Is Auto WB good enough?

Tom Reichner wrote in post #19464980
.
I found the image(s) I was talking about in the quote post (above).

I actually shot it at 6,500 Kelvin, not 6,000 as I had thought.

See how doggone blue it is, even at 6,500 Kelvin?

Hosted photo: posted by Tom Reichner in
./showthread.php?p=194​64980&i=i44849359
forum: RAW, Post Processing & Printing

Here is the edited version, with the color temperature cranked all the way up to 10,000 Kelvin. . I actually would like it a bit warmer, but my software simply won't allow for it to go any warmer with the color temperature slider.

Hosted photo: posted by Tom Reichner in
./showthread.php?p=194​64980&i=i82560117
forum: RAW, Post Processing & Printing

So this is an example of why it can be advantageous to shoot with the Kelvin setting instead of the AWB setting. . Trust me, the auto WB setting would have been much cooler than my 6,500 K setting. . Auto settings are often good when things are more or less normal, but when you get to real extremes, that is where automatic settings break down - they always seem to underestimate just how far from normal the conditions really are.

.

I thought that lower numbers are warmer and higher numbers are cooler?

Bob_A wrote in post #19464696
I still don’t understand why for any situation setting WB in camera matters when you shoot in RAW and it can be changed in seconds in a RAW converter.

If I have a grey card or Color checker in the scene I can change WB faster in RAW than fiddling with the camera.

If I’m outdoors and I want a huge burst at 5400K I can also select the images and change them all to 5400K faster in LR than in camera (I’m talking about ME … YMMV).

I don’t think anyone uses the results from AWB. It’s just that it doesn’t matter all that much whether you want to set WB in camera or after the fact.

I’m here to learn though. What am I missing?

I've been reading this thread because I know almost nothing about white balance. I've been doing it one way since I started digital, and I've always wondered if I was doing it wrong.

I just shoot raw on daylight. I was there, so I know what the light was like, and I can usually get there from daylight in post. It's muscle memory for me at this point. It works outdoors, and indoors, I'll pull out a flash if it looks tough.

After reading this thread, I think my dumb-luck solution is fine for what I shoot. It's good to be skeptical of auto-anything.

TeamSpeed wrote in post #19465047
I thought that lower numbers are warmer and higher numbers are cooler?

I think he meant that he would prefer a bit more warmth in the photo, which would entail an even higher WB value than his software permitted, >10000K.

Tom Reichner wrote in post #19464980

Hosted photo: posted by Tom Reichner in
./showthread.php?p=194​64980&i=i82560117
forum: RAW, Post Processing & Printing

This picture is good for illustrating the point, but it's also just a cool picture regardless!

.

TeamSpeed wrote in post #19465047
.
I thought that lower numbers are warmer and higher numbers are cooler?
.

.
No, low Kelvin values are cool, high Kelvin values are warm.

If you want to test this, just pick up a camera, set your white balance setting to Kelvin, dial in 4000 degrees, and snap a shot of something. . Now reset the Kelvin setting to 10,000 and shoot the same thing. . You will easily see that the image shot at 10,000 is far far far warmer than the image shot at 4000 degrees.
.

--------------- --------------- --------------- --------------- ---------------
.

dasmith232 wrote in post #19465140
.
This picture is good for illustrating the point, but it's also just a cool picture regardless!
.

.
Thanks so much!


.

what we have here is ambiguity created by usage of terms

1. 2900K houshold incandescent light is warmer than 3200K photoflood light is warmer than 5500K sunlight light is warmer than 6500K cloudy sky light

   
   "What is considered warm lighting?
   A warm light has a temperature between 2700°K and 3500°K, therefore a low temperature."

2. A setting of 10000K will result in a warmer looking captured JPG than a 5500K setting, which makes a warmer looking captured JPG than 3200K setting, which makes a warmer looking captured JPG than one taken 2900K setting...the photo looks warmest at HIGH K value and looks coolest at LOW K value.
   

Wilt wrote in post #19465147
what we have here is ambiguity created by usage of terms

1. 2900K houshold incandescent light is warmer than 3200K photoflood light is warmer than 5500K

   sunlight light is warmer than 6500K cloudy sky light
   "What is considered warm lighting?
   A warm light has a temperature between 2700°K and 3500°K, therefore a low temperature."

2. A setting of 10000K will result in a warmer looking captured JPG than a 5500K setting, which makes a warmer looking captured JPG than 3200K setting, which makes a warmer looking captured JPG than one taken 2900K setting...the photo looks warmest at HIGH K value and looks coolest at LOW K value.
   

Further complicating matters is the issue that the kelvin temperatures we use are the temperatures of black body radiators. Such as object glowing red at 2500k is cooler than when it glows white at 5500k. Yet we call 2500k warm and 5500k cool.

The yellow/blue cast comes from a mismatch between the light's temperature and what we tell the camera it is. If we shoot under a 2500k source, but tell the camera its 5500k, the camera will add too much yellow to offset the cool/high kelvin light it expects.

mike_d wrote in post #19465150
Further complicating matters is the issue that the kelvin temperatures we use are the temperatures of black body radiators. Such as object glowing red at 2500k is cooler than when it glows white at 5500k. Yet we call 2500k warm and 5500k cool.

The terminology seems contradictory but that is because a word like warm has more than one meaning. 5500K is much hotter than 2500K, but gives off a bluish light which is cooler emotionally. That's because blue is the color of water and snow and ice. The sky is blue and relatively cool compared to the much hotter yellow sun. The sky is not blue because it is at a high kelvin. It is blue because of scattering of light.

Adobe's use of color temperature is just a device and maybe an unfortunate one because temperature is not actually involved. Of course you can ignore the temperature slider and go to curves where the RGB values can be adjusted without any thought of temperature.

The more precise term is "Correlated Color Temperature" -

The Planckian radiator is also called an ideal "black-body" radiator - here, radiator is referring to the radiated, continuous electromagnetic energy spectrum emitted from the black body (an idealized opaque, non-reflective body).

The perceived color of a body that is heated to some temperature is often demonstrated qualitatively by heating a metallic conductor until it glows (gives off visible electromagnetic radiation). As the temperature of the metal is increased, the perceived color of the body changes, from red to orange to yellow at lower temperatures, to blue and similar hues at higher temperatures all the way to "white".

As has been pointed out, the descriptive terms "warmer" and "cooler" are the inverse of the actual temperature trend that produces those perceived colors when heated material radiates electromagnetic energy. Lower (cooler) temperatures produce warmer perceived color, and higher (hotter) temperatures produce cooler perceived color. Such are the vagaries of physics.

WB is used to neutralize ("balance") the prevailing light (electromagnetic radiation falling on the scene and being reflected into the camera) so that neutral surfaces appear neutral ("white"). In many raw conversion utilities, correlated color temperature (CCT) is the paradigm used to specify the balance of the blue-yellow component of WB. When you dial in a "temperature" in a raw converter, you are essentially telling the converter the CCT of the prevailing light falling on the scene. Thus, for "warm" (yellow) scenes (say tungsten lighting) you would tell the raw converter that the CCT was low or "warm" (like 2300 °K, for example) - the raw converter would adjust the R, G, and B exposure of the raw data to shift the image toward higher or "cooler" (bluer) output to neutralize the low CCT "warm" lighting.

While a blue sky illuminating a scene ("shade") is not the result of the sky being heated to high Kelvin temperatures, the spectra of the sky's radiation falling on the subject (as pointed out, due to scattering of sunlight in the blue wavelength range) correlates with the perceived color of black-body radiation at higher temperature (say 6500°K or even greater).

Intense, non-continuous lighting spectra (like LEDs and lasers, for example) often pose problems for WB because the ideal black-body radiation assumption (continuous spectrum) is not valid and the R, G, and B channels are wildly different such that clipping occurs in one or two channels while the remaining channel(s) are drastically underexposed.

Even continuous lighting subject to filtration (like stage lighting through vibrant, intense color gels) can pose issues with WB and exposure, again due to the clipping of one or more channels way before the remaining channels.

Etc.

Unfortunately, no camera manufacturer has seen fit to produce a camera that shoots raw files AND provides a raw historgram (one exception being a Canon camera running Magic Lantern firmware, another being the Pixii rangefinder camera). When posed with a challenging exposure situation (intense color, odd lighting) one has to rely on the JPEG histogram and experience - the JPEG histogram is bound to the chosen white balance set in the camera. Some internet resources provide instructions or a raw file that appears magenta to load onto your compatible camera - you specify this image as the custom WB image and the camera's histogram is then forced to simulate a UniWB histogram - where there is no channel scaling by the WB operation. It is a pseudo-raw histogram. Better then nothing; however, your JPEGs will appear very green!

Kirk

Sorry, I got myself a bit mixed up. Since we are redoing all bulbs in our house, I have mixed up the effects of "temperature" on these bulbs, using sites like this to pick our bulbs (my wife DOES NOT like bulbs that go beyond the 3500K level, too "blue" or "operating room" like). After going through all the can lights in the house and returning bulbs for others, I have this on my mind and mixed me up.

Yes, on my camera when I set up a specific temperature manually, going lower makes the scene "bluer" or "colder", and raising the K setting warms it. I often have to do this during night games and marching band, as it is just too hard to shoot a custom white balance beforehand, especially when you are part of the pit crew, and the only time you have access to the field is when you are busy scooting sound gear and props for the competition. I can however spin that K dial and get an instant readout on the mirrorless during those 3-5 minutes that we are sitting at the end of the field, so that is what I end up doing there.

Here is a comparison of a "custom" WB (based on the Color Checker neutral patch) and a UniWB (no channel scaling) rendering. The yellow arrows on each image point to the R, G, and B scaling of exposure (in Stops) applied to each channel to get the desired WB.

These screenshots are from Raw Photo Processor, a Mac-only raw converter that bucks convention and is loaded with tools that appeal to more photographic methods for adjusting images.

Kirk

Here is a fun little experiment to help visualize WB in terms of per-channel exposure instead of CCT.

I shot a CC target in daylight (late afternoon sun). I shot a reference image as one normally would, slightly underexposed so as not to clip any channels. I fixed this exposure and did not change it throughout the shooting exercise. I then placed filters in front of the lens to simulate different color lighting (yellow, orange, red, magenta).

In my raw converter (Raw Photo Processor) I added exposure to each shot so that the GREEN channel fell in the middle of Zone VI (that is, I equalized exposure across all of the images based on the green channel's exposure) for the CC patch that is used for neutral white balance (see yellow arrow in the attached image). I then took a screenshot of the rendered raw file with a raw histogram of the R, G, and B channels, where no white balance has been applied - the histogram is just for the neutral patch indicated by the yellow arrow in the top image (not the entire image). For each lighting condition, one can observe a couple of things:

1) the amount of exposure adjustment changes to get the GREEN channel to the reference value - this is due to the filter factor of each filter (the amount of green light passed by the filter).

2) the more interesting thing to note is the RELATIVE exposure between channels, and how that changes with the different lighting conditions. With no filter, it is apparent that more green exposure is captured, followed by blue, and then red. This makes sense - there are twice as many green photo sites on a sensor than there are blue or red, and there is more blue light from daylight illumination than there is red.

For a camera to WB these different channel exposures, exposure scaling of each channel is done linearly - like adding exposure compensation on a per-channel basis - to make a target neutral, neutral. In a histogram, the exposure adjustment to each channel will make the R, G, and B channels of the neutral patch all fall on top of each other. So, in the first image, with no filter applied, the blue channel would have to be boosted about one-half of a stop and the red channel a little over a full stop to get the R, G, and B values of the neutral patch exposed identically (i.e., neutral).

The red filter image is very interesting - the RED channel in the histogram is gone! It is clipped to beyond the histogram because the red filter cuts so much green light that the image had to bee boosted 3.7 stops to get the green reference properly adjusted. This image would be impossible to WB in post, and I also tried shooting a WB target and doing a custom WB in camera - no luck.

The magenta filter is also interesting - compared to the no filter, you can see that the magenta filter lifts the blue and red exposure so that the blue exposure equals the green, and the red is about one half a stop below those. This filtration is a trick to use in daylight to help boost the signal-to-noise ratio of the R and B channels when shooting in high-contrast conditions where you will be lifting the shadows a lot (increasing SNR equates to decreasing the noise). This is not so relevant any more with camera sensors getting better and better, but back in my Canon 5D days, this trick helped in certain situations. A more intense magenta filter will decrease the overall exposure (higher filter factor) but bring the blue and red channels closer to green - magenta is the complement of green, so this makes sense.

However, if any filter gets too intense and one or two channels are exposed way differently compared to the others (like in the red example above), then the overall color rendering suffers, even if you can possibly WB the image. The color profile for the camera cannot cope with the extreme transformations required to render the image from the heavily biased raw data.

[/nerd]

Yay!

Kirk

Here is a comparison of the color rendering of each of the above images, WB'ed and exposure adjusted on the green channel using the neutral patch.

As one might expect, the more channel scaling required (more extreme WB adjustments) the more the color suffers.

Kirk

And here are no filter images, UniWB and exposure adjusted for the GREEN channel in the neutral patch for:

Daylight (top)
Open shade (middle)
Incandescent (bottom - maybe some sort of CFL, hard to tell).

Note the relative exposures amongst the channels - do they make sense to you? interestingly, in all lighting conditions, the spread of exposure across the R, G, and B channels is about 2 stops. So I would imagine that color rendering profiles are able to tolerate this range of exposure difference across channels. When the exposure difference gets greater than this, perhaps that is when color profiles starting falling apart.

Kirk

It would be interesting to add one more comparison, where you talk a picture of the body cap then apply that as your custom WB, then see what adjustments are needed.