Parametric adjustments in Photoshop and Lightroom

[jisner]

I'm starting a new thread to discuss parametric adjustments and how they are most likely implemented in Photoshop and Lightroom.

In a Photoshop RGB document, every adjustment layer can be represented by three curves: one for R, one for G, and one for B. Curves, in turn, are splines, which can be represented by a table of floating point (x,y) pairs. This is why adjustment layers add so little to the size of a Photoshop document.

Furthermore, multiple adjustment layers can be reduced to a single set of R, G, and B curves. This is brilliantly demonstrated by Greg Apodaca's "adjustment layer inspector," which you can download here.

The inspector is just a PSD file, so there's nothing to install. Just open the file in Photoshop and watch the composite R, G, and B curves change as you modify the adjustment layers. Here's a screenshot of the composite curves when I've got a Brightness/Contrast, a Levels, a Curves, an Exposure, a Vibrance, a Hue/Saturation, a Color Balance and a Photo Filter in the Layers panel!

As to whether Photoshop implements adjustment layers this way, we will never know because the software is proprietary. For the same reason, we will never know if Lightroom implements its parametric adjustments this way. But the inspector illustrates that the mathematics is there to be taken advantage of. And I'm guessing that Adobe's developers are smart enough to take advantage of it.

[pnodrog]

I very much doubt that the you can reduce multiple adjustment layers to a single set of RGB curves. Adjustment layers can be applied to other layers with partial image information and different blending modes, can have their own layer mask, transparency settings and applied in different orders.

The curves as you say add little to the size of a Photoshop document are only an indication of the global outcome or adjustments and do not hold sufficient information to provide the required adjustment to every pixel in the image.

The overall curves reflect what has been done but could not reconstruct what has been done.

[Manfred M]

Like Paul, it would surprise me very much if this is the way that Adobe has implemented adjustment layers. I suggest it is unlikely that they do for the following reasons.

Every time we create a new adjustment layer, we also get a layer mask created at the same time. This allows us to apply a layer mask globally, which is the default condition when a new layer mask is created or to apply the mask locally by painting on the layer mask. This can obviously done with pixel level precision, so each adjustment layer will have to store this selection data as well. Two different adjustment layers could have an operation that applies to the same pixel or set of pixels.

We can also apply a blending mode to every layer which suggests a global mathematical operator is applied globally to the entire layer. This suggests to me that data for each new adjustment layer will have to have data stored independently of the other ones.

Add a clipping mask to the adjustment layers, and this gets even more complicated. The clipping mask can be applied globally or locally as well, so again, the specific areas can be painted onto the mask.

This is not a hypothetical argument either; as I use these techniques when I dodge and burn an image non-destructively.

[Mike Buckley]

Though I appreciate that whatever method Adobe does or does not use to implement adjustment layers might be of interest as one way of enjoying a hobby in photography, does making an attempt to understand that proprietary information help us make better photos? If so, how?

[DanK]

Though I appreciate that whatever method Adobe does or does not use to implement adjustment layers might be of interest as one way of enjoying a hobby in photography, does making an attempt to understand that proprietary information help us make better photos? If so, how?

presumably not, but I am curious anyway. I'd be very interested in seeing an explanation, but I doubt Adobe has published any.

If applied globally, curves are splines, and it seems plausible that the same is true of other global adjustment layers, such as saturation and levels. It seems plausible to me that these could be combined. However, it seems to me that as soon as one adds any location information, such as using a mask or the clone tool or sharpening with a high-pass filter, it would become impossible to represent the adjustment as a simple spline relating previous and final levels in the image as a whole.

[Manfred M]

Though I appreciate that whatever method Adobe does or does not use to implement adjustment layers might be of interest as one way of enjoying a hobby in photography, does making an attempt to understand that proprietary information help us make better photos? If so, how?

I would clearly answer "yes" to your question, Mike, but there is a matter of degree (i.e. how deep does the knowledge have to go). As photography has a technical component; whether we are using a camera or working on an image in PP, understanding how something works makes me a better photographer as it lets me chose the most appropriate approach.

Let me give you an example.

I tend to use a non-destructive workflow wherever possible because it makes tweaking my work far easier and I can easily do something about it, when I over correct.

I do a lot of dodging and burning in my retouching work. In B&W work, this is easy because all we are doing is changing the luminosity of the area we are working on. The moment I start doing so in a colour image we not only affect the luminosity, but also the saturation. This is especially apparent in burning and a common side effect can be a local increase in saturation. Knowing this, I can set up a workflow (mentioned in #3) where I can counteract this issue, which is what I do.

In order to come up with / use this particular workflow, I need to have an understanding of what John's question is all about. While I can't know for sure, testing and playing with adjustment layers, layer masks, blending modes and clipping masks have lead me to something that seems to be working for me.

[DanK]

Thanks, Manfred. I often use luminosity blending with a curves tool for that reason, but for some reason it never occurred to me that burning raises the identical issue.

[Mike Buckley]

In order to come up with / use this particular workflow, I need to have an understanding of what John's question is all about.

He didn't ask or even imply a question, so you lost me.

Even so, the example you provided seems to be your understandable response to having observed the effects of Adobe's technology. It has nothing to do with understanding the proprietary information about how and why Adobe's technology produces the effects you observed.

[jisner]

I very much doubt that the you can reduce multiple adjustment layers to a single set of RGB curves. Adjustment layers can be applied to other layers with partial image information and different blending modes, can have their own layer mask, transparency settings and applied in different orders.

The curves as you say add little to the size of a Photoshop document are only an indication of the global outcome or adjustments and do not hold sufficient information to provide the required adjustment to every pixel in the image.

The overall curves reflect what has been done but could not reconstruct what has been done.

To be convinced, try Greg Apodaca's inspector. It's just a PSD file. It has all nineteen adjustment layers. There's nothing for you to do except adjust the properties and observe the effect on the composite curve. You can add masks, change blending modes and transparency, and observe how these changes affect the composite curve. By the way, as a teacher, I find the Inspector to be an extremely useful tool for teaching blending modes.

The composite curve is applied to original photo (or to the most recent pixel layer) to render the preview you see on the screen. It has all the information needed to adjust every pixel in the image. When I say "applied" of course I do not mean that the original photo (or most recent pixel layer) is modified. A better word would be "transformed" because the pixel layer is transformed into the preview by applying the composite curve.

I agree that the overall curve cannot re-construct the various adjustment layers, any more than your bank balance can re-construct your transaction history. Why would you even want to do this when you have the individual curves in each adjustment layer? (They're like your transaction history.)

[DanK]

I agree that the overall curve cannot re-construct the various adjustment layers, any more than your bank balance can re-construct your transaction history. Why would you even want to do this when you have the individual curves in each adjustment layer? (They're like your transaction history.)

That's the point. in your first post, you first wrote:

Furthermore, multiple adjustment layers can be reduced to a single set of R, G, and B curves.

The key is "reduced." The reduction shows that changes were made to SOME pixels at a given level of luminance, but it doesn't show TO WHICH pixels the changes were made, and if local adjustments were made, this matters: some pixels at a given level of luminance will have been changed, while others will not have been changed, and this information is lost. In other words, it doesn't represent the full editing process, which is why it can't recreate it. I incorrectly read your first post to mean that the reduction represents the edits.

[jisner]

...some pixels at a given level of luminance will have been changed, while others will not have been changed, and this information is lost. In other words, it doesn't represent the full editing process, which is why it can't recreate it. I incorrectly read your first post to mean that the reduction represents the edits.

The composite curve represents the cumulative effect of the adjustment layers. It cannot recreate them. It is only used for transforming the most recent pixel layer into a screen preview. It is disposable, transient, ephemeral, because it can be easily re-constructed. For example, suppose you turn off the visibility of one of your adjustment layers. Oops, a new composite curve must be computed from the adjustment layers that remain visible. Fortunately, the computation is very fast.

[Black Pearl]

Just in case........

[jisner]

Just in case........

I think you have just invoked Godwin's Law for this discussion. Bye!

[xpatUSA]

I think you have just invoked Godwin's Law for this discussion. Bye!

I struggled to find a mention of Hitler or Nazis in this thread but failed. Does "Bye!" mean that you are leaving this thread or are you actually suggesting that the Member from Whitburn leaves !!

Looks like this thread is headed for the same fate as the last one; hopefully sooner rather than later . .

[Manfred M]

It has nothing to do with understanding the proprietary information about how and why Adobe's technology produces the effects you observed.

Carefully said, proprietary information simply means the details of how something has been accomplished has not been published outside of the organization that owns the algorithms. That certainly does not mean we do not have a reasonably good idea as to how this was accomplished. This is what "Reverse Engineering" is all about.

Reverse Engineering does not necessarily provide a perfect copy of the proprietary solution, but it may range from being "good enough" to being "better" than the original solution. Reverse Engineering a piece of software like Photoshop relies a good working knowledge of mathematics (as this is really what the core of any algorithm is about), some physics (optics) and of course knowledge of programing. If you don't believe me; have a look at Serif's Affinity Photo, Corel Paintshop Pro and Gimp to name just a few. Their algorithms may not be identical to Photoshop, but they are very close.

[Mike Buckley]

I'm very confused, Manfred. Are you saying that you've reverse engineered this stuff and that you have a better, more helpful understanding thanks to having done that?

[Cantab]

I struggled to find a mention of Hitler or Nazis in this thread but failed.

Looks like this thread is headed for the same fate as the last one; hopefully sooner rather than later . .

Ted, I'm concerned that you may be correct. I'd never heard of Godwin's law but once I had looked it up I was concerned about the tone set by the comment you quoted -- if I understood the point behind the comment correctly (which I may not have done -- the dangers of written communication with no body language).

later edit:
Ted, your edit after I posted my original comment clarifies the uncertainty I have about what was meant by John's comment.

[Manfred M]

I'm very confused, Manfred. Are you saying that you've reverse engineered this stuff and that you have a better, more helpful understanding thanks to having done that?

No Mike, I'm saying others may have reverse engineered this stuff and rolled it out in their own software that is available to buy or in the case of Gimp, to download. I suspect this is likely the case with the Infinity product as it is fairly new and when I played around with it earlier this year, it had incorporated a lot of the key functionality we find in Photoshop (although it sometimes seemed a bit "clunky", so some of the algorithms worked, but not necessarily as efficiently as the competitions. When it comes to the Corel product; it has been around so long it is quite possible that Adobe reverse engineered some of its functionality and included them in their own products. This is actually pretty common in the software industry as much of this stuff is not protected by either patent law or copyright law (so long as you develop it yourself and don't steal the code).

That being said, at a high level, I understand the subjects (math, physics, programing) well enough to understand the basis of the underlying functionality and how it could be implemented. This is no different than your knowledge in the line of business you worked in; you knew your own product lines extremely well and had a pretty good handle on the competition's products. Enough so that you could have a good discussion with your clients and why they should be buying the products you were selling and not those offered by the competition.

[DanK]

Just in case........

There was no cause for this.

[Mike Buckley]

Thanks for trying to explain everything, Manfred. Either I don't understand (likely), didn't ask the question accurately enough (also likely), or you're not understanding my question for whatever reason. Not worth pursuing any further.