameter adjustment is ever going to give more than an approximation to correct colour, for the 'average' person.

I understand that dichromats (genetically missing a type of colour sensor) were used militarily for spotting camouflaged military hardware because the camouflage designers had done a good job matching the perceived colours of the nets and cloths to the natural background but only when looked at by normal trichromatic eyes. The atypical colour perception of a dichromat meant that this colour matching was totally void - so the camouflage stood out because to that viewer, it was to them, literally, a different colour.

It's common practice now to apply a massively multi-variate output transformation (we call it a profile) to the display device, be it screen or printer, rather than just a two variable spectrum adjustment. It would make more sense to be able to apply a full affine transformation between the raw colourspace of the camera sensor and a standard colour space based on the spectrum of the light source in use at the time.

Richard,

I dug around a bit and managed to find this video on the rawworkflow.com website: http://www.whibalhost.com/_Tutorials/Photoshop_LR/07/index.html

It shows how you can easily use a shot of the WhiBal to fix white balance afterwards, even if you leave the camera in Auto White Balance while you're actually shooting.*

Seeing is much easier than explaining. Dontcha just luv the Internets sometimes?

Best,
Adam

* Be sure to start at 1:00 and stop at 2:35. At 2:35 he starts talking about how to auto-adjust the exposure on multiple pictures at the same time and it's a little confusing at first, since it initially sounds as though he is still talking about setting the white balance.

Of course, all of this is for naught if your lighting falls outside the range of the temperature-and-tint model used by Apple and Adobe.

Indeed, Adobe and Apple seem to completely ignore the in-camera WB data for at least some cameras. I run up against this in infrared photography with my modified G5.

http://homepage.mac.com/n6mod/infrared/ (or click my name above)

Adobe doesn't seem interested in dealing with the results of converted cameras, and Apple has never responded to any of my queries.

I have had Big Names In Digital Photography demonstrate their ignorance of how cameras (digital or otherwise) work by insisting that what I expect is impossible.

Yet dcraw works just fine...

Starting from Adobe Camera Raw 4.5 and Lightroom 2.0, Adobe introduced camera profiles for use with their converters. This change is deeply powerful in terms of color matching and control.

You can start with a shot of the Gretag Macbeth Colorchecker with the 24 squares and it can generate a profile based on that calibration. You can then further map the colors to your liking.

In regards to Zandr's issue with IR, there is a white balance calibration slider in the profile editor to adjust the white balance mapping.

I suggest anyone who is even mildly interested in color in these raw convertors to read the FAQ:
http://labs.adobe.com/wiki/index.php/DNG_Profiles_FAQ

The comments to this entry are closed.