Picture Postcard Workflow: LAB Color Boost or RGB Saturation Boost?
Color Boost: what is it?
The Color Boost is a step in the PPW workflow. It comes after the contrast enhancing steps, when color is revisited. In the overview (see PPW Overview and Comments) it is step 9b. Dan Margulis' advice is to combine it with the Modern Man from Mars (step 9a) which I am hesitant about because shifting and boosting color in one go is quite intimidating.
Color Boost in the PPW panel is a simple action, running in the LAB color space. It consists of a Curves Adjustment layer where very steep curves on the color channels are applied. The curves cross the origin so that neutrals are preserved. For the large majority of photographs, this causes colors to go wild. This is deliberate: the challenge is to tone down the effect back to reality, by masking, reducing opacity, or using blend-if sliders.
Effective use of the Color Boost action assumes colors to be "correct", or better: hues to be correct. This is covered by step 1 of the workflow. If not, then incorrect colors would be boosted, which is obviously undesirable. In other words, the Color Boost action is supposed to increase saturation while leaving hues intact. (Later in this article, we'll see that this is not always true, but the assumption sounds reasonable.)
The obvious question then is, wouldn't a simple Saturation boost from the Hue/Saturation adjustment in RGB have a similar effect?
Dan Margulis in his latest book assures that this is not the case, but he doesn't explain why. He devotes just a few sentences to this subject, for example: "LAB is the favored way to enhance color, usually producing a more attractive result than RGB."
All right, I am willing to accept this, but still I want to understand what happens. Can we, by reasoning or by looking at RGB and/or LAB numbers, explain the differences and thus understand why the LAB Color Boost gives a superior result?
That is the subject of this article.
One example image
I have decided to work on one image, apply the first PPW steps, make sure that hue and contrast are good and then start the comparison.
Figure 1 is the original after a rather conservative raw processing in Camera Raw (as per advice).
I loaded the result in Photoshop, applied a slight color correction, made lightest and darkest points neutral (see How to set proper endpoints in the PPW workflow) and applied the blending and curving step. No shadows/highlights, no hammers, no H/K and no MMM were applied. (In an image like this, MMM would be a natural choice, but its application may interfere with the comparison, so I skip it for now.)
The starting point for our investigation is figure 2. Note how dull the colors look. This is as expected and not an issue since we are about to improve that. But the image covers good range in all channels, and no blown out areas in any channel.
So I duplicate the image. The original remains in RGB. On it, I add a Hue/Saturation adjustment layer and set Saturation to +100. This has a ridiculous effect, but before the colors burn into my monitor, I set layer blend mode to Color. This is deliberate: in the LAB version we only touch colors, so in RGB we do the same. I think this keeps the comparison fair.
(As an aside note, it is surprising how much the image changes when we move to Color mode. The adjustment was supposed to change saturation only, wasn't it? I have no explanation for this, but I don't care for now.)
I convert the second image to LAB, and then I hit the CB button of the PPW panel. This creates two layers: Color Boost and Endpoint Adjustment. I delete the second one, it is not relevant now.
The Color Boost comes in three flavors, one with the A curve more steep than the B curve, one the other way round, and one with equally steep curves. The PPW default is A more steep than B, but I override this and set them equal.
This is also deliberate: the RGB Saturation adjustment also allows different adjustment per color (Red, Yellow, Green, Cyan, Blue and Magenta), but again we strive for as equal conditions as possible for the competitors.
The question of course is, is +100 Saturation in RGB about as strong as the PPW Color Boost at its default 75% opacity? Are the two versions similar enough to make a comparison meaningful?
Looking at the results, I think yes. See below for the two images and judge for yourself.
Image boosted with RGB Hue/Saturation
Image boosted with LAB Color Boost
The first difference that catches the eye is that the LAB version is darker. Or better, the RGB version is lighter. This is easily checked by looking at the LAB equivalents of the RGB values of the left picture. Clicking for example in the reds of the flower reveals L 47 for the original, and 55 for the updated version. This is not an exception, but almost a rule. I did not find any pixel that got a lower L value, and many have become higher.
Obviously, this is not the case for the right picture, because in LAB we leave the L curve alone.
Again, the RGB saturation adjustment is supposed to boost saturation, and not touch luminosity. Also the layer blend mode was set to Color. The explanation for the lightening is that the Luminosity component in the RGB model is different from the L value of the LAB space.
We can check that. The sample point shows 206,43,62 original and 255,22,49 on the Hue/Sat layer. Luminosity in RGB is defined as 0.3*R + 0.6*G + 0.1*B.
Using this formula, the original has luminosity 93.8, after the boost it has become 94.6. Not exactly equal, but not far off either: probably a rouding error.
My conclusion is that LAB's L is more reliable as a measure for lightness, for the RGB boosted image looks clearly lighter than the original, whereas the LAB version doesn't.
The noted difference is not very offensive, but my judgment is that LAB performs slightly better in this respect.
First difference. Saturation boost in RGB makes images lighter, even if it is done on a layer set on Color.
The next difference is clearly visible in the areas of duller color. For example, look at the leaves on the bottom left of the image. In the original, they measure around RGB 125,150,110 which is a desaturated green. The RGB boosted version comes at 75,196,1, and the LAB version, recomputed to RGB, at 98,157,71. See figure 3 for the original.
To understand the difference, we have to realize what saturation really means in terms of pixel values. Roughly said, in RGB the more the individual channels differ, the more saturated a color is. RGB 0,100,200 is clearly more saturated than 80,100,120. Even more simple, one could say that the difference between the highest and lowest of the R,G and B values is a reasonable measure for the saturation.
In LAB, it is even simpler. Saturation increases as A or B move further away from 0. L doesn't matter. So definitely 50,10,-10 is less saturated than 50,-30,30.
Now let's look at the green we just looked at. One doesn't have to be a mathematician to notice that the RGB figures 75,196,1 denote a much more saturated color than the LAB alternative 98,157,71.
So whence the huge difference?
To invoke a saturation boost, Photoshop tries to increase the highest of the R, G and B and decreases the lowest. The middle gets in-between such that the hue remains the same, and the three together retain the lightness of the original.
This may seem difficult, but it is easy to understand that there is plenty room for a saturation boost when the initial RGB values are each far away from the extreme values 0 and 255. Remember what we have here: 125,150,110, all around a channel's middle value 128. Asked to maximally increase saturation, Photoshop takes all the room it can get, and makes a bright yellow-green: 75,196,1.
In LAB, the mechanism is different. Color is boosted according to curves. In order to understand what happens, let's look at the LAB equivalents of the original pixel values. They are 59,-15,18. Obviously the L value won't change, but the A and B will. Thanks to the linear shape of the curve, we can state that the closer to 0 the original A or B value is, the less it will increase. A value of 0 will not change at all.
In the current case, the resulting values after the application of the curves are A -32 and B 38. That's a lot more color than the original but a lot less than the RGB variant which, converted to LAB, yields A -56 and B 68.
The huge saturation increase of these dull areas in RGB is objectionable because it is not in proportion to other, more saturated areas. In other words, the distinction between less saturated and more saturated color gets largely lost in RGB.
See below for a comparison. The LAB version is clearly better.
Dull greens boosted in RGB
Dull greens boosted in LAB
Second difference. Hue/Saturation in RGB gives near-neutral midtones a disproportionate boost in comparison with light or dark colors.
The next difference is related to the previous one. Let's have a look at the lighter hues of the flower. A pale yellow, coming in at RGB 250,213,125. See figure 4.
Contrary to the previous case, these RGB figures give no room to blow saturation. The highest figure (red) is already close to its maximum of 255, so what to do? Photoshop arrives at 252,212,117 which is just slightly more saturated than the original. In the image, it shows. The yellow gets boosted just slightly, not more.
Now how does this compare to LAB?
First, let's look at the result of the curves adjustment. A and B start at 5 and 49. Yellow it is, but (as we can also read from the RGB numbers) more on the orange than on the green side of yellow.
Anyway, the curves do their work, and the resulting values are 10 for the A and 104 for the B. In the meantime, L has not changed. It was and remains 87.
Now something special happens. The combination L 87, A 10, B 104 is out of the RGB gamut. It is very light and very yellow at the same time, something the RGB color space cannot reach.
(As an aside note, we are looking at sRGB here. How this would turn out in a large gamut like ProPhotoRGB may be subject of another article.)
Fair enough, the color boost has done its work, but what are we looking at then? We don't see something outside the RGB gamut, do we?
True. Given these excessive numbers, Photoshop follows a strategy of compromise. It is unable to show something both very yellow and very light, so it comes up with a color that is both less yellow and less light. Convert the 87,10,104 back to RGB and the RGB values of 255,210,0 appear. This is somewhat darker than the original but also a lot more yellow. Given the goal to "boost color" this is definitely more successful than what is accomplished in RGB.
See below. The yellow areas of the flowers look better in LAB than in RGB. Again, the LAB version is better.
Light yellows boosted in RGB
Light yellows boosted in LAB
Third difference. Colors that are both pure and very light are not affected much in an RGB Saturation boost, but they are in LAB.
A fourth difference is most apparent in the yellow-green leaves in the bottom left quarter of the image. See figure 5. After boosting, in RGB the color retains its yellow-green hue, in LAB it seems to shift to green. Is that so and if yes, can we explain?
First, let's check the pixel values. Sample one color in this area, the original measures RGB 141,180,61. That's green on the yellow side. The RGB saturated version arrives at 132,197,1. Here the R is still not far below the G, so the hue is still green with a strong yellow component.
The LAB boosted color measures (converted to RGB) 73,192,0. Note that the green and blue are approximately the same but the R is considerably lower. So we are looking at a green that has lost almost half its yellow. More green, less yellow. See below for evidence.
Yellow-greens boosted in RGB
Yellow-greens boosted in LAB
So, despite making sure that the AB curves move through the origin, we notice a hue shift in the LAB version, something that the RGB version doesn't show.
Question again is, can we explain?
Let's have a look at the LAB figures. The original shows A -27 and B 53. After application of the curves, the figures are -58 and 108 resp. This means: a good bluish green combined with an excessive yellow.
Obviously, this is out-of-RGB-gamut again, but what do we actually see?
The RGB values of 73,192,0 convert back to LAB 69/-55/67. Compare this to what the LAB Boost gave us: 69,-58,108. So, lightness is kept, greenness turns in 3 points, and yellowness 41 points. I assume this explains the effect we see: the B had become so high that Photoshop had to cut off more than a third, where the A could almost be retained. So the color loses some of its yellow and returns more green than it was before the boost.
I assume. This indeed is a hypothesis only. The same effect - hue shift - can be seen in the red of the flowers. See figure 6 for the original.
In LAB, the red seems to shift towards pink, magenta. So again let's look at the numbers. The original is 217,96,92 in RGB. Red with just slightly more green than blue, which means: a red that hangs just a tiny bit to the orange side. In LAB, this color translates to 57,48,27.
After the boost, the A gets 101, the B 57. You bet that's out-of-gamut for RGB, so Photoshop has to give in. Similarly to what we saw for the yellow greens, we can expect the A to be cut off more than the B, right? So the expectation is that we lose more magenta than yellow, and find a shift towards orange.
Not so. The resulting color is more magenta than the original. See below for a comparison again.
Reds boosted in RGB
Reds boosted in LAB
The boosted LAB 57,101,57 is translated back to RGB as 255,0,49. So the blue has remained positive where the green has vanished. Clearly we're on the pink/magenta side of red, unlike before the boost. Convert RGB 255,0,49 back to LAB and the result is 54,81,51. Again, the highest turns in most (A, from 101 to 81) whereas the lowest decreases only 6 points (B, from 57 to 51). Still the result appears to have gained in magenta compared to the original.
I have no explanation for this. It must be a side-effect of moving excessive out-of-gamut colors into the RGB space. Obviously there is no clear unambiguous rule that covers these conversions. (And if there is, I don't know it.) Hues move, but we don't know where.
For this image, it's a positive effect. Making the leaves more green and the flowers more magenta means that the colors are driven apart. Color contrast is increased, which is good from an aesthetic point of view.
But does it always work like that?
I am not sure. This difference leaves me in doubt. The workflow dictates correct hues in step 1. Do we want the hue to shift as a side effect of color boost in step 9? I am inclined to say no.
So, RGB wins here, although I say that with hesitation.
Fourth difference. The LAB color boost shows a hue shift, especially in the more saturated colors. RGB Saturation doesn't. Depending on the image, this may or may not be beneficial.
Conclusion? LAB Color Boost works better, at least for the rather extreme settings that we used.
Of course, we only studied one picture, but the differences are explicable and thus will probably show up in most, if not all, photos.
Are you in for a surprise?
There is another, more modern, saturation slider in Photoshop, included in the Vibrance Adjustment. Just out of curiosity, I tried it and compared it with the other versions. See below for this version next to the LAB boosted verison.
Image boosted with Saturation slider in Vibrance adjustment
Image boosted with LAB Color Boost
At first glance, this version is very well comparable with the LAB boosted version, even in full power.
So why not have a look at above differences and see if they are still there for this alternative saturation boost.
Zooming in on particular pixels reveals that their L value (when converted to LAB) does not always remain equal. But the general appearance of the RGB boosted version is equally light as the LAB boosted version.
My conclusion is that the RGB Saturation boost using the Vibrance adjustment is about as good as the LAB boost. It may be a little better or a little worse depending on how you appreciate the hue shift that LAB Color Boost causes.
Gerald Bakker, 3 March 2015
Copyright © 2015-2017 Gerald Bakker. All Rights Reserved.
Figure 1. Original image
Figure 2. After first steps