Finding neutral grey for color correction, without a grey card.

[Thonord]

I am familiar with using a solid color grey layer, blending with Difference and using Threshold to locate neutral grey.

However, I don't understand it!

In my mind, Yes, you will find neutral grey pixels in the image, but that cannot be what we are looking for.

Are we not really looking for pixels that differ from the image rendition of grey and what was grey in reality.
Is it not the pixels that are not grey - but should be, we are looking for?

 

[thebestcpu]

Hi Thonord
I will give you my take on the procedure you have mentioned.

First, it really only works on images where the original scene or the corrected image includes actual neutral mid range tones. If the image is not of that nature then the process procedure will not work. This is a big "IF" assumption yet a fair number of images will meet that criteria.

Another assumption (fair assumption) is that the color shift that needs correcting is uniform over the entire image.

A third assumption is that the color shift that the image includes does not shift any true existing colors to neutral gray. This is also a fair assumption yet some color shifts can turn some colors to be neutral. In such cases the above procedure would also not work.

So if the image meets the criteria where the procedure will work, here the theory behind the procedure.

Since the image will not contain any neutrals grays with the above criteria, what the procedure does is find those areas of the image that are the closest to neutral gray. Since any any non-neutral area would be pushed further away from neutral gray, then the areas of the image that are closest to neutral mid tone gray will be the gray areas that were color shifted. Then the procedure says to do color balance on those points in the image which should only contain the amount of color shift from true gray.

The biggest problem I have with this approach is that there are a reasonable amount of images that do have neutral color points that are not mid tone gray (i.e. closer the black or closer to white). The above procedure does not work with those images and you would still be better off by guessing what should be a neutral color in the image and use that as the starting point for color correction.

Not sure what I wrote is clar enough or not so ask more questions if it was (or I was) not clear enough.

Hope this helps some.

 

[Thonord]

Thank you very much. Your explanation is a paragon of lucidity. I understood it the first time I read it.

Question answered. Now I have to find out how to mark a thread "Answered"

 

[thebestcpu]

Question answered. Now I have to find out how to mark a thread "Answered"

I think you just did

and your welcome.

Or is that better said "bare hyggelig"

 

[MikeMc]

Excellent answer, to pick a bit more, many times (correcting old faded colors) finding the black point , then the white point, never finding neutral grey...most images are well balanced with 2 of three points found.

 

[Tom Mann]

Yup, the comments by TheBestCPU (John) are right on the mark, particularly, his cautions about using automated methods to do white balancing.

Let me add one thing to his comments. Often, while there may not be anything in the image that you know for sure is a neutral gray, there may be areas in the image that do have well known colors -- just not grays.

So if you can make these come out the right color (instead of working with grays), you often can get a very acceptable color balance, quite similar to what one would get had a gray area been present and used. For example, a commonly occurring, reproducible color in many images is skin, particularly, if it is someone (eg, member of your family) that you have photographed previously with good success w.r.t. color balance.

The way to do this is to use a "Curves" adjustment layer. Read off the RGB values of the skin in the image that needs to be corrected (TBC), as well as the RGB values of the skin in your reference skin image, and enter Red_TBC on the x-axis of the red curve, and Red_goal on the y-axis of the red curve, then do the same for the Green and Blue curves.

Here are some web pages that describe this technique:
http://www.zuberphotographics.com/co...rves-color.htm (scroll about 1/3 rd of the way down the page to the section titled, "Color Channels Technique")
http://www.zuberphotographics.com/co...hop/curves.htm (probably too basic for you).

And here is a post I wrote in this forum where the goal was to adjust a recently acquired photo of a tooth to make it have the same color as a reference (ie, goal) tooth:
https://www.photoshopgurus.com/foru...r-analysis-post1533709929.html#post1533709929

If you have any questions, don't hesitate to ask.

Cheers,

Tom M

 

[Thonord]

Mike MC,Quote: ".most images are well balanced with 2 of three points found." UnQuote
Duly noted. Didn't know that. Very interesting

Tom Mann, Quote: "The way to do this is to use a "Curves" adjustment layer. Read off the RGB values of the skin in the image that needs to be corrected (TBC), as well as the RGB values of the skin in your reference skin image, and enter Red_TBC on the x-axis of the red curve, and Red_goal on the y-axis of the red curve, then do the same for the Green and Blue curves." UnQuote
Point well taken, thank you.

Ok, so maybe you are not quite as lucid as "thebestcpu". I had to read that twice

And; don't worry about being :"too basic". I'm autodidact, and sometimes it worries me how much basic knowledge I miss. I only know how to do, what I have needed to do - or stumbled across.
Talk about stumbling? Zuberphotographics. Their downloadable curves and stuff, with explanation - are going to keep me busy for weeks