Halo Help Please

[DanK]

It just reminds me to that other thread where was posted that sequence isn't important in a parametric editor. So now I'm a bit surprised.

In the case of a parametric editor, the sequence in which the user does the edits does not matter. The sequence in which the software applies them does. The two are not the same in the case of Lightroom. They are in the case of pixel editors. That was what I and others were trying to explain to you in earlier threads.

[george013]

We can't control the sequence of how things are applied in a parametric editor; that is something that the programmers have decided for us users. That is effectively why sequence is not important (and this is not ideal, in my opinion).

From a mathematical standpoint (which is what both parametric and pixel based editors use), so long as the operations are commutative (i.e. order of operations does not matter; for instance addition and multiplication); sequence is not important. The moment this is not the case, order is indeed important, hence my general uneasiness with all parametric editors.

That being said, most computer programmers have a strong math background, so hopefully they have designed their code in such a way where this is taken care of properly.

Again, is it important to do sharpening before noise reduction? Also when you don't have control over the sequence? Why bother than?

I think you made a mistake by saying "order of operations does not matter; for instance addition and multiplication". You probably mean division and multiplication. But even then it only counts for a lineair approach of the whole image, and nothing gets outside of the range.

Even when that program is build by the biggest fool, it's a peace of software. Everything has it's place. Happens in a kind of sequence. The question is what is that sequence.

When I'm convinced of your approach of parametric, you'll be the first one to know. But I doubt that will ever happen. But never say never.

George

[Manfred M]

George - Let me show you an example of a very noisy image.

1. Here I sharpened and then applied noise reduction. Raw conversion - Adobe Camera Raw. Sharpening and Noise Reduction in Photoshop.





2. Here I applied noise reduction and then sharpened. The amount of sharpening was identical in both images. Raw conversion and noise reduction in DxO Optics Pro. Sharpening in Photoshop. I find that Optics Pro has the best noise reduction on the market.





3. Reference image - no sharpening or noise reduction applied. Raw conversion - Adobe Camera Raw.





4. Reference image - no sharpening or noise reduction applied. Raw conversion - DxO Optics Pro 11




When you sharpen before noise reduction, you end up sharpening the noise, and the noise reduction cannot eliminated it. Reduce the noise first, and it is not sharpened and the image is cleaner.

[Manfred M]

I think you made a mistake by saying "order of operations does not matter; for instance addition and multiplication". You probably mean division and multiplication. But even then it only counts for a lineair approach of the whole image, and nothing gets outside of the range.

No, I meant addition and multiplication. Subtraction and division are not commutative.

Linearity has nothing to do with it as all basic mathematical equations use these four operations.

[george013]

No, I meant addition and multiplication. Subtraction and division are not commutative.

Linearity has nothing to do with it as all basic mathematical equations use these four operations.

I thought you meant the combination of the operations. Long time ago I thought of these definitions. I don't see the importance of it in this discussion.

I'm aware the sequence of noise reduction and sharpening can be important. What do I miss if you also state that the sequence is not important. Looks like a contradiction to me.

George

[DanK]

I'm aware the sequence of noise reduction and sharpening can be important. What do I miss if you also state that the sequence is not important. Looks like a contradiction to me.

George

Please read post #21, which answered this question.

[xpatUSA]

In my view, neither ACR nor Lightroom should be used for sharpening, just because you cannot control the order of how the software applies the changes.

RawTherapee has recently acknowledged that fact in Version 5 by providing a post-resize deconvolution sharpening option in addition to the existing one.

"Cumulative" editors are quite useful because one can sharpen again and again. For example, in less fancy editors, one can up the local contrast in USM with a larger pixel width but lower amount and then, on top of that, apply a fractional pixel width at a much higher amount. Not for the purist but it works well for quick-and-dirt processing. I do a lot of that.

[xpatUSA]

I'm aware the sequence of noise reduction and sharpening can be important. What do I miss if you also state that the sequence is not important? Looks like a contradiction to me.

George

I'm finding it difficult to understand why the Member from Amsterdam continues to argue against the simple truths revealed in this thread and so well illustrated by Manfred in post #23.

[Manfred M]

I thought you meant the combination of the operations. Long time ago I thought of these definitions. I don't see the importance of it in this discussion.

I don't think I quite understand your comment. The combination of the operations is what gives us the final output. In some cases the order of the operations is important.

The reason I showed the examples I did in #23 is that the order of the operations DOES matter. Noise reduction must happen before sharpening. Both of these operations can occur in a parametric editor, but we have no idea in which order they are performed.

I'm aware the sequence of noise reduction and sharpening can be important. What do I miss if you also state that the sequence is not important. Looks like a contradiction to me.

I don't think I have ever said the sequence is not important. What I have said is that the user cannot control the sequence that a parametric editor processes the data. We don't know in what sequence the various edits are applied, because the software publishers have not chosen to tell us and we have to trust that they have implemented things in a way that works.

This is why I do virtually no editing in the parametric editor and do as much as possible in a pixel based editor, simply because I do control the sequence there, based on the order that I place layers in the layer stack.

[george013]

I don't think I quite understand your comment. The combination of the operations is what gives us the final output. In some cases the order of the operations is important.

The reason I showed the examples I did in #23 is that the order of the operations DOES matter. Noise reduction must happen before sharpening. Both of these operations can occur in a parametric editor, but we have no idea in which order they are performed.

If you're limiting the editing of an image to a simple addition and multiplying than ok. But I can hardly believe you mean that.
I just noticed that your examples are done in 2 different editors. How did you pass the image to the other image? Through a diskfile or just by passing it directly? In both ways you're passing the raster image, either as a diskfile, saving and opening, or as a memory address, directly.

I don't think I have ever said the sequence is not important. What I have said is that the user cannot control the sequence that a parametric editor processes the data. We don't know in what sequence the various edits are applied, because the software publishers have not chosen to tell us and we have to trust that they have implemented things in a way that works.

This is why I do virtually no editing in the parametric editor and do as much as possible in a pixel based editor, simply because I do control the sequence there, based on the order that I place layers in the layer stack.

Do you say now that when saving and opening that file again the results might be different?


Let me make some general statements.
1. There're roughly 2 ways to edit an image: vector based and pixel based.
2. A vector based editor is parametric.
3. An pixel based editor can be parametric.
4. The raw-converters are parametric pixel based editors.
5. Editing is done on the actual status of the image. Done in a fixed sequence.
6. In image editing the original is the disk file. That file is never changed when editing, parametric or not.
7. For the end user the difference is when saving. A non-parametric editor overwrites the original. A parametric editor is adding a list of commands to that original in some way, leaving the original untouched.
8. When passing the image from a paramatric converter to another editor you're passing the memory address of that raster image, with all the editing done. The receiving program just continues.


Ths should be a normal way for a non parametric editor.
Load image in memory.
Call procedure Darken(x,y,z,....)
Call procedure Levels(.....)
Call procedure Saturisation(.....)
etc.
Save.

In a parametric editor it would be something like
Load image in memory. Create CommandList.
Call procedure Darken(x,y,z,....), Add procedure to commandlist
Call procedure Levels(.....) Add procedure to CommandList
Call procedure Saturisation(.....) Add procedure to CommandList
etc.
Save CommandList to disk.

George

[DanK]

Mods: I suggest closing this thread. The issue of halos has been resolved. It appears that nearly all of the participants understand the difference between what are nearly universally called parametric and pixel-based editors, and several of the participants here have clearly explained it for those who didn't. The fact that the order in which sharpening and noise reduction are applied to an image is well-known and has been demonstrated here. The reason why the order of the user's edits doesn't translate into the order in which the edits are applied in the case of parametric editors has been explained. Going off into another discussion of "parametric pixel based editors" will only serve to confuse some people.

[xpatUSA]

This should be a normal way for a non parametric editor.
Load image in memory.
Call procedure Darken(x,y,z,....)
Call procedure Levels(.....)
Call procedure Saturisation(.....)
etc.
Save.

In a parametric editor it would be something like
Load image in memory. Create CommandList.
Call procedure Darken(x,y,z,....), Add procedure to commandlist
Call procedure Levels(.....) Add procedure to CommandList
Call procedure Saturisation(.....) Add procedure to CommandList
etc.
Save CommandList to disk.

George

George's ordering of the calls shows him to believe that his "command list" sequence follows the temporal order of his parametric editing!

So the question becomes: is an XMP sidecar file George's "command list" or not?

If the XMP sidecar file IS George's "command list", why are XMP files always in the same XML order?

XMP files are editable. Obviously that question could be answered by changing the order of a function or two and rendering the raw with the original and then the altered XMP. It should then be obvious (even to George):

If the renderings turn out different - then the editing steps are applied in XMP order, not temporal order.

If the renderings are identical, then George is possibly right.

Bets, anybody?

On the other hand, if George's "command list" is NOT the XMP sidecar file, what and where is this "command list" and how come only George knows about it? Perhaps George will tell us . . . . . . .

+1 to Dank's last suggestion re: Mod action.

[Manfred M]

Thread has been closed.