Fun with a model! How bad can it look?

[Kevin Stecyk]

Scenario 1: We are speaking of obtaining a custom white balance setting for the camera to use with in-camera white balance color correction, from a "calibration" shot taken by the camera of a neutral target exposed to the scene illumination.

In a Canon camera, when we do that, the camera, when led to the "calibration" shot, regards the center of the frame. There is no opportunity for the user to "focus on" one portion or another. If we want the camera to regard a certain portion of the target at that time, we must frame the calibration shot to put that portion in the critical center area.

If the part of the calibration shot frame regarded by the camera is blown out (however that came to be), then the process is useless for white balance.

This was my area of interest Doug. My target has a small grid in the middle of the gray stripe so that camera can focus and shoot. If I am attempting to photograph a something with no detail, the camera's focusing system can't get a lock. Then I can't shoot. I then have to focus on the edge of the white card or something, lock focus, then shoot the white card. With my white card, however, I can focus on the gray stripe because of its grid.

If we look at Will's shot of his gray card, we note that the white is blown and the black is more variable. But as pointed out earlier, if Will filled the entire frame with his gray card, then his camera's sensor should have used the gray stripe in the middle for the white balance. So I have no clue what the issue is.

 

[Doug Kerr]

Hi, Kevin,

This was my area of interest Doug. My target has a small grid in the middle of the gray stripe so that camera can focus and shoot. If I am attempting to photograph a something with no detail, the camera's focusing system can't get a lock. Then I can't shoot. I then have to focus on the edge of the white card or something, lock focus, then shoot the white card. With my white card, however, I can focus on the gray stripe because of its grid.

Sure. Another way out is to switch to manual focus and leave the focus set wherever is handy - there is no need for the calibration shot to be in focus.

If we look at Will's shot of his gray card, we note that the white is blown and the black is more variable. But as pointed out earlier, if Will filled the entire frame with his gray card, then his camera's sensor should have used the gray stripe in the middle for the white balance. So I have no clue what the issue is.

I've lost track of what his scenario was.

Best regards,

Doug

 

[Deleted member 55]

Canon only uses the center 2.4% at the center of the frame for custom white balance with the EOS 1DsMKIII.

 

[Doug Kerr]

Attention was recently drawn here to the potential influence of reflections from the photographer's clothing on the chromaticity of the illumination on the scene.

Indeed, I have drawn attention to that while decrying the practice of identifying white balance measurements made (a) with the "instrument" (e.g., camera equipped with a measurement diffuser) at the subject location and (b) with the "instrument" at the camera position as (a) "incident light" and (b) "reflected light" measurements, respectively.

I point out that, in fact, when making the measurement with the instrument at the camera position for the shot (not a good idea, in any case), what is actually measured is the incident light on the camera position (part of which may have been reflected by the subject). (That technique, by the way, is only effective if the incident light on the camera location happens to be essentially of the same chromaticity as the incident light on the subject.)

And, when making the measurement with the instrument at the subject location, which measures the incident light there, much of that light may in fact nevertheless be "reflected", even from the photographer's garb.

(Note that this does not have to do with "gray card" white balance measurement.)

Best regards,

Doug

 

[Kevin Stecyk]

I forget what "deficiency" in the delivered image suggested inappropriate or ineffectual white balance color correction.

We have slightly green eye whites, blown red channels on the wall, and her skin is very warm., meaning overly red (very low cyan).

In his book Skin: The Complete Guide to Digitally Lighting, Photographing, and Retouching Faces and Bodies, Lee Varis states on page 125:

A good skin tone value for a Caucausian will have the highest values almost equal in magenta and yellow, with yellow a bit higher (more on this later). Cyan will be one-fourth to one-third of the high value for yellow...

He is using CMYK colorspace to judge the colors. Later on page 134, he states:

Asian skin is more yellow; the trick is not to let it get too yellow.

In looking at Will's photo, examine her skin. Look at the cyan values. In many cases the cyan value is zero or very near zero.

 

[Asher Kelman]

A Portrait in Mixed Light: How Bad Can it Look?"

Guys,

That will innocently asked, with his unusual humor, "How bad can it look?" with his single picture of a lovely model challenged us to examine the obvious salmon hue to the entire picture. Actually, the look of the picture is exactly as Will wants! Still, serendipitously, nevertheless, this thread then served as an a natural opportunity to look at color WB in one real practical setting. Your discussion shows how one investigates such an phenomenon and so might be useful for others. Who hasn't themselves ever been faced with odd color hues after using WB procedures seemingly sufficient for the purpose.

Will called me and described all the steps and then questioned him. What I have written in post # 87 summarizes that with my best efforts and understanding.

Original portrait by Will Thompson on the left and edited by Joachim on the right​

As a result of this discussion, which is by no means settled, we have at least made progress in 2 main areas. We have demonstrated the following:


1. One might discover significant inconsistencies in the test conditions by checking

• The Target: LAB or RGB values of the target used at different points over the target. If they are not closely uniform, then the base line calibration shot for auto WB is deemed flawed. But we do not know the delta or variation that would require this shot to be rejected.
  
  
• The Image: Color hues are obvious to our eyes in many cases. We also can check RGB and LAB values of the eyes and skin.

2. The take home messages, suspicions and concerns:

• The material of the WB target:[/B] A fabric collapsable white balance target may not uniformly reflect light as a flat card
  
  
• Mixed light: of ring flash and sunlight might be changed as the auto flash becomes closer for geometric and even at a push of the imagination, flash temp reasons
  
  
• The clothes: and face of the photographer might reflect and so modulate light falling on the target at very close distances.
  
  
• Tonal Recovery feature of Canon Cameras: Over-exposure and pulling back with the tonal recovery feature of the camera might result in incorrect calibration of WB, but we do not know that or if if this is a practical concern in the conditions described in post # 87.

Asher

 

[Doug Kerr]

Hi, Kevin,

We have slightly green eye whites, blown red channels on the wall, and her skin is very warm., meaning overly red (very low cyan).

Hard to believe that an ineffectual white balance color correction could be the cause of blown reds (unless the color correction vector adopted by the camera was very peculiar).

I don' t have the original (ex camera) file, so I can't look into that.

Best regards,

Doug

 

[Deleted member 55]

To all:

Please note the corrected white balance target towards the beginning of this thread. (post #30)

I was wearing all black with a dark brown shirt. (maybe my shiny white head was the culprit?)

 

[StuartRae]

maybe my shiny white head was the culprit?

You don't happen to have a large red beard?

 

[Asher Kelman]

You don't happen to have a large red beard?

No, Stuart!

But with a brown shirt, he might as well, be sporting the beard, LOL! I'd hope we'll see a test for that.

When doing a monitor calibration with Eye One of other spectrophotometer, we use a dim light, grey shirt and no colored posters or furniture anywhere in the room. So for a shot, against which subsequent shots will be referenced, one would think that being just 3 feet from the reference target and using a flash might be a significant departure from advisable practice.

Asher

 

[Doug Kerr]

Will was kind enough to send me the original JPG file (ex camera) of the Orange Girl shot.

Indeed, the White balance color correction that was applied (by way of CWB) was for an incident light with a very high blue component and a very low red component; the camera-computed color temperature of the assumed incident light (not the whole story, of course) was 8382 K!

I can't imagine getting even close to that except by shooting a blue target for the WB calibration shot - maybe not even then.

Will, did you ever wash your fabric target with a detergent with brighteners?

The high red component in the red coefficient of the correction vector might indeed have contributed to the red channel blowout (point for Kevin!).

The AWB (automatic) correction vector would have implied an assumed color temperature of the incident light of 6030 K.

Best regards,

Doug

 

[Deleted member 55]

A color corrected version of the original post.

 

[Doug Kerr]

Will tells me that this version was white-balance color corrected during raw development with DPP based on the white of the eye as the neutral reference, and the development was reduced by one stop from normal.

Best regards,

Doug

 

[Asher Kelman]

Will,

This is a remarkable transformation! Kudos for you for bearing with us as we struggled. We still do not know the basis for the hue shift in the first place. That you correct it, by your own explorations, deserves kudos! It is so frustrating when one does what should assure great color and then the result falls short like this did at first.

Will Thompson: Model

Original Left and Color Corrected, Right
(RAW, Canon's DPP, white of eye as WB reference)
​

It's as if you have removed a veil of salmon-hued light. Now, she looks so vivacious and alive. Her lips are red and believable and face is a vivaciously healthy pink. Even her blue jeans no longer seem aged! So to me, it's a great improvement. But what do you think, do you also prefer your latest version?

Asher

 

[Deleted member 55]

And yet another correction!

​

 

[Asher Kelman]

She's even more real!

Will, if you go on any further she's going to climb out of the frame!

So, do you know what went wrong?

Asher

 

[Kathy Rappaport]

Nice

Will the last rendition is lovely. Now if you can only get her not to scowl at you. Did Grommet growl at her and she growled back at him?

 

[Deleted member 55]

2AM color correction!

 

[Doug Kerr]

Hi, Will,

2AM color correction!

Do we get to see it, or are you just reporting in?

Best regards,

Doug

 

[Doug Kerr]

Hi, Will,

And yet another correction!

​

This is quite nice. What recipe?

Best regards,

Doug

 

[Joachim Bolte]

I'm not behind a PS-pc right now, and I don't know if my monitor gives good colors... but is there a little bit of purple/blue glow in her hair?

Suits her better than the reddish tint, it hints more towards the asian blueish black or bourgundy that I would suspect on both of these models.

 

[Doug Kerr]

Hi, it's me - you know, the engineer-snapshooter.

I thought I would take the original raw file (ex camera) for Will's (in)famous "Orange Girl" and process it the way a snapshooter would.

I used Canon DPP. I did white balance color correction based on eyedropper pickup from the manifest sclera of the subject's right eye. I had the program use its "automatic" mode of dealing with the remainder of the image processing.

I saved the image as a JPEG file, then downsized it for delivery here, applying sharpening at that time.

Here's what that did (the full frame):

William Robert Thompson: Orange Girl, 2010
(edited: Douglas A. Kerr)​

Here is the Photoshop histogram of the file posted here:

Minimum headroom in the JPEG file (R channel) (99.93 percentile basis) is about 1/3 stop.


Best regards,

Doug

 

[Asher Kelman]

Doug,

I must say, I'm liking this picture more and more, as we peel away the changes made to her. This is a delight, especially now that you have given her a milieux and the rest of her body. This, IMHO, is a far better picture with her smashing fun shoes on the decorated foreground concrete. How delightful!

More so, there are a few simple extras that others might have removed; several leaves under the bench and the gap in the paint on the upper right. They suggest our susceptibility to the passage of time. This then, accentuates the precious and perilous nature of youth. That, as all life's journey's is especially transient, vanishing away while we were anxious to become older! The paradox of youth!

So, with all richness, now, thanks to you gentlemen, we have ourselves a picture!

Will Thompson,

You are asked, " How bad?"

Not bad at all, my good friend, not bad!

Asher

 

[Doug Kerr]

Hi, Asher,

There is no dough that won't crack if we knead it enough.

Best regards,

Doug

 

[Kevin Stecyk]

While Doug might have processed the image as a snapshooter might have done, there's room for improvement.

First, scan her skin and compare the values using CMYK to the values Lee Varis describes. Please see post:

http://www.openphotographyforums.com/forums/showpost.php?p=103285&postcount=95

You'll see heavy cyan component, stronger than it ought to be.

Second, scan her hair on camera left hand side: It's blue.

Third, look at the concrete, particularly the concrete running alongside the red bricks. It's blue.

Putting this altogether, there's a heavy blue cast in this picture. We've gone from overly red to overly blue.

Color is hard. You need to pay close attention to what you are doing. Relying on a dropper or simple techniques can easily lead you astray. I've learned that lesson again (repeatedly) recently with some of my images. It's easy to get caught up on the technicalities without watching the bigger picture.

So now that I've critiqued Doug's fine work, how would I approach it? I would just set a few sample points (you're allowed four) on some critical areas. Then I would adjust curves in rgb mode to eliminate casts. You've got some good points to work with. Her bracelet appears that it should be white. Even if it wasn't white, I'd make it white and bright. Inside her hair look pretty dark. So now I've defined by bright and dark points, and they should be neutral.

Next, I would make sure her eyes whites were close to neutral, and if not neutral, at least warm.

I'd watch the skin and concrete for their values, making sure that they were reasonable.

I would also watch her hair for reasonableness too. Women's hair can be tricky because you don't know if it is dyed or not. A quick glance at her hair seems to suggest that it should be near black.

Even though you are limited to four points, you should monitor all significant color areas to make sure that they hang together. Just sample them as you move about adjusting curves.

I never let Adobe adjust my colors for me using droppers. It only takes a minute or two to do it properly by sampling and adjusting.

 

[Asher Kelman]

I'm impressed, as always, Kevin!

Asher

 

[Joachim Bolte]

First, scan her skin and compare the values using CMYK to the values Lee Varis describes. Please see post:

Please eleborate: Why would you use CMYK numbers for color evaluation, when your input is RGB, and your output is also? Sure, printed on paper this picture will present itself differently, but we are only viewing it on RGB devices...

And could you make a corrected version like you would do it, so we got something to compare workflows and quality?

 

[Doug Kerr]

Hi, Joey,

Please eleborate: Why would you use CMYK numbers for color evaluation, when your input is RGB, and your output is also?

There is a widely-held school of thought that says that one can characterize what is expected of various ethnic skin types in terms of the CMY model. That is, for a certain skin "type", we find more uniformity in the ratio among the C, M, and Y coordinates than among the R, G, and G coordinates. (Its like one of those mathematical situations where the ratios among <1 minus the variables> tell us what we want to know rather than the ratios among the variables.

Thus, goes the notion, by pursuing those expected ratios for the skin type in hand while examining the image in CMY coordinates, one can make the skin color in the image conform to the appropriate one of those models.

I was introduced to this by one of the guys here a while ago while in a big discussion about "beautifying" skin colors that arose in connection with a special color correction target offered by Drew Strickland.

I'm not sure I buy into the notion, but I wanted to illuminate it for you, as it is probably what Kevin is alluding to.

I did a little playing around to probe the credibility of it but I got tired early.

Best regards,

Doug

 

[Joachim Bolte]

I can imagine RGB being not the best colormodel to proof skin-color, but I don't think CMYK is either. Especially not when reading CMYK values directly from the RGB image. What does that do to the 'reds' that have to be compressed to fit into the CMYK space? And what CMYK-space does one use for that, for CMYK is device dependend just as RGB is. Using this method, everybody will get different values.

I prefer to use Lab so my color is much less dependend on the Luminocity value. Then it's quite easy, caucasian skin has a a* value of about 13, and a b* value of about 17 (a bit more yellow than red, one could say). Dark skin is a little redder, asian skin a little more yellow... L* values can be adapted separately, caucasian is about 38% L* I think...

After correction you convert back to your output colormodel, wether that is RGB or CMYK.

 

[Doug Kerr]

Hi, Joey,

I prefer to use Lab so my color is much less dependend on the Luminocity value. Then it's quite easy, caucasian skin has a a* value of about 13, and a b* value of about 17 (a bit more yellow than red, one could say). Dark skin is a little redder, asian skin a little more yellow...

It's hard to imagine that there would be consistency of the a* and b* values at different values of L*.

For example, these two colors have the same chromaticity:

L*a*b* 75,13,17
L*a*b* 56,10,13

This is of course because the L*a*b* space is a "luma-chroma" space; in fact, for the same chromaticity, a* and b* scale directly with L*.

L* values can be adapted separately, caucasian is about 38% L* I think...

Typically true if we are speaking of the L*a*b* coordinates of the reflective color of the skin.

But of course the L* value of a patch of skin in an image is hardly constant (unless we always use a "zone-system" approach to exposure and all instances of the skin are under the same illumination (no shadowing, for example).

Best regards,

Doug