Problem getting Zerene Stacker 3-D Rocking to work

[Iainp]

I'm very pleased with the increase in the quality of images I'm getting now using Zerene Stacker. This is a dried out moth I found on the windowsill the other day. It needs tidying up, and I should have arranged a different background, cleaned away dust etc, and I need to seriously sort out my lighting arrangements but this is still light years ahead of anything I was able to produce before:

(96 image stack)



My problem is that I can't get the 3D Rocking to work for me. This may just be because I'm not grasping conceptually what's going on.

I went to Options > Preferences > Stereo/Rocking in ZS and I got a choice of settings. I tried the default settings first:

First image maximum shift in X and Y +3
Last image maximum shift in X and Y -3

I chose Number of output images: 3 - just to speed things up - then ran the stacking in PMax.

I only selected a sequence of 10 images from my 96 to keep it simple.

The result is:




One image " smeared" to the left, one fine, one smeared to the right.

Assuming that the issue was something to do with the setting, I ran it again using all combinations of X and Y, first and last image, and +4 to -4 but always the same problem. Here's the result at the minimum settings: +1 and -1, number of output images 4:






The smearing is reduced at minimum settings but still very evident. I've tried this with different stacks of images with the same result. Can anyone help?

Many thanks,

Iain

[rjlittlefield]

Iain,

I am strapped for time right now, so let me just give recommendations. I will post an explanation later today.

Recommendations

1. Set Options > Preferences > Preprocessing > Image Pre-sizing to around 25%.

2. Set Options > Preferences > Stereo/Rocking parameters to 7 images with maximum shifts of -3 and +3 in X, 0 in Y.

3. Put a checkmark on "Generate stereo pair or rocking sequence".

4. Add all 96 images to your project. Do not add just 10.

5. Run Stack > Align And Stack All (PMax).

This should produce seven images, each apparently from a slightly different viewpoint. The center image will be the same as if you did not have a checkmark on "Generate stereo pair or rocking sequence". The other six images will be apparently from various viewpoints left and right of center.

There should be little or no smearing in the output images that are closest to center. The more extreme images may or may not be smeared, with settings as listed above.

--Rik

Edit: fix typo in name.

[Iainp]

I'm on to it. Thanks

[Iainp]

Gosh it works
I'll post results tonight

[rjlittlefield]

Iain,

Here is part two of the answer.

Explanation

What stereo/rocking does is conceptually straightforward.

Imagine that your source frames have been properly aligned, then they have printed on card stock and arranged into a vertical stack.

If you close one eye and look straight down at that stack from directly above it, what you see is the standard alignment.

If you now open both eyes and shift your head slightly so that your nose is straight over the stack, then neither eye sees the standard alignment. Instead, your left eye sees cards at the top of the stack as being shifted slightly to the right with respect to lower cards, and your right eye sees them shifted the other way.

If the cards were made of transparent plastic and all the OOF regions were simply erased, then the stack of cards would look much like PMax or DMap output. Close one eye and look straight down, and you see the standard output. Open both eyes and look a little bit from each side, and you see two side views that make a stereo pair.

What the stereo/rocking parameters do is simply a bit of math to create those side views computationally. The center frame of the stack is taken as a fixed reference, and frames in front and in back of center are shifted a small amount in proportion to their depth in the stack.

It is here that the numbers start to get important.

The maximum shift values specify how much the top and bottom of the stack should be shifted with respect to each other, and the units are fractions of the maximum frame dimension. For example, if your images end up 800 pixels wide for posting, then a maximum shift of 3% means that the top and bottom of the stack will be shifted by 24 pixels with respect to each other (800*0.03 = 24). This shift is distributed evenly throughout the stack.

You can see now why there is a big difference between 10 frames and 96 frames. With 96 frames, the shift per frame will be only 24/(96-1) = 0.25 pixels. But with only 10 frames, the shift per frame will be 24/(10-1) = 2.7 pixels. The smaller value will look fine, perhaps with a bit of streaking, but the larger value will show nasty echoing as illustrated in the images from your early experiments.

I recommended to use the full stack and reduce your image size with preprocessing because that runs fairly quickly and gives a good approximation to what you would get at full size if only you waited long enough.

If you did want to process at full size but with fewer frames, then the trick would be to reduce the maximum shift value in proportion to the smaller frame count. Instead of 3% with 96 frames, you might use 0.28% with 10 frames. The value 0.28 will not be offered by the ZS user interface, but you can just type it in.

I'm glad to hear that the recommendation worked, and I'll be interested to see the results.

Thanks for asking -- I'm sure the question will come up again.

--Rik

[Iainp]

Now I get it! Thanks very much for the explanation and here is the result:

This one is untouched:

http://iainpetrie.typepad.com/.a/6a00e3 ... 0ba970b-pi

(takes a few seconds to load and run)

And this one has been sharpened and adjusted in Photoshop. I think I may have overdone it though:

http://iainpetrie.typepad.com/.a/6a00e3 ... 073970b-pi

Now I'm going to experiment with the Stereo Pairs

[rjlittlefield]

Bingo -- you got it.

Yes, the sharpening does seem overdone. It exposes some real detail that the unsharpened one did not, but it also introduces and emphasizes quite a few artifacts. I suspect something in between would have been better.

Here's another thought...

With rocking animations it's better to use more frames over a smaller range, while for a stereo pair it's better to use the most separation that gives acceptable artifacts. That's because in a rocking sequence, it becomes easy to spot artifacts like streaks that come and go and come and go as the animation repeats. With a stereo pair, especially if the pair is symmetric (same offset to left and right), whatever artifacts there are tend to be similar in the two views, hence not obvious.

At least this is my current thinking. I haven't run any controlled studies of the issue, but this is the gut feeling and later analysis that comes from my experiences in building up the page of stereo samples.

--Rik

[Iainp]

Right, stereo pairs experimentation will be today's project, as well as making my first snoot, playing around with microscope set up, more stacking, making a moth trap, looking for insects....Thank goodness it's the weekend

[Iainp]

I thought making an anaglyph, for someone like me with very limited computer skills, might be tricky but in the end it only took a week

I found the free StereoPhoto Maker (http://stereo.jpn.org/eng/stphmkr/) mentioned on the Zerene Stacker site very difficult to get to grips with. Granted, this probably says more about me than the software, but in the end I found this tutorial on Youtube which showed me how to do it in Photoshop in just a couple of minutes:
http://www.youtube.com/watch?v=8TyzhHpF-UM

Here's the result. As we say in England, (well, in the north anyway) I'm dead chuffed, and I've already sent out the red/green glasses to everyone who knows me.



I'm surprised how much "extra information" the anaglyph seems to reveal, compared to the original live view through the microscope. It appeared pretty much flat at the time whereas the anaglph shows the 3D form really well, especially on the body, and where the white "feathery" bits overlap the lower right edge of the eye.

[rjlittlefield]

That's a clever trick, using the advanced blending mode in Photoshop. Open, Open, Select All, Copy, Paste, double-click layer, uncheck the R channel under Advanced Blending, align/crop if needed, and save.

The way I would have handled this task using StereoPhoto Maker is as follows.

Assuming I had two separate images, then to make the anaglyph version: launch StereoPhoto Maker, Open Left/Right Images, click the Color Anaglyph button to create a "color (red/cyan)" version, click the swap sides button if needed, then Save Stereo Image to a new file.

If I had a side-by-side pair packaged as a single image, then Open Stereo Image instead of Open Left/Right Images.

In either case, if cropping is needed, then use the SPM crop tool before saving.

Do these sequences fail for you, or <insert own description> ?

The anaglyph works very well, by the way.

--Rik

[Iainp]

Ah you make it sound so simple! I'll revisit and try again (I must admit, I failed to even upload my photos to the StereoMaker in the first place... )

I would give it a go now, but I'm off to try out my snoot. I'd never heard the word until you mentioned it just over a week ago now I've made my own. I've come a long way