Where a feature in the foreground overlaps with one in the background, a halo of unsharpness remains in the stack. In none of the stack images this zone is sharp: when the foreground feature is sharp, the background is out of focus; when the background feature is in focus, it is partly covered by the blurred foreground feature.
I have found that one way to deal with this is to take a second stack with maximum DOF, compromising sharpness for diffraction. But with some luck, there is some detail to fill in the halos.
However, even though I am photographing plants, the time it takes to shoot two stacks is long enough to see movement in the subject (They don't grow that fast, but hygroscopic movement is notable).
What other strategies exist to avoid these halos?
Ghost busting
Moderators: rjlittlefield, ChrisR, Chris S., Pau
-
iconoclastica
- Posts: 486
- Joined: Sat Jun 25, 2016 12:34 pm
- Location: Wageningen, Gelderland
Ghost busting
--- felix filicis ---
-
rjlittlefield
- Site Admin
- Posts: 23608
- Joined: Tue Aug 01, 2006 8:34 am
- Location: Richland, Washington State, USA
- Contact:
iconoclastica, welcome aboard!
Short answer: consider using the default brush in Zerene Stacker to copy details from a PMax output into a DMap output. This approach is discussed in some detail in the "How To Use DMap" tutorial, the section titled "Dealing With DMap's Limitations", HERE. If you're using Helicon Focus, you may be able to do something similar between outputs from Methods C and B.
Long answer follows...
It may help to define the problem more precisely.
You wrote: "In none of the stack images this zone is sharp: when the foreground feature is sharp, the background is out of focus; when the background feature is in focus, it is partly covered by the blurred foreground feature."
The latter two parts are exactly correct, but the first part may be misleading. Background near to the foreground edge is not exactly "unsharp"; rather it is contaminated by light coming from the out-of-focus foreground object.
When focus is placed on the background, then near to a simple edge, the background is seen clearly by half of the lens aperture, while the other half of the lens aperture is blocked by foreground.
As a result, if foreground and background are equally bright, then for background areas near the foreground edge, about half the light comes from unobstructed background while the other half comes from out-of-focus foreground.
The effect of the unfocused foreground light is essentially the same as veiling glare: it reduces contrast, mainly by brightening what should be dark features of the background.
There's nothing you can do about the basic problem, short of dissecting away foreground as you shoot the stack. (Don't laugh -- I've seen that done!)
But what you can do is to preserve and potentially enhance what information remains about the focused background.
Depth map methods by themselves (DMap in Zerene Stacker, Method B in Helicon Focus) are prone to producing what I call "loss of detail" halos, in which background detail near a foreground edge is completely lost because the depth map does not slew quickly enough from foreground to background.
In contrast, the pyramid methods (PMax and Method C) can retain focused background detail right up to the edge of foreground. (Often they retain focused background even past the foreground edge, which leads to the "transparent foreground" artifact, but that's another discussion.)
So, the trick summarized as "short answer" simply exploits the pyramid method's ability to retain background information next to the edge, while retaining (most of) the depth map method's ability to be completely faithful to colors and contrasts in the original source images.
I hope this helps. The contamination problem is a tough one. It's especially bad when the background is dark and the foreground is light, because then most of the light near the edge ends up coming from OOF foreground, cutting the background contrast even more. I've even seen cases where focused background is seen only though a narrow slit in foreground, where background is so badly contaminated that not even the pyramid methods could manage to preserve it.
--Rik
Short answer: consider using the default brush in Zerene Stacker to copy details from a PMax output into a DMap output. This approach is discussed in some detail in the "How To Use DMap" tutorial, the section titled "Dealing With DMap's Limitations", HERE. If you're using Helicon Focus, you may be able to do something similar between outputs from Methods C and B.
Long answer follows...
It may help to define the problem more precisely.
You wrote: "In none of the stack images this zone is sharp: when the foreground feature is sharp, the background is out of focus; when the background feature is in focus, it is partly covered by the blurred foreground feature."
The latter two parts are exactly correct, but the first part may be misleading. Background near to the foreground edge is not exactly "unsharp"; rather it is contaminated by light coming from the out-of-focus foreground object.
When focus is placed on the background, then near to a simple edge, the background is seen clearly by half of the lens aperture, while the other half of the lens aperture is blocked by foreground.
As a result, if foreground and background are equally bright, then for background areas near the foreground edge, about half the light comes from unobstructed background while the other half comes from out-of-focus foreground.
The effect of the unfocused foreground light is essentially the same as veiling glare: it reduces contrast, mainly by brightening what should be dark features of the background.
There's nothing you can do about the basic problem, short of dissecting away foreground as you shoot the stack. (Don't laugh -- I've seen that done!)
But what you can do is to preserve and potentially enhance what information remains about the focused background.
Depth map methods by themselves (DMap in Zerene Stacker, Method B in Helicon Focus) are prone to producing what I call "loss of detail" halos, in which background detail near a foreground edge is completely lost because the depth map does not slew quickly enough from foreground to background.
In contrast, the pyramid methods (PMax and Method C) can retain focused background detail right up to the edge of foreground. (Often they retain focused background even past the foreground edge, which leads to the "transparent foreground" artifact, but that's another discussion.)
So, the trick summarized as "short answer" simply exploits the pyramid method's ability to retain background information next to the edge, while retaining (most of) the depth map method's ability to be completely faithful to colors and contrasts in the original source images.
I hope this helps. The contamination problem is a tough one. It's especially bad when the background is dark and the foreground is light, because then most of the light near the edge ends up coming from OOF foreground, cutting the background contrast even more. I've even seen cases where focused background is seen only though a narrow slit in foreground, where background is so badly contaminated that not even the pyramid methods could manage to preserve it.
--Rik
-
iconoclastica
- Posts: 486
- Joined: Sat Jun 25, 2016 12:34 pm
- Location: Wageningen, Gelderland
Thank you, Rik, for this elaborate reply. You are right, of course, rephrasing the issue. Contaminated by OOF foreground is a better way to put it.
So, I understand PMax ought to yield sharp transitions between foreground and backgound? That is not what I see, so perhaps hygroscopic movement here too may be the cause of the issue. Hmm, I'll have to build a high moisture chamber than to prevent such drying. Or indeed cutting away the foreground, I thought about that too, but how to prevent moving the setup then...
New experiments to do tomorrow.
So, I understand PMax ought to yield sharp transitions between foreground and backgound? That is not what I see, so perhaps hygroscopic movement here too may be the cause of the issue. Hmm, I'll have to build a high moisture chamber than to prevent such drying. Or indeed cutting away the foreground, I thought about that too, but how to prevent moving the setup then...
New experiments to do tomorrow.
--- felix filicis ---
-
rjlittlefield
- Site Admin
- Posts: 23608
- Joined: Tue Aug 01, 2006 8:34 am
- Location: Richland, Washington State, USA
- Contact:
PMax does not always yield a sharp transition between foreground and background.
Because PMax is relentless about preserving the highest contrast detail at each pixel position, it is very vulnerable to the "transparent foreground" artifact. In some cases this can merge together foreground and background in an ugly way.
See page 100 at http://zerenesystems.com/presentations/ ... 151009.pdf. (This is the page titled "Transparent Foreground". Pages 85-98 show more about the overall stack and individual frames.)
It is better to think that PMax will preserve the best available sharp detail in the background, right up to the foreground edge, and potentially beyond. This is what makes it suitable for use as retouching input, where human intelligence can be used to brush that detail from PMax to DMap only where it's appropriate.
--Rik
Because PMax is relentless about preserving the highest contrast detail at each pixel position, it is very vulnerable to the "transparent foreground" artifact. In some cases this can merge together foreground and background in an ugly way.
See page 100 at http://zerenesystems.com/presentations/ ... 151009.pdf. (This is the page titled "Transparent Foreground". Pages 85-98 show more about the overall stack and individual frames.)
It is better to think that PMax will preserve the best available sharp detail in the background, right up to the foreground edge, and potentially beyond. This is what makes it suitable for use as retouching input, where human intelligence can be used to brush that detail from PMax to DMap only where it's appropriate.
--Rik
-
iconoclastica
- Posts: 486
- Joined: Sat Jun 25, 2016 12:34 pm
- Location: Wageningen, Gelderland
No, it isn't about the transparent foreground. That is an issue indeed, but also a problem that sofar I have been able to solve. I will see if I can upload some pictures here.
OK, that was easy. Above is my stack. It is artificial, but in the real ones I see the same thing. Here's the PMax:
PMax does a decent job, but needs some retouching. I think I am still with you. But when I start retouching, some parts (foreground object) become perfect, but the other parts cannot be perfected, for the missing parts are not in the input stack:
I'd be inclined to think I am over-ambitiously trying to do the impossible, better to change the viewpoint to avoid the overlap, but then I see examples of insect hairs and legs that should have shown the same problems. What did they do differently?
OK, that was easy. Above is my stack. It is artificial, but in the real ones I see the same thing. Here's the PMax:
PMax does a decent job, but needs some retouching. I think I am still with you. But when I start retouching, some parts (foreground object) become perfect, but the other parts cannot be perfected, for the missing parts are not in the input stack:
I'd be inclined to think I am over-ambitiously trying to do the impossible, better to change the viewpoint to avoid the overlap, but then I see examples of insect hairs and legs that should have shown the same problems. What did they do differently?
--- felix filicis ---