Guys, please have a look at the following. Not really about the thingin the picture, just the 3D effect here. I've stacked 4 images for each, did do a stack on 'auto' (offset, rotation etc) and non-auto, ie, stacking without any deviation. I really thought that would be necessary to get the proper effect, but it seems the automatic functions work better!
nonauto:
auto:
This is an example of the single focus points:
Critical evaluation please (Rik?). I was afraid the stacking wouldn't work at all, or the 3d effect would be the same as for a single frame-pair, instead of a combined amount of depth from all individual frame pairs?!
Thanks, Rene.
BTW, it's a chlorophyte colony from freshwater, flushed out into the Northsea. Preserved with iodine (blue starch!). Image taken with 60/0.8 achromat, cells are about 7 µm across.
3D and stacking in helicon focus
Moderators: MacroMike, nzmacro, Ken Ramos, twebster, S. Alden
-
- Posts: 727
- Joined: Fri May 06, 2005 11:57 pm
- Location: Richland, WA, USA
- Contact:
Rene,
Nice work, great post! This is really interesting.
I have spent many minutes studying these things. I agree completely with your assessment.
The single-focus pair is clearly 3D, no fusion problems at all, and it unambiguously establishes what's where, front-to-back. Posting this was a stroke of genius.
The nonauto pair is just a mess. The two images don't fuse cleanly, which is no great surprise, but the 3D is completely messed up -- it's much "flatter" than it should be, and there are even a couple of back-to-front inversions. That is a surprise. I've never seen that effect before and offhand I don't know what's going on.[*]
The auto pair is much better. There's one place where the fusion jitters a little, but that's mainly just because there are now two pretty sharp objects, one in front of the other. (This is outer row, at about the 11-o'clock position. ) 3D placement seems to be identical to the single-focus pair, and we've picked up depth of field in the bargain.
For only a 4-image stack, I'd say you got yourself a fine result!
--Rik
[*] OK, while writing, I figured it out. The viewing geometry for your micro world consists of two stacks skewed in opposite directions. It's not quite the same as the inclined cones in the macro worlds that I'm used to, but it's close enough to make sense. In your world, objects in front of the focus plane will be fuzzy and appear to be a bit left or right depending on view, while objects behind the focus plane will be fuzzy and a bit too right or left (opposite direction). When you move the focus plane, you have to let the images adjust so that the fuzzy and sharp images of the same thing line up properly. Otherwise the fuzzy and sharp images fight with each other, with the fuzzy images trying to give some impression of depth while the sharp ones try to kill it. (Images in the focus plane are never displaced by aperture changes. Your cut-off-half-the-light trick only affects the out-of-focus stuff!)
I've never run into this in macro because if I turn off auto-adjustment, my images get radially smeared. I think what's happening in your case is that the focus planes are far enough apart to hide the smearing, so we think that the non-adjusted stuff is aligned OK when it really isn't.
A good problem... Yes, you want to allow auto-adjustment.
Nice work, great post! This is really interesting.
I have spent many minutes studying these things. I agree completely with your assessment.
The single-focus pair is clearly 3D, no fusion problems at all, and it unambiguously establishes what's where, front-to-back. Posting this was a stroke of genius.
The nonauto pair is just a mess. The two images don't fuse cleanly, which is no great surprise, but the 3D is completely messed up -- it's much "flatter" than it should be, and there are even a couple of back-to-front inversions. That is a surprise. I've never seen that effect before and offhand I don't know what's going on.[*]
The auto pair is much better. There's one place where the fusion jitters a little, but that's mainly just because there are now two pretty sharp objects, one in front of the other. (This is outer row, at about the 11-o'clock position. ) 3D placement seems to be identical to the single-focus pair, and we've picked up depth of field in the bargain.
For only a 4-image stack, I'd say you got yourself a fine result!
--Rik
[*] OK, while writing, I figured it out. The viewing geometry for your micro world consists of two stacks skewed in opposite directions. It's not quite the same as the inclined cones in the macro worlds that I'm used to, but it's close enough to make sense. In your world, objects in front of the focus plane will be fuzzy and appear to be a bit left or right depending on view, while objects behind the focus plane will be fuzzy and a bit too right or left (opposite direction). When you move the focus plane, you have to let the images adjust so that the fuzzy and sharp images of the same thing line up properly. Otherwise the fuzzy and sharp images fight with each other, with the fuzzy images trying to give some impression of depth while the sharp ones try to kill it. (Images in the focus plane are never displaced by aperture changes. Your cut-off-half-the-light trick only affects the out-of-focus stuff!)
I've never run into this in macro because if I turn off auto-adjustment, my images get radially smeared. I think what's happening in your case is that the focus planes are far enough apart to hide the smearing, so we think that the non-adjusted stuff is aligned OK when it really isn't.
A good problem... Yes, you want to allow auto-adjustment.
-
- Posts: 727
- Joined: Fri May 06, 2005 11:57 pm
- Location: Richland, WA, USA
- Contact:
Rene,
Good luck chewing on the explanation -- the ideas are very hard to get across without illustrations.
Hhmm...maybe illustrations already exist...
A few months ago I banged my head for a long time on what turns out to be a similar problem in a different arena.
There, I was concerned with avoiding parallax in stitched panoramas.
But understanding the no-parallax problem really turned out to hinge on understanding what aperture placement does to the images of in-focus and out-of-focus points.
The basic ideas about that apply to this microscopy problem too.
I wrote an article to capture what I learned from the no-parallax work. Some of the illustrations and discussions may help here. See http://www.janrik.net/PanoPostings/NoPa ... xPoint.pdf, especially Figures 1-3 and the text around them. An excerpt from the bottom of page 5:
How to proceed? The only thought that comes to mind is that auto-adjustment relies on matching image features between adjacent images. I would expect it to work more reliably when the step in focus between images is small enough that some features are reasonably sharp in adjacent images. It seems to have worked well in the case shown here, despite what looks to be pretty large step in focus. If it gives you trouble in future tests, you might play around with reducing the focus step.
Hope this all helps -- I really like the images you're getting!
--Rik
Good luck chewing on the explanation -- the ideas are very hard to get across without illustrations.
Hhmm...maybe illustrations already exist...
A few months ago I banged my head for a long time on what turns out to be a similar problem in a different arena.
There, I was concerned with avoiding parallax in stitched panoramas.
But understanding the no-parallax problem really turned out to hinge on understanding what aperture placement does to the images of in-focus and out-of-focus points.
The basic ideas about that apply to this microscopy problem too.
I wrote an article to capture what I learned from the no-parallax work. Some of the illustrations and discussions may help here. See http://www.janrik.net/PanoPostings/NoPa ... xPoint.pdf, especially Figures 1-3 and the text around them. An excerpt from the bottom of page 5:
When you block half the condenser, you are imposing a small aperture in an unusual position. This makes the blur circles for out-of-focus points shift left/right, as described above and illustrated in figures 2a and 2c in the article. Most of the information about 3D is contained in that shift. When you change focus, the shifts change too. If you combine the images without adjustment, then you are combining a lot of different shifts that are not consistent with each other. What auto-adjustment does, in this case, is to lock in the shifts corresponding to your first image, adjusting all the other images to match.Summarizing so far... In general, when we impose a small aperture [in an unusual position], the in-focus image stays the same, except for getting dimmer as we make the aperture smaller. The out-of-focus image also gets dimmer, by the same amount on average, but this is accomplished by leaving intact the portion of the blur that corresponds to having the center of perspective at the aperture, while eliminating all other portions of the blur. Not only do the blur circles get smaller, but their centers shift location. If you have objects at different distances, then their images shift and scale by different amounts that are appropriate to having the center of perspective at the aperture. All this happens by just clipping the ray fans.
How to proceed? The only thought that comes to mind is that auto-adjustment relies on matching image features between adjacent images. I would expect it to work more reliably when the step in focus between images is small enough that some features are reasonably sharp in adjacent images. It seems to have worked well in the case shown here, despite what looks to be pretty large step in focus. If it gives you trouble in future tests, you might play around with reducing the focus step.
Hope this all helps -- I really like the images you're getting!
--Rik