Hi, Mike! Thoughts while shaving . . . .
I suspect my acceptance of the false gray info in the Difference layer is due to the fact that PS has a better eye than I.
I was working on an image yesterday where the subject was wearing a big, black Western hat. Sure looked black to me but PS correctly showed it as having an excess of blue under a cloudy sky. Essentially, it appears that PS was better at finding a 128-level gray point than I and I stopped at tht point.
Here’s the trail I followed in coming up with an approach that I believe resolves the apparent conflict of perception vs measurement and I think is an absolute path to color cast correction.
A true gray sample at any level will have all three RGB levels exactly the same, provided there is no color cast. Assumig there is a color cast — say blue for now — any true gray sample will have the R and G channels very close in value while the blue channel will have a higher value.
We can use this idea in setting gray points at *any* level on our curve.
Find and mark sample points that you assume are true gray. Check the Info palette to verify that the R and G channels are, say, within 5 levels of one another and the blue channel is somewhat higher. Do this at varying levels in the background layer. Then create a curve and set the deviant blue channel to the average of the R and G channels at these points.
Here are some values I got in looking for true gray points: (126/129/138) (172/175/183).
At the dark end the differences are not as great but the percentages are about the same: (32/32/35)
When I set the blue channel to the average of the other two channels on the curve, my color cast was corrected.
(Yes, I marked and corrected the end points first, using a Threshold layer.)
All the Difference layer was doing for me was leading me directly to the best 128/128/128 approximations in the image. (That’s where I got the 128/128/128 sample above.) If we used a 25% (64) gray fill layer, the Difference blend should take us directly to a 64/64/xx point. Similarily for a 75% (190) gray fill layer.
There’s more to this but I know it gets tedious.
I’d be beholden to be corrected.
cu . . . . patrick
ps: I left the previous thread attached as I presume the topic is of general interest.
"Mike Russell" wrote in message
Hi Patrick,
Thanks for your explanation! I’m intrigued enough to give it a try. It may
well be that flagging the near gray values in this way can be of value in finding the neutrals in an image. If that’s the case, and it may well be, then I’ve learned something valuable today!
Mike
patrick wrote:
OK, Mike, you got me! First off, let me say that you response was the most diplomatic exception I have ever encountered on the net. Thanks for that!
First the cop-out and then my own reservations as to why it should not work. The cop-out. . . . it works for me. It is the best and most direct route to color cast correction that I have tried. So far, it has never failed. And I have the certitude of numbers as opposed to guesses as to achieve the practical results. So much for anecdotal evidence. ("There you go again!")
My queasy feeling that it is wrong and what wakes me up at night is the fact that the Difference layer is displaying pixels that it *knows* are identical to the 50% gray fill as pure black. BUT it "knows" under distorting lighting. Given that there is a problem with the ambient lighting, it *must* misread and falsely report what it thinks is gray.
So the logic is all wrong.
Howsomever . . . . let’s get pragmatic here. Try this:
Use a threshold layer to mark the hi/low points.
Create a new layer, fill it with 50% gray and set the blend to Difference. Any pixels that match the 50% fill will be displayed as black.
Mark at least two "gray" spots. I look for spots whose RGB values lie within a range of 10 levels..
Go to you curve layer and use the eyedroppers on all four points.
I always end up with the best color cast correction I can get — ready for tweaking for effect.
The exception I take to the altenative — "finding something that you know should be neutral" — is that, unless you photographed a gray card or MacBeth chart under the same lighting, you’re making a judgment as to what "should be neutral." Lots of varieties of pavement out there, lots of cloud bases, lots of old gray mares. (Back in the days of yesteryear with light meters bigger than today’s GPS units, I missed a lot of shots by trusting that "grass is a good approximation to 18% gray.)
I stumbled on this with the intent of averaging the black pixels reported by the Difference layer at varying tone levels. That makes better logic but too old to argue with success..
Again, Mike, thanks for the inquiring (as opposed to disparaging) tone of your response.
cu . . . . patrick
"Mike Russell" wrote in message
patrick wrote:
[re using a 50% gray difference layer to locate neutrals]
I emphatically agree with Chris. This is a very direct technique and becomes the simplest with just a few iterations. Note that it also identifies neutral areas at varying tone levels for still greater effect.
I’m intrigued then. If you are claiming that using a difference layer to locate values that are "almost" neutral is helpful, I question the value of
that, compared with finding something that you know should be neutral, such
as a bride’s dress, newspaper, sidewalk, or certain areas of clouds..
OTOH, it sounds as if you are very convinced that this works, and it does at
least have the feature that it references the image in some way. Can you illustrate with an example, or describe the technique some more?
Combine it with the use of a threshold layer to find and monitor the lo/hi points and you have a nice start for your curve layer. . . . . patrick
Using threshold is indeed an excellent technique for locating the highlight
and shadow.
—
Mike Russell
www.curvemeister.com
www.geigy.2y.net