Picture Postcard Workflow: MMM Color Split
A new dive into the MMM action
It’s been well over one and a half year since the last time I published an article. Sometimes readers contact me, asking if I’m fine – don’t worry, I am happy and healthy, just giving priority to different things in life. But photography never escapes my interest, and nor does post-processing and the PPW. In fact, I keep a list of possible subjects that I could write about. Most of the items on that list will never make it to an article, but some remain in the back of my head, and pop up every now and then because I am confronted with them in some image.
One such subjects is “MMM and color noise”. Run MMM followed by Color Boost as the PPW prescribes, and all looks wonderful - except for an ugly lot of color noise that has been evolved out of a seemingly clean and flat original. How come, and what to do about it? Reducing opacity of MMM’s Color layer helps, but it also reduces the beneficial effect. Running another Color Noise reduction also helps, but it’s slow and often not as effective as one would like.
After some thorough investigation, I now have a much better idea of what causes the color noise (MMM is guilty of couse, Color Boost only intensifying the crime). I also have an idea how to moderate it, albeit by a manual action. And as a side-effect of all this analysis, I developed a slightly more versatile version of the MMM Finetuned action. Read on!
Have a look at figure 1. It’s an original image of the Italian city San Giminiano, taken from a large distance. The hazy conditions caused a flat and dull result, so I started to correct the range and arrived at figure 2. This is the basis for our investigations.
Figure 2 is not bad at all, except that the town, main subject of the scene, lacks color. It’s a clear candidate to run a combined MMM (Finetuned) plus Color Boost. So here is what I did:
Done! Figure 3 is the resulting image. Lovely isn’t it? Well, until we zoom in. Figure 4 is a fragment of the town area shown at 100%. Disgusting noise! (Click the image for an even better view.) One may argue that in print it’s not so bad (which I believe is true, unless maybe for a poster-size print). But it bothers me anyway. I want to understand what causes this noise – why not find out if a different approach exists that gives a cleaner result?
Noise: a crime scene investigation
So back to the layer stack. The original is practically noise-free. The image was taken with ISO 200, and the preliminary steps do not evoke any noise at all. To get a feeling of what happens with the colors, I show you the color channels (A and B).
Figure 5. Fragment of A (left) and B (right) channels before applying the MMM action
Figure 6. The same fragments after applying MMM
The original A and B of the town area are shown in figure 5. Note that the A is almost flat, while the contours of the buildings are much better visible in the B. Stated differently, contrast in the A channel here is much lower (in fact, almost nonexistent) than in the B channel.
Now think back of what the MMM action does. (If you don’t know that, have a look at my first analysis of the MMM action which contains a step-by-step explanation.) The heart of the action is a call of the Equalize adjustment, which tries to spread out channel values as found in a given "advisory" selection over their full range. Now think about what that means for our color channels. As we saw, the original A of the selected area contains almost no contrast – yet the Equalize command squeezes as much detail as possible from it. The result is mostly noise, and given the strength of LAB’s color channels, a considerable amount of it. The B is also spread out: existing detail is amplified (as desired, this is what the MMM action is all about), and of course noise creeps in as well, but not as much as in the A channel.
Figure 6 shows the same image area, A and B channel, after an application of the MMM Finetuned action, where the Color layer was set to 50%.
Simply stated: while the original A contains much less variation than the original B, the MMM action treats the A just like the B. This feels not quite right, doesn’t it? The net effect is a lot of color noise, and it’s mainly located in the A channel. Look back at figure 4. Do you agree that the noise is dominantly green/magenta? That’s the result of trying to increase variation in a channel that contains virtually none. I feel that, at least in some images, MMM should give precedence to one channel over the other. In the case of our example image, shouldn’t the A be spread out less than the B?
That question doesn’t have a simple answer. Toning down the A in favor of the B indeed reduces noise a bit. But obviously it also reduces the green/magenta separation where it happens to be beneficial. In the town area of the image, the MMM action gives the yellowish roofs an orange tint: the result of the addition of magenta. If we reduce the green/magenta component, the roofs move back to yellow, which may be closer to their original color but also less vivid. On the other hand, the same MMM action makes the foreground bush too green in my opinion. Diminishing the A means a slight improvement here (see figure 7 for a comparison).
Figure 7. Result after applying MMM default (left) and after halving the effect of the A channel (right)
Separating channels and colors
I think we have to take the point of channel separation beyond the issue of noise. Shouldn’t the retoucher always have the choice to emphasize one color over another in the MMM process? Here is another example image that I think shows my point. Figure 8 is the image that is ready for a color treatment.
Figure 8. Original image, starting point for an MMM application
Not knowing which color ranges I wanted to emphasize, I drew a more or less arbitrary selection (figure 9). The default MMM Finetuned provides figure 10 from this. Not exactly a successful result if you ask me. An overdose of honey-sweet colors, definitely unfaithful to the pastels of the originals. So what to do? Reduce the opacity of the MMM Color layer, that’s the easy choice. The result would be a compromise between figures 8 and 10.
I think we can do better. Just realize that MMM's spreading of colors takes four directions: towards green and towards magenta over the A channel, and towards yellow and towards blue over the B. What if we split these four moves into four separate layers, to allow the retoucher to control and manipulate each color completely independent of the others?
This may sound like an overly complex structure, but it’s not difficult at all to accomplish. The following outlines the necessary steps:
Figure 11d. Blending Options for Yellows
Figure 12. Layer stack after the split
The layer stack now looks as in figure 12. Nothing has changed with the image itself, but of course the action has opened a myriad of possibilities to further fine-tune the result of the MMM action.
Playing with opacities, blend-if and masks
Let’s return to our example image to see what can be done. Starting from figure 10:
Figure 13. MMM Finetuned default (left) and after adjustments on the split color layers (right)
Figure 13 puts the default MMM Finetuned version and the now achieved version next to each other. I definitely prefer the latest one, but for the connoisseur even more fine-tuning is possible. While the circle in the center has gotten a nice blue, other areas may be considered too blue. It’s the Blues layer that causes the spotty appearance of the yellow frames just above the center circle. So I reopened the Blend If for this layer, switched to Blend If “Lightness” and excluded the lightest pixel values (see figure 14 for before, the Blending Options frame and after).
Figure 14. Fragment of fig. 13b (left), another blend-if on the MMM Blues layer (middle) and result (right).
Of course, much more manipulation is possible. Note that each of the four color layers now has its own mask. It could be that some color works well in one image area and is unfavorable elsewhere. Why not mask out the problematic area?
This is the clue of the article. Split the MMM Color layer into 4 layers, one for each of LAB’s primary colors. The split itself does nothing, but it enables the retoucher to emphasize some colors over others in a multitude of ways.
Some leftover questions...
1. MMM or MMM Finetuned?
Question: The examples of this article were processed with the MMM Finetuned action. But the basic philosophy of MMM Finetuned was to limit its effect to the colors of the initial, advisory selection. So how does that relate to the Color Split as suggested above? Why bother with all LAB’s colors?
Answer: Indeed, in many cases, splitting colors doesn’t add much value after a run of MMM Finetuned. But note the chosen selections of the two example images. In the first, it was a bunch of near-neutrals. In the second, it was a not-so-limited collection of colors. For exactly these two categories, the split has the most, and in fact a lot of potential.
And it is obvious that the original MMM action (the one with a button on the PPW panel) is always suitable for a color split. Definitely more so than MMM Finetuned, because the original MMM doesn’t limit itself to the colors of the initial selection.
2. What about the noise?
Question: The initial goal of this article was to find a way to limit color noise resulting from MMM and/or CB. Have we now achieved this goal? Is the color split the solution to that problem?
Answer: First, have a look at the following comparison (figure 15a and b).
Left: the result of MMM Finetuned with default opacities. Right: the result after applying the color split as explained, and then adjusting opacities to the values as listed above. Significantly less color noise! So yes, a color split is the magic trick to attain a clean and noise-free image, even after a full run of the PPW.
All kidding aside, of course it’s not so simple. The opacities that I chose to come to figure 15b were driven by artistic considerations and not by attempting to reduce color noise. So the actual reduction that we see in figure 15b was an accidental side-effect. It could as well have gotten worse.
Only if a reduction of color noise is the main focus, and the retoucher is willing to analyze noise patterns in the different color channels, then I believe the “split” color layers can help.
3. How good is it?
Question: Is this MMM Color Split really a groundbreaking idea? Can a clever usage lead to final results that otherwise would be impossible?
Answer: I don’t think so. A retoucher who applies Color Boost in combination with MMM (Finetuned) and who is willing to fiddle with opacities and blend-if sliders, can likely obtain similar results with or without splitting the MMM Color layer. In that sense, one could argue that this whole split is superfluous.
But for me, the intuitivity of the 4-layer interface is what works. And thinking about it, Color Boost could benefit from the same 4-layer split, couldn’t it?
... and two new actions
I made the split available as two new actions, downloadable in one package. One splits the MMM Color layer of the original MMM action. The other splits the MMM Fine Color layer of the MMM Finetuned action. If you want to experiment, run MMM (Finetuned) first and then the corresponding Split Color action. You can download the actions as MMM Split Color from the downloadable actions page.
Gerald Bakker, 16 January 2023
Picture Postcard Workflow
Copyright © 2015-2023 Gerald Bakker. All Rights Reserved.