Saturation and Exposure - busting a Myth

[xpatUSA]

When a shot contains highly saturated colors that are out-of-gamut in your color space of choice, the common advice is to use less exposure in the shot in order to reduce the color-saturation. However, this advice does fly in face of color theory; but why that is so is very hard to explain. So it is that my previous geeky attempts at explanation have largely fallen on deaf ears and glazed-over eyes.

While loading some applications onto an empty hard drive, I re-discovered a nifty application called ShowImage, see:

http://www.efg2.com/Lab/Graphics/Colors/ShowImage.htm

The big deal with this app is that it can show a histogram of the color saturation in a image (as determined in many editors by the HSB or HSV color-picker). So I went and found a yellow flower shot. Please guess, right now, which areas of the following shot are the most color-saturated . .



I was very careful to select an image that was not over-exposed anywhere, especially on the petals.

Now, another thing ShowImage can do is show a histogram of the tone levels in an image: the brightness (value) part of the HSB model:



In the histogram, brightness is represented as 0-255 instead of the more normal units of percent but ya get the idea: the image is not over-exposed and looks much like a gray-scale version of the original.

Now we come to saturation. To some, the following image may come as a shock:



We see that the highest color-saturation is in the less-exposed parts of the flower petals. Do notice the difference between the sunlit and the shadowed parts of that horizontal petal in the lower middle of the flower.

Therefore, had this flower been shot with less exposure, there would have been even more out-of-gamut colors in it. On the other hand, exposing to the right a bit more would have reduced the number.

Of course our eyes are very tolerant of excesses in color and, often, a horribly color-clipped image of a red rose can look "better" than one that actually has some color-contrast in it's petals.

P.S. sorry to say that ShowImage is Windows-only.

[IzzieK]

That a little interesting program...even if it is a 1995 version. I will read up on it as I am always guilty of overprocessing my image. It will be an interesting study...thanks for the link.

[xpatUSA]

That's a interesting little program...even if it is a 1995 version. I will read up on it as I am always guilty of overprocessing my image. It will be an interesting study...thanks for the link.

Thanks right back atcha, Izzie.

It's like me; old but good

I should have mentioned that it doesn't open TIFFs or PNGs.

Here's what it says in the file opening dialog: .bmp, .ico, .emf, .wmf, .jpg.

Does open in Windows XP but I don't have 7 or 8 so can't advise about that, sorry.

[Manfred M]

Interesting explanaton, Ted, although I'm not sure that you've really demonstrated anything other than what a particular piece of software displays.

I suspect that the "underexpose to increase saturation" is a carry over thought from the colour film days where it was a known practice that underexposing your film by 1/3 of a stop gave a more saturated print or transparency. A thinner negative (or transparency) would result in a higher level of saturation when printed. Some higher end camera manufacturers were accused of playing with the light meter settings to enhance this effect.

Out of gamut was never a consideration as each individual film type had its own specific characteristics and the same could be said about the papers. Slower films tended to have a wider colour range than faster ones; that translates reasonably well in the digital age as lower ISO values give higher colour ranges too.

I would have thought that the handling of out of gamut (OOG) colours would have more to do with how they are mapped and handled by the software that converts RAW data into image files. Perceptual and Relative colorimetric tend to be the most commonly used rendering intents, but even there, we are at the mercy of the software developers and their mapping algorithms. Open source software aside, we have no idea of how the knee and shoulder curves are actually determined.

[xpatUSA]

Interesting explanaton, Ted, although I'm not sure that you've really demonstrated anything other than what a particular piece of software displays.

I suspect that the "underexpose to increase saturation" is a carry over thought from the colour film days where it was a known practice that underexposing your film by 1/3 of a stop gave a more saturated print or transparency.

By "the common advice" I meant the advice that is often given when someone is complaining about an image that is excessively saturated. That advice is opposite to what is said above i.e. "underexpose to increase saturation". I'll go and edit the OP to make that more clear.

A thinner negative (or transparency) would result in a higher level of saturation when printed. Some higher end camera manufacturers were accused of playing with the light meter settings to enhance this effect.

Out of gamut was never a consideration as each individual film type had its own specific characteristics and the same could be said about the papers. Slower films tended to have a wider colour range than faster ones; that translates reasonably well in the digital age as lower ISO values give higher colour ranges too.

Interesting but I should made it clearer that I was only talking about digital images.

I would have thought that the handling of out of gamut (OOG) colours would have more to do with how they are mapped and handled by the software that converts RAW data into image files. Perceptual and Relative colorimetric tend to be the most commonly used rendering intents, but even there, we are at the mercy of the software developers and their mapping algorithms. Open source software aside, we have no idea of how the knee and shoulder curves are actually determined.

Quite so. However, I am assuming that the processing pipe-line remains the same and, equally, the subject. The advice to which I refer implies that the shooter can go back and shoot as least something similar then come back and use the same processing with the result being a better image. And the advice remains incorrect in that reducing exposure does not reduce saturation.

If we now introduce different ISO values, different converter mappings, different rendering intents and film vs. digital, I fear that this thread will obfuscate itself into oblivion

[Manfred M]

Ted, I was writing about digital too. I find it interesting how some (many?) of the practices and thoughts from the film days seem to have carried over to digital without any analysis to see if the same parameters / assumptions even apply in the digital world. Film chemistry and digital sensors are totally different creatures.

In this case, I rather suspect this is where the advice originates and is therefore something that needs to be questioned.

[chauncey]

I've always been somewhat anal when it comes to proper in camera exposure and was unaware,
until recently, how much influence color temperature has on exposure.

That said, the term saturation has always confused me, it appears to me that it is nothing
more than altering the RGB numbers which would be displayed in the resulting histogram.

FWIW, properly exposed, in my mind is in the vicinity of RGB 245.
To suggest that underexposing somehow coughs out a superior image, as compared to a
properly exposed image, is way beyond my grasp.

[xpatUSA]

I've always been somewhat anal when it comes to proper in camera exposure and was unaware,
until recently, how much influence color temperature has on exposure.

William, color temperature has nothing to do with my post, sorry.

That said, the term saturation has always confused me, it appears to me that it is nothing
more than altering the RGB numbers which would be displayed in the resulting histogram.

Lost me completely there, I'm afraid. If only it were that simple. Have a look here:

https://en.wikipedia.org/wiki/Colorfulness

To suggest that underexposing somehow coughs out a superior image, as compared to a
properly exposed image, is way beyond my grasp.

I'm pleased that I suggested no such thing, instead implying that more exposure is "better" in this case.

[xpatUSA]

Interesting explanaton, Ted, although I'm not sure that you've really demonstrated anything other than what a particular piece of software displays.

To respond meaningfully, it is sadly necessary to go back to eye-glazing stuff. I was demonstrating that a particular piece of software displays my main point better than any other particular pieces of software on my HD. For example, I have, in the past, posted 3D gamut diagrams in Lab space and tried to make the same point. Few got it, I'm afraid.



It's the same flower image as in the OP . . . but it's not exactly obvious that the darker yellows are more saturated.

[DanK]

Ted,

A very interesting thread. Thanks for posting.

This is new to me, but the math in the Wikipedia article seems clearer to me than the words. The second saturation equation gives it in LAB notation:

S_ab=C/L=[(a^2+b^2)^1/2]/L

Chroma, the numerator, is just the distance from neutral in the plane defined by the two LAB color dimensions, A and B, right? Even though the word "colorful" apparently is supposed to mean something else, that is in lay terms what this is: how far from free of color the image is. Saturation is then defined as the ratio of this quantity of chroma to luminance. That gives the result you posted. Drop exposure, and the denominator shrinks while the numerator stays the same.

Am I right so far?

If so, my next question would be: what are the practical ramifications of this? There are two different issues: going out of gamut, and losing detail. With respect to the second: I routinely drop exposure, but it is not in response to saturation properly defined, but in response to the problem that brightly exposed areas with a lot of--I was going to say "saturation", but I think I should say "chroma"--sometimes lose detail. Decrease exposure a bit, and the detail is preserved.

So for this first issue, isn't the problem more incorrect terminology than incorrect technique?

The second issue is going out of gamut. Here I am confused. The quantity [(a^2+b^2)^1/2] defines how far from neutral the area is, and in what direction. Isn't that what determines whether the image is out of gamut, not [(a^2+b^2)^1/2]/L?

Dan

[xpatUSA]

Ted,

A very interesting thread. Thanks for posting.

This is new to me, but the math in the Wikipedia article seems clearer to me than the words. The second saturation equation gives it in LAB notation:

S_ab=C/L=[(a^2+b^2)^1/2]/L

Chroma, the numerator, is just the distance from neutral in the plane defined by the two LAB color dimensions, A and B, right?

a* and b*

Even though the word "colorful" apparently is supposed to mean something else, that is in lay terms what this is: how far from free of color the image is. Saturation is then defined as the ratio of this quantity of chroma to luminance. That gives the result you posted. Drop exposure, and the denominator shrinks while the numerator stays the same.

Am I right so far?

Yes, if we are now talking about L*a*b*, but the OP was only made in terms of HSB or HSV. For those, saturation is calculated differently. Additionally, the Wiki distinguishes between colorfulness, chroma and saturation clearly enough, I thought:

colorfulness, chroma, and saturation are related but distinct concepts referring to the perceived intensity of a specific color. Colorfulness is the degree of difference between a color and gray. Chroma is the colorfulness relative to the brightness of another color that appears white under similar viewing conditions. Saturation is the colorfulness of a color relative to its own brightness.

So, it is saying that chroma and saturation are not the same thing. I would really prefer to continue with the HSV model for this discussion, if possible.

If so, my next question would be: what are the practical ramifications of this? There are two different issues: going out of gamut, and losing detail. With respect to the second: I routinely drop exposure, but it is not in response to saturation properly defined, but in response to the problem that brightly exposed areas with a lot of--I was going to say "saturation", but I think I should say "chroma"--sometimes lose detail. Decrease exposure a bit, and the detail is preserved.

So for this first issue, isn't the problem more incorrect terminology than incorrect technique?

Sorry, I can't make out which terminology is incorrect in the OP, please clarify.

Perhaps the ShowImage views are still not illustrative enough.

The second issue is going out of gamut. Here I am confused. The quantity [(a^2+b^2)^1/2] defines how far from neutral the area is, and in what direction. Isn't that what determines whether the image is out of gamut, not [(a^2+b^2)^1/2]/L?

No, and neither does [(a^2+b^2)^1/2]/L, IMHO. But L has to be in there somewhere because a color space is a volume which includes a lightness dimension.

Any one color in the 3D diagram is given by the co-ordinates of L*, a* and b*. Therefore L* must be part of any expression that determines whether a color is in or of gamut. In other words a color in 3D space has three coordinates, not two. I'm not so hot on solid geometry but surely the inclusion of L* requires something more than a simple division?

In CIELAB space, each color space has a known gamut represented by a volume in that space. In my response to Manfred above the color space gamut volume shown is sRGB. Anything outside of that volume is out-of-gamut and, for a good few colors, the lightness ordinate L* can determine whether a color is in or out.

Brain is beginning to hurt, Dan. I'll sign off for now . . .